perfsonar-dev - nmwg: r293 - trunk/nmwg/doc/dLS

Subject: perfsonar development work

List archive

nmwg: r293 - trunk/nmwg/doc/dLS

From:
To: ,
Subject: nmwg: r293 - trunk/nmwg/doc/dLS
Date: Sun, 28 Oct 2007 23:51:38 -0400

Author: swany
Date: 2007-10-28 23:51:36 -0400 (Sun, 28 Oct 2007)
New Revision: 293

Modified:
trunk/nmwg/doc/dLS/dLS.html
trunk/nmwg/doc/dLS/dLS.pdf
trunk/nmwg/doc/dLS/dLS.xml
Log:

edits to the dLS document. mostly cleanup and clarification. the
debate about the choice of the election metric remains.

Modified: trunk/nmwg/doc/dLS/dLS.html
===================================================================
--- trunk/nmwg/doc/dLS/dLS.html 2007-10-17 14:03:15 UTC (rev 292)
+++ trunk/nmwg/doc/dLS/dLS.html 2007-10-29 03:51:36 UTC (rev 293)
@@ -1,10 +1,6 @@
<!DOCTYPE html
PUBLIC "-//W3C//DTD HTML 4.01//EN">
-<html lang="en">
-<head>
-<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
-<title>Distributed Lookup Service (dLS) in the perfSONAR Framework</title>
-<style type="text/css" title="Xml2Rfc (sans serif)">
+<html lang="en"><head><meta http-equiv="Content-Type" content="text/html;
charset=iso-8859-1"><title>Distributed Lookup Service (dLS) in the perfSONAR
Framework</title><style type="text/css" title="Xml2Rfc (sans serif)">
a {
text-decoration: none;
}
@@ -293,91 +289,7 @@
content: normal;
}
}
-</style>
-<link rel="Author" href="#rfc.authors">
-<link rel="Copyright" href="#rfc.copyright">
-<link rel="Chapter" title="1 Introduction" href="#rfc.section.1">
-<link rel="Chapter" title="2 System Specific Operation"
href="#rfc.section.2">
-<link rel="Chapter" title="3 Bootstrapping" href="#rfc.section.3">
-<link rel="Chapter" title="4 Structures and Messages" href="#rfc.section.4">
-<link rel="Chapter" title="5 Result codes" href="#rfc.section.5">
-<link rel="Chapter" title="6 Appendices" href="#rfc.section.6">
-<link rel="Chapter" href="#rfc.section.7" title="7 References">
-<meta name="generator"
content="http://greenbytes.de/tech/webdav/rfc2629.xslt, Revision 1.291,
2006/10/29 09:03:19, XSLT vendor: SAXON 8.8 from Saxonica
http://www.saxonica.com/";>
-<link rel="schema.DC" href="http://purl.org/dc/elements/1.1/";>
-<meta name="DC.Creator" content="Boote, J">
-<meta name="DC.Creator" content="Glowiak, M">
-<meta name="DC.Creator" content="Swany, M">
-<meta name="DC.Creator" content="Zurawski, J">
-<meta name="DC.Date.Issued" scheme="ISO8601" content="2007-10">
-</head>
-<body>
-<table summary="header information" class="header" border="0"
cellpadding="1" cellspacing="1">
-<tr>
-<td class="front left">perfSONAR</td>
-<td class="front right">J Boote</td>
-</tr>
-<tr>
-<td class="front left">
-</td>
-<td class="front right">Internet2</td>
-</tr>
-<tr>
-<td class="front left">Intended status: Informational</td>
-<td class="front right">M Glowiak</td>
-</tr>
-<tr>
-<td class="front left">
-</td>
-<td class="front right">PSNC</td>
-</tr>
-<tr>
-<td class="front left">
-</td>
-<td class="front right">M Swany</td>
-</tr>
-<tr>
-<td class="front left">
-</td>
-<td class="front right">UDel</td>
-</tr>
-<tr>
-<td class="front left">
-</td>
-<td class="front right">J Zurawski</td>
-</tr>
-<tr>
-<td class="front left">
-</td>
-<td class="front right">Internet2</td>
-</tr>
-<tr>
-<td class="front left">
-</td>
-<td class="front right">October 2007</td>
-</tr>
-</table>
-<p class="title">Distributed Lookup Service (dLS) in the perfSONAR
Framework</p>
-<h1>
-<a id="rfc.status" href="#rfc.status">Status of this Memo</a>
-</h1>
-<p>This memo provides information for the perfSONAR community. It does not
specify any standards or technical recommendations. Distribution is
unlimited.</p>
-<h1>
-<a id="rfc.copyrightnotice" href="#rfc.copyrightnotice">Copyright Notice</a>
-</h1>
-<p>Copyright � The perfSONAR Consortium (2007). All Rights Reserved.</p>
-<hr class="noprint">
-<h1 id="rfc.section.1" class="np">
-<a href="#rfc.section.1">1.</a> <a id="intro"
href="#intro">Introduction</a>
-</h1>
-<p id="rfc.section.1.p.1">This document describes the Distributed Lookup
Service (dLS) in the perfSONAR (pS) system. This functionality extends the
basic Lookup Service (LS) functionality that has been present in the system
for some time. The basic LS supports the storing and querying of perfSONAR
Service information as well as metadata about data stored or gathered by a
particular pS service instance.</p>
-<p id="rfc.section.1.p.2">From clients' perspective, Lookup Service
operation involves registration, deregistration, querying and obtaining query
results. Clients want to discover the services that are running in the
network. LS enables this by gathering information from the services and then
using it to fulfill client queries. Figure below presents basic LS
interactions.</p>
-<div id="ls-op">
-</div>
-<div id="rfc.figure.1">
-</div>
-<p>LS Operation</p>
-<pre>
+</style><link rel="Author" href="#rfc.authors"><link rel="Copyright"
href="#rfc.copyright"><link rel="Chapter" title="1 Introduction"
href="#rfc.section.1"><link rel="Chapter" title="2 System Specific Operation"
href="#rfc.section.2"><link rel="Chapter" title="3 Bootstrapping"
href="#rfc.section.3"><link rel="Chapter" title="4 Structures and Messages"
href="#rfc.section.4"><link rel="Chapter" title="5 Result codes"
href="#rfc.section.5"><link rel="Chapter" title="6 Appendices"
href="#rfc.section.6"><link rel="Chapter" href="#rfc.section.7" title="7
References"><meta name="generator"
content="http://greenbytes.de/tech/webdav/rfc2629.xslt, Revision 1.291,
2006/10/29 09:03:19, XSLT vendor: SAXON 8.8 from Saxonica
http://www.saxonica.com/";><link rel="schema.DC"
href="http://purl.org/dc/elements/1.1/";><meta name="DC.Creator"
content="Boote, J"><meta name="DC.Creator" content="Glowiak, M"><meta
name="DC.Creator" content="Swany, M"><meta name="DC.Creator"
content="Zurawski, J"><met
a name="DC.Date.Issued" scheme="ISO8601"
content="2007-10"></head><body><table summary="header information"
class="header" border="0" cellpadding="1" cellspacing="1"><tr><td
class="front left">perfSONAR</td><td class="front right">J
Boote</td></tr><tr><td class="front left"></td><td class="front
right">Internet2</td></tr><tr><td class="front left">Intended status:
Informational</td><td class="front right">M Glowiak</td></tr><tr><td
class="front left"></td><td class="front right">PSNC</td></tr><tr><td
class="front left"></td><td class="front right">M Swany</td></tr><tr><td
class="front left"></td><td class="front right">UDel</td></tr><tr><td
class="front left"></td><td class="front right">J Zurawski</td></tr><tr><td
class="front left"></td><td class="front right">Internet2</td></tr><tr><td
class="front left"></td><td class="front right">October
2007</td></tr></table><p class="title">Distributed Lookup Service (dLS) in
the perfSONAR Framework</p><h1><a id="rfc.status" href="#r
fc.status">Status of this Memo</a></h1><p>This memo provides!
informa
tion for the perfSONAR community. It does not specify any standards or
technical recommendations. Distribution is unlimited.</p><h1><a
id="rfc.copyrightnotice" href="#rfc.copyrightnotice">Copyright
Notice</a></h1><p>Copyright � The perfSONAR Consortium (2007). All Rights
Reserved.</p><hr class="noprint"><h1 id="rfc.section.1" class="np"><a
href="#rfc.section.1">1.</a> <a id="intro"
href="#intro">Introduction</a></h1><p id="rfc.section.1.p.1">This document
describes the Distributed Lookup Service (dLS) in the perfSONAR (pS) system.
This functionality extends the basic Lookup Service (LS) functionality that
has been present in the system for some time. The basic LS supports the
storing and querying of perfSONAR Service information as well as metadata
about data stored or gathered by a particular pS service instance.</p><p
id="rfc.section.1.p.2">From the clients' perspective, Lookup Service
operation involves registration, deregistration, querying and obtaining query
result
s. Clients want to discover the services that are running in the network.
The LS enables this by gathering information from the services and then using
it to fulfill client queries. The following figure presents basic LS
interactions.</p><div id="ls-op"></div><div id="rfc.figure.1"></div><p>LS
Operation</p><pre>
_____ __________
| | Register/De-register | |
| LS | <------------------------> | Service |
@@ -387,49 +299,7 @@
----------------> | Client/Service |
Query for Services |_________________|

- </pre>
-<p>Services interacting with an LS</p>
-<p class="figure">Figure 1</p>
-<p id="rfc.section.1.p.4">This document describes the support necessary to
extend basic Lookup Service to a distributed mode of operation. dLS reduces
overhead in a single service and maintains consistent information. There are
a few key facets of this mode of operation:</p>
-<ul>
-<li>Summarization - to reduce the amount of information sent over the
network or to anonymize sensitive data, some form of data reduction must take
place.</li>
-<li>Scope - to enable a hierarchy of systems, some form of scoping must
exist that forms logical communication and data exchange channels.</li>
-<li>Search - information location is key and the way in which distributed
location and search is handled is the crux of this service.</li>
-</ul>
-<p id="rfc.section.1.p.5">Additionally we present solutions to issues
related to allow seamless operation of this service including bootstrapping
(i.e. how service find other parts of the system) and domain specific
concerns.</p>
-<hr class="noprint">
-<h1 id="rfc.section.2" class="np">
-<a href="#rfc.section.2">2.</a> <a id="system" href="#system">System
Specific Operation</a>
-</h1>
-<h2 id="rfc.section.2.1">
-<a href="#rfc.section.2.1">2.1</a> <a id="overview"
href="#overview">Overview</a>
-</h2>
-<p id="rfc.section.2.1.p.1">The first step of information flow is when a pS
service registers with an LS. The service may know the name of an LS via
static configuration (the most common case for legacy deployments), or other
forms of bootstrapping such as multicast may occur. A service registers a
"service metadata" record about itself and full metadata (i.e. containing all
information such as subject, eventType(s), and any parameters, see <a
href="#service-metadata" title="Service metadata
example">Section 4.1</a>) about stored data it has knowledge of. Such a
record is called Lookup Information (see <a href="#lookup-info" title="Lookup
Information">Section 4.2</a>).</p>
-<p id="rfc.section.2.1.p.2">The idea is to move the metadata from a local
XML data store to a specialized LS with additional searching capabilities.
While a service instance may support limited searching, this is not necessary
as they should be focused on storing or gathering data and leave the lookup
functionality to the LS. Possible exceptions are rapidly changing metadata
like the most recent timestamp and full details of data stored in long-term
archival MAs.</p>
-<p id="rfc.section.2.1.p.3">The architecture of the dLS protocol assumes the
existence of logical rings of LS instances. The current proposal involves two
levels of rings: lower scope and upper scope. Lower scope hides inter-domain
relationships between LS instances while upper scope hides intra-domain
relationships.</p>
-<h2 id="rfc.section.2.2">
-<a href="#rfc.section.2.2">2.2</a> <a id="summary"
href="#summary">Summarization</a>
-</h2>
-<p id="rfc.section.2.2.p.1">The LS that a service contacts to register
becomes the "Home LS" (HLS, see <a href="#glossary"
title="Glossary">Section 6.1</a>) of that particular service. It is the
responsibility of the HLS to make summary data about the all of the pS
services it knows of available to the larger enterprise and to draw relevant
queries to itself.</p>
-<p id="rfc.section.2.2.p.2">The construction of such a summary is important
to the overall success of this service; summaries prevents other LS instances
from being overloaded by information, must be general enough to allow for
easy creation and exchange but also must retain enough information to provide
a rich query interface able to locate the distributed information. That means
service metadata information must be filtered (summarized) as it propagates
through the LS cloud. We start by making an observation that summarization is
best based on scope (see also <a href="#scope" title="Scope
Forming">Section 2.3</a> for forming scope), simply put this means that
we should attempt to summarize the "most" the "further" away from the source
that we get. This creates a smaller data set that travels the furthest away
while keeping the larger and more informative data sets closer to the source.
We present the strategies as such: </p>
-<ul>
-<li>Summarization for the "lower scope" (formerly known as "local
scope")</li>
-<li>Summarization for the "upper scope" (formerly known as "global
scope")</li>
-</ul>
-<p> We limit the discussion in this case to these two scopes, although
extension to "n" scopes is possible. As the number of scopes increases
additional "aggregation" will be necessary to combine information thus
reducing the size of the data sets further.</p>
-<h3 id="rfc.section.2.2.1">
-<a href="#rfc.section.2.2.1">2.2.1</a> <a
id="lower_scope_summarization" href="#lower_scope_summarization">Lower Scope
Summarization</a>
-</h3>
-<p id="rfc.section.2.2.1.p.1">This first level of summarization, between HLS
instances internal to a domain, consists of simply dropping superfluous
information from the metadata descriptions provided by registered services.
For now we define this to be simply removing additional "parameter" elements
from the metadata. Special consideration must be given to the
"supportedEventType" parameter by simply converting this to actual eventType
elements. This will ensure interoperability with legacy services.</p>
-<p id="rfc.section.2.2.1.p.2">Future iterations may choose to drop
additional pieces of information deemed unnecessary or private such as parts
of topological descriptions. This sort of modification is encouraged as long
as the data remains "symmetrical" and conforms to the schematic definitions
for a given metadata description. It should be noted that such modifications
will affect the searching procedure and could isolate the source services.</p>
-<p id="rfc.section.2.2.1.p.3">The mechanics for performing this level of
summarization can use any number of technologies. Either Extensible
Stylesheet Language Transformation (XSLT) documents or the XQuery language
(see <a href="#glossary" title="Glossary">Section 6.1</a>) may be used
to prepare the initial data for exchange in this first level. Since the
exchange of this local information will occur frequently, a simple operation
that is scheduled or on demand should be employed by the individual
implementations to ensure the regular LS functions are not impeded.</p>
-<p id="rfc.section.2.2.1.p.4">In order to make information available to the
LS cloud, the HLS will advertise this summary information to other LS
instances to propagate the appropriate information. Information exchange will
be handled using a "taking turns" protocol such as token ring. The holder of
the token will then perform the information exchange to other known instances
(see <a href="#glossary" title="Glossary">Section 6.1</a>).</p>
-<div id="hls-cloud">
-</div>
-<div id="rfc.figure.2">
-</div>
-<p>Token passing between HLS cloud</p>
-<pre>
+ </pre><p>Services interacting with an LS</p><p
class="figure">Figure 1</p><p id="rfc.section.1.p.4">This document describes
the support necessary to extend the basic Lookup Service to a distributed
mode of operation. Distributing LS functionality is necessary to scale the
system up in terms of the amount of information that can be stored and
searched. There are a few key facets of this mode of
operation:</p><ul><li>Summarization - to reduce the amount of information
sent over the network or to anonymize sensitive data, some form of data
reduction must take place.</li><li>Scope - to enable a hierarchy of systems,
some form of scoping must exist that defines local and remote communication
groups.</li><li>Search - information location is key and the way in which
distributed location and search is handled is the crux of this
service.</li></ul><p id="rfc.section.1.p.6">Additionally we present solutions
to issues necessary to allow effective operation of this service in
cluding bootstrapping (i.e. how service finds other parts of the system) and
domain-specific concerns.</p><hr class="noprint"><h1 id="rfc.section.2"
class="np"><a href="#rfc.section.2">2.</a> <a id="system"
href="#system">System Specific Operation</a></h1><h2 id="rfc.section.2.1"><a
href="#rfc.section.2.1">2.1</a> <a id="overview"
href="#overview">Overview</a></h2><p id="rfc.section.2.1.p.1">The first step
of information flow is when a pS service registers with an LS. The service
may know the name of an LS via static configuration (the most common case for
legacy deployments), or other forms of bootstrapping such as multicast may
occur. A service registers a "service metadata" record about itself and full
metadata (i.e. containing all information such as subject, eventType(s), and
any parameters, see <a href="#service-metadata" title="Service metadata
example">Section 4.1</a>) about stored data it has knowledge of. Such a
record is called Lookup Information (s
ee <a href="#lookup-info" title="Lookup Information">Section!
 4.
2</a>).</p><p id="rfc.section.2.1.p.2">The idea is to move the metadata from
a service-local XML data store to a specialized LS with additional searching
capabilities. While a service instance may support limited searching, this is
not necessary as they should be focused on storing or gathering data and
leave the lookup functionality to the LS. Possible exceptions when a client
may need to contact a service directly are when metadata rapidly changes,
like the most recent data's timestamp and full details of data stored in
long-term archival MAs.</p><p id="rfc.section.2.1.p.3">The architecture of
the dLS protocol assumes the existence of logical rings of LS instances. The
architecture should allow for multiple levels. The current proposal involves
two levels of rings: lower scope and upper scope. Lower scope hides
inter-domain relationships between LS instances while upper scope hides
intra-domain relationships. </p><h2 id="rfc.section.2.2"><a
href="#rfc.section.2.2">2.2</a>&n
bsp;<a id="summary" href="#summary">Summarization</a></h2><p
id="rfc.section.2.2.p.1">The LS that a service contacts to register becomes
the "Home LS" (HLS, see <a href="#glossary"
title="Glossary">Section 6.1</a>) of that particular service. It is the
responsibility of the HLS to make summary data about the all of the pS
services it knows of available to the larger enterprise and to draw relevant
queries to itself.</p><p id="rfc.section.2.2.p.2">Summarization is important
to the overall success of this service as summaries prevents other LS
instances from being overloaded by information. They must be general enough
to allow for easy creation and exchange but also must retain enough
information to provide a rich query interface able to locate the distributed
information. That means service metadata information must be reduced
(summarized) as it propagates through the LS cloud. We start by making an
observation that summarization is best based on scope (see also <a href=
"#scope" title="Scope Forming">Section 2.3</a> for form!
ing scop
e). Simply put, this means that we should attempt to summarize "more" the
"farther" away from the source that we get. This creates a smaller data set
that travels the farthest away while keeping the larger and more informative
data sets closer to the source. We present the strategies as such:
</p><ul><li>Summarization for the "lower scope" (formerly known as "local
scope")</li><li>Summarization for the "upper scope" (formerly known as
"global scope")</li></ul><p> We limit much of the discussion in this document
to two scopes, although an arbitrary number of scopes is possible without
extension to the model. As the number of scopes increases additional
"aggregation" will be necessary to combine information thus reducing the size
of the data sets further. </p><h3 id="rfc.section.2.2.1"><a
href="#rfc.section.2.2.1">2.2.1</a> <a id="lower_scope_summarization"
href="#lower_scope_summarization">Lower Scope Summarization</a></h3><p
id="rfc.section.2.2.1.p.1">This first level of
summarization, between HLS instances internal to a domain, consists of
simply eliding detailed information from the metadata descriptions provided
by registered services. For now we define this to be simply removing
additional "parameter" elements from the metadata. Special consideration must
be given to the "supportedEventType" parameter by simply converting this to
actual eventType elements. This will ensure interoperability with legacy
services.</p><p id="rfc.section.2.2.1.p.2">Future iterations may choose to
drop additional pieces of information deemed unnecessary or private such as
parts of topological descriptions. This sort of modification is encouraged as
long as the data remains "symmetrical" and conforms to the schematic
definitions for a given metadata description. It should be noted that such
modifications will affect the searching procedure and could isolate the
source services.</p><p id="rfc.section.2.2.1.p.3">The mechanics for
performing this level of summari
zation can use any number of technologies. Either Extensible!
Stylesh
eet Language Transformation (XSLT) documents or the XQuery language (see <a
href="#glossary" title="Glossary">Section 6.1</a>) may be used to
prepare the initial data for exchange in this first level. Since the exchange
of this local information will occur frequently, a simple operation that is
scheduled or on demand should be employed by the individual implementations
to ensure the regular LS functions are not impeded.</p><p
id="rfc.section.2.2.1.p.4">In order to make information available to the LS
cloud, the HLS will advertise this summary information to other LS instances
to propagate the appropriate information. Information exchange will be
handled using a "taking turns" protocol such as token ring. The holder of the
token will then perform the information exchange to other known instances
(see <a href="#glossary" title="Glossary">Section 6.1</a>).</p><div
id="hls-cloud"></div><div id="rfc.figure.2"></div><p>Token passing between
HLS cloud</p><pre>
_____ _____
| | | |
| LS1 | <----------------- | LS2 |
@@ -440,15 +310,7 @@
| | | |
|-------> | LS3 | ---------|
|_____|
- </pre>
-<p>HLS instances communicating via a token message</p>
-<p class="figure">Figure 2</p>
-<div id="hls-cloud1">
-</div>
-<div id="rfc.figure.3">
-</div>
-<p>Broadcast summary</p>
-<pre>
+ </pre><p>HLS instances communicating via a token message</p><p
class="figure">Figure 2</p><div id="hls-cloud1"></div><div
id="rfc.figure.3"></div><p>Broadcast summary</p><pre>
_____ _____
| | | |
| LS1 | < > | LS2 |
@@ -461,58 +323,7 @@
|____/T\ Token
\_/

- </pre>
-<p>The holder of the token (LS3) will inform everyone of its summary
information.</p>
-<p class="figure">Figure 3</p>
-<p id="rfc.section.2.2.1.p.7">Once exchanged, the details regarding storage
in the XML database backend (see <a href="#glossary"
title="Glossary">Section 6.1</a>) are also left to individual
implementations. It is understood that this information, in the possession of
non HLS instances, is provided as a convenience and should be treated in the
same way that directly registered information is (i.e. purged on expiration).
When requested, credit must also be attributed to the HLS as being
(non)authoritative for each piece of information.</p>
-<h3 id="rfc.section.2.2.2">
-<a href="#rfc.section.2.2.2">2.2.2</a> <a
id="upper_scope_summarization" href="#upper_scope_summarization">Upper Scope
Summarization</a>
-</h3>
-<p id="rfc.section.2.2.2.p.1">A designated member of each lower scope will
be required to interact with the upper scope. The mechanics of how we learn
who is the designated leader is discussed in <a href="#tokens" title="Token
Passing">Section 2.3.2</a>. The leader of each lower scope (and the
designated backup) will be responsible for examining each member's summary
information and building a summarization/aggregation that best describes the
contents of the ring.</p>
-<p id="rfc.section.2.2.2.p.2">The most natural summarization is based on the
topology of the network (like in network routing). Thus, topology-based
summarization will include this information as well as eventType information
from the other LSs. Summarization will be performed using specialized summary
algorithm. Topology information such as IP addresses will be summarized using
algorithms basing on Radix Tree (see <a href="#IP-summary" title="IP
addresses summarization algorithm">Section 2.2.2.1</a>).</p>
-<p id="rfc.section.2.2.2.p.3">Other information can be summarized in a less
programmatic fashion through the use of either Extensible Stylesheet Language
Transformation (XSLT) documents or the XQuery language as discussed in the
previous section. These mechanisms will take into account the XML elements
that represent the network topology currently used in metadata subjects as
well as additional items such as eventTypes.</p>
-<p id="rfc.section.2.2.2.p.4">The output of this process becomes a "service
summary" that represents a breadth of the original input. See <a
href="#LSControl-Summary-lower" title="LS Summary Message
(Lower)">Section 4.6</a> or <a href="#LSControl-Summary-upper" title="LS
Summary Message (Upper)">Section 4.7</a> for a mock-up of the summary
output. Additional transformations, while aggressive, will strive to preserve
as much information as possible to remain useful during the search
procedures.</p>
-<h4 id="rfc.section.2.2.2.1">
-<a href="#rfc.section.2.2.2.1">2.2.2.1</a> <a id="IP-summary"
href="#IP-summary">IP addresses summarization algorithm</a>
-</h4>
-<p id="rfc.section.2.2.2.1.p.1">To properly summarize a list of strings such
as IP addresses we must rely on the help of special algorithms and data
structures. The <a href="http://en.wikipedia.org/wiki/Radix_tree";>Radix
Tree</a> algorithm is useful in creating "associative arrays" where the keys
are expressed as strings. Radix Tree implementations have been used in
practice to index things such as text documents as well as more menial tasks
such as dictionaries. We are most interested in indexing IP addresses with
their natural hierarchy, where it is necessary to describe a large ranges of
values and have few exceptions.</p>
-<p id="rfc.section.2.2.2.1.p.2">A detailed explanation of the nuances of the
Radix Tree is well beyond the scope of this document, but a brief overview is
presented for completeness. The structure itself is best described as a
binary tree where nodes contain whole values and the edges are the primary
source of navigation. The edges of the tree can be based on a single
character or perhaps on even longer strings (one of the features that leads
to efficiency for small data sets). The algorithm should support the
following basic operations: </p>
-<ul>
-<li>Lookup: Indicate that something is or is not contained in the tree.</li>
-<li>Insert: Like in most inserts we attempt to place something into the
structure. We first start by doing a Lookup to see if it exists already; the
point where we stop is where we will insert the object. We are careful to
utilize the longest matching prefix of any other nearby edge to our
advantage. This last part is normally referred to as "splitting" and ensures
that each node has no more than two children.</li>
-<li>Delete: Delete an object from the tree. This operation will be
complicated by "collapsing" parents that have a single child and merging the
edges.</li>
-</ul>
-<p id="rfc.section.2.2.2.1.p.3">Once constructed, it is possible to consult
the structure in creating IP summaries as well as constructing information
regarding netmasks.</p>
-<h2 id="rfc.section.2.3">
-<a href="#rfc.section.2.3">2.3</a> <a id="scope" href="#scope">Scope
Forming</a>
-</h2>
-<p id="rfc.section.2.3.p.1">The next question is how to form the lower and
upper scopes. The simplest answer is that the lower scope be formed based on
the domain name of the participating systems. That would allow e.g.
internet2.edu, geant2.net, and pionier.gov.pl to potentially operate more
than one LS instance inside their own domains (for performance and
scalability.) As LS instances come online they will invoke bootstrapping
procedures to find and join a lower scoped group first.</p>
-<p id="rfc.section.2.3.p.2">The scopes should be named based on URIs. This
will allow a domain-level scope to take the form <a
href="http://internet2.edu";>http://internet2.edu</a>, with subdomain scopes
named <a href="http://internet2.edu/foo";>http://internet2.edu/foo</a>, etc.
The top-level scope can be called <a
href="http://perfsonar.net";>http://perfsonar.net</a> with potential for
geographic divisions later if necessary for performance (such as <a
href="http://eu.perfsonar.net";>http://eu.perfsonar.net</a>).</p>
-<p id="rfc.section.2.3.p.3">The major algorithms used to form and maintain
the ring structure of the dLS, no matter which scope we are talking about,
are as follows: </p>
-<ul>
-<li>Join Procedure</li>
-<li>Token Passing</li>
-<li>Summarization Notification</li>
-</ul>
-<p> Each of these procedures is paramount to keeping members of the
distributed "service" functioning correctly. The algorithms will be presented
in the frame of the lower scope, but will be used in the same manner for the
upper scope as well.</p>
-<h3 id="rfc.section.2.3.1">
-<a href="#rfc.section.2.3.1">2.3.1</a> <a id="join" href="#join">Join
Procedure</a>
-</h3>
-<p id="rfc.section.2.3.1.p.1">When an LS instance comes online it will have
some bootstrapping knowledge of potential peers (both inter and intra
domain). This information is contained in LSRing file (see <a href="#LSRing"
title="LS Ring File Structure">Section 4.3</a>). The inter-domain
knowledge is used first to establish a connection to an already in progress
ring, or perhaps to start a ring that may not exist yet.</p>
-<p id="rfc.section.2.3.1.p.2">A candidate LS will continuously search it's
LSRing information and send an LSControl mesage to its know LS instances with
a "join" eventType (see <a href="#LSControl-Join" title="LS Joining Message
for Joining a Ring">Section 4.4</a>) until a successful response is
seen. The LS candidate will then search through the successful
LSControlResponse to this message and update its LSRing with the returned
information. This can mean updating the "contentSize" and "active" parameters
as well as adding new LS instances. These parameters are indicative of the
size of each LS (i.e. how many services are registering with it) and
"live-ness" (i.e. were we successful in contacting it recently). The
contacted LS will also update the local copy of LSRing to add the new member
to its "available" list.</p>
-<p id="rfc.section.2.3.1.p.3">After updating, the newly joined LS will
broadcast another LSControl message with a "summary" eventType (see <a
href="#LSControl-Summary-lower" title="LS Summary Message
(Lower)">Section 4.6</a>, or if we are dealing with the upper level see
<a href="#LSControl-Summary-upper" title="LS Summary Message
(Upper)">Section 4.7</a>) to all of the "active" LS instances from its
LSRing. Again the responses will be parsed to get any useful updated
information. At the end of this process joining LS will possess an LSRing
file reflecting the state of the dLS cloud. Each of the recipient LS
instances which hasn't heard anything from this joining LS previously will do
the same, including adding this new member to their own lists (as they didn't
know of it's existence yet).</p>
-<p id="rfc.section.2.3.1.p.4">After this initial warm-up the LS will observe
the rules of token etiquette and remain silent until it is contacted with a
token, or it has not seen one in a very long time (see <a href="#tokens"
title="Token Passing">Section 2.3.2</a>).</p>
-<h4 id="rfc.section.2.3.1.1">
-<a href="#rfc.section.2.3.1.1">2.3.1.1</a> <a id="join_algorithm"
href="#join_algorithm">Join Algorithm</a>
-</h4>
-<p id="rfc.section.2.3.1.1.p.1">
-</p>
-<div id="join-example2">
-</div>
-<div id="rfc.figure.4">
-</div>
-<p>Illustration of LS Join Algorithm</p>
-<pre>
+ </pre><p>The holder of the token (LS3) will inform everyone of
its summary information.</p><p class="figure">Figure 3</p><p
id="rfc.section.2.2.1.p.7">Once exchanged, the details regarding storage in
the XML database backend (see <a href="#glossary"
title="Glossary">Section 6.1</a>) are also left to individual
implementations. It is understood that this information, in the possession of
non HLS instances, is provided as a convenience and should be treated in the
same way that directly registered information is (i.e. purged on expiration).
When responding to queries for this information, the LS must indicate whether
or not it is authoritative.</p><h3 id="rfc.section.2.2.2"><a
href="#rfc.section.2.2.2">2.2.2</a> <a id="upper_scope_summarization"
href="#upper_scope_summarization">Upper Scope Summarization</a></h3><p
id="rfc.section.2.2.2.p.1">A designated member of each lower scope will be
required to interact with the upper scope. The mechanics of how we
learn who is the designated leader are discussed in <a href="#tokens"
title="Token Passing">Section 2.3.2</a>. The leader of each lower scope
(and the designated backup) will be responsible for examining each member's
summary information and building a summarization/aggregation that describes
the contents of a given scope. This summary will serve as input to the
higher-level scope.</p><p id="rfc.section.2.2.2.p.2">The most natural
summarization is based on the topology of the network (like in network
routing). Thus, topology-based summarization will reduce available service
instances in the same way that IP addresses are summarized into network
numbers. They will indicate the eventTypes that a service has and ones that
can it can generate. Summarization will be performed using specialized
summary algorithm. Topology information such as IP addresses will be
summarized using algorithms based on Radix Tree (see <a href="#IP-summary"
title="IP addresses summarization algor
ithm">Section 2.2.2.1</a>).</p><p id="rfc.section.2.2.2!
.p.3">Ot
her information can be summarized in a less programmatic fashion through the
use of either Extensible Stylesheet Language Transformation (XSLT) documents
or the XQuery language as discussed in the previous section. These mechanisms
will take into account the XML elements that represent the network topology
currently used in metadata subjects as well as additional items such as
eventTypes.</p><p id="rfc.section.2.2.2.p.4">The output of this process
becomes a "service summary" that represents a breadth of the original input.
See <a href="#LSControl-Summary-lower" title="LS Summary Message
(Lower)">Section 4.6</a> or <a href="#LSControl-Summary-upper" title="LS
Summary Message (Upper)">Section 4.7</a> for a mock-up of the summary
output. Additional transformations, while aggressive, will strive to preserve
as much information as possible to remain useful during the search
procedures.</p><h4 id="rfc.section.2.2.2.1"><a
href="#rfc.section.2.2.2.1">2.2.2.1</a> <a id=
"IP-summary" href="#IP-summary">IP addresses summarization
algorithm</a></h4><p id="rfc.section.2.2.2.1.p.1">To properly summarize a
list of strings such as IP addresses we must rely on the help of special
algorithms and data structures. The <a
href="http://en.wikipedia.org/wiki/Radix_tree";>Radix Tree</a> algorithm is
useful in creating "associative arrays" where the keys are expressed as
strings. Radix Tree implementations have been used in practice to index
things such as text documents as well as more menial tasks such as
dictionaries. We are most interested in indexing IP addresses with their
natural hierarchy, where it is necessary to describe a large ranges of values
and have few exceptions.</p><p id="rfc.section.2.2.2.1.p.2">A detailed
explanation of the nuances of the Radix Tree is well beyond the scope of this
document, but a brief overview is presented for completeness. The structure
itself is best described as a binary tree where nodes contain whole values
and the
edges are the primary source of navigation. The edges of th!
e tree c
an be based on a single character or perhaps on even longer strings (one of
the features that leads to efficiency for small data sets). The algorithm
should support the following basic operations: </p><ul><li>Lookup: Indicate
that something is or is not contained in the tree.</li><li>Insert: Like in
most inserts we attempt to place something into the structure. We first start
by doing a Lookup to see if it exists already; the point where we stop is
where we will insert the object. We are careful to utilize the longest
matching prefix of any other nearby edge to our advantage. This last part is
normally referred to as "splitting" and ensures that each node has no more
than two children.</li><li>Delete: Delete an object from the tree. This
operation will be complicated by "collapsing" parents that have a single
child and merging the edges.</li></ul><p id="rfc.section.2.2.2.1.p.3">Once
constructed, it is possible to consult the structure in creating IP network
summaries.</p><h2
id="rfc.section.2.3"><a href="#rfc.section.2.3">2.3</a> <a id="scope"
href="#scope">Scope Forming</a></h2><p id="rfc.section.2.3.p.1">The next
question is how to form the lower and upper scopes. The simplest answer is
that the lower scope be formed based on the domain name of the participating
systems. That would allow e.g. internet2.edu, geant2.net, and pionier.gov.pl
to potentially operate more than one LS instance inside their own domains
(for performance and scalability.) As LS instances come online they will
invoke bootstrapping procedures to find and join a lower scoped group
first.</p><p id="rfc.section.2.3.p.2">The scopes should be named based on
URIs. This will allow a domain-level scope to take the form <a
href="http://internet2.edu";>http://internet2.edu</a>, with subdomain scopes
named <a href="http://internet2.edu/foo";>http://internet2.edu/foo</a>, etc.
The top-level scope can be called <a
href="http://perfsonar.net";>http://perfsonar.net</a> with potential f
or geographic divisions later if necessary for performance (!
such as
<a href="http://eu.perfsonar.net";>http://eu.perfsonar.net</a>).</p><p
id="rfc.section.2.3.p.3">The major algorithms used to form and maintain the
ring structure of the dLS, no matter which scope we are talking about, are as
follows: </p><ul><li>Join Procedure</li><li>Token
Passing</li><li>Summarization Notification</li></ul><p> Each of these
procedures is important to keeping members of the distributed "service"
functioning correctly. The algorithms will be presented in the frame of the
lower scope, but will be used in the same manner for the upper scope as
well.</p><h3 id="rfc.section.2.3.1"><a
href="#rfc.section.2.3.1">2.3.1</a> <a id="join" href="#join">Join
Procedure</a></h3><p id="rfc.section.2.3.1.p.1">When an LS instance comes
online it will have some bootstrapping knowledge of potential peers (both
inter and intra domain). This information is contained in LSRing file (see <a
href="#LSRing" title="LS Ring File Structure">Section 4.3</a>). The
inter-domain kno
wledge is used first to establish a connection to an already in progress
ring, or perhaps to start a ring that may not exist yet.</p><p
id="rfc.section.2.3.1.p.2">A candidate LS will continuously search its LSRing
information and send an LSControl mesage to its known LS instances with a
"join" eventType (see <a href="#LSControl-Join" title="LS Joining Message for
Joining a Ring">Section 4.4</a>) until a successful response is seen.
The LS candidate will then search through the successful LSControlResponse to
this message and update its LSRing with the returned information. This can
mean updating the "contentSize" and "active" parameters as well as adding new
LS instances. These parameters are indicative of the size of each LS (i.e.
how many services are registering with it) and "live-ness" (i.e. were we
successful in contacting it recently). The contacted LS will also update the
local copy of LSRing to add the new member to its "available" list.</p><p
id="rfc.section.2.
3.1.p.3">After updating, the newly joined LS will broadcast !
another
LSControl message with a "summary" eventType (see <a
href="#LSControl-Summary-lower" title="LS Summary Message
(Lower)">Section 4.6</a>, or if we are dealing with the upper level see
<a href="#LSControl-Summary-upper" title="LS Summary Message
(Upper)">Section 4.7</a>) to all of the "active" LS instances from its
LSRing. Again the responses will be parsed to get any useful updated
information. At the end of this process the joining LS will possess an LSRing
file reflecting the state of the dLS cloud. Each of the recipient LS
instances which hasn't heard anything from this joining LS previously will do
the same, including adding this new member to their own lists (as they didn't
know of it's existence yet).</p><p id="rfc.section.2.3.1.p.4">After this
initial warm-up the LS will observe the rules of token etiquette and remain
silent until it is contacted with a token, or it has not seen one in a very
long time (see <a href="#tokens" title="Token Passing">Section 
2.3.2</a>).</p><h4 id="rfc.section.2.3.1.1"><a
href="#rfc.section.2.3.1.1">2.3.1.1</a> <a id="join_algorithm"
href="#join_algorithm">Join Algorithm</a></h4><p
id="rfc.section.2.3.1.1.p.1"> </p><div id="join-example2"></div><div
id="rfc.figure.4"></div><p>Illustration of LS Join Algorithm</p><pre>
(5,6)
______________________________________________________
| |
@@ -530,39 +341,7 @@
(5,6)

- </pre>
-<p>LS2, LS3 and LS4 are in the ring. LS1 wants to join the dLS cloud.</p>
-<p class="figure">Figure 4</p>
-<p> <p id="rfc.section.2.3.1.1.p.2">The algorithm for joining the ring
works as follows:</p> </p>
-<dl class="empty">
-<dd>1. LS1 - Candidate sends LSControlRequest message with the
http://perfsonar.net/services/LS/join eventType to selected LS2 which is
already a member of the ring. LS2 was found in LSRing initial file during
bootstrapping process.</dd>
-<dd>2. LS2 receives join message from L1 and decides whether to accept it or
not. A security policy may occur here.</dd>
-<dd>3. LS2 sends the LSControlResponse answer indicating success joining
(result code: http://perfsonar.net/result/success/LS/join) or failure (error
code).</dd>
-<dd>4. After receiving the response from LS2, LS1 parses the results of
LSControlResponse to discover all members of its scope.</dd>
-<dd>5. LS1 broadcast its summary info (lower or upper) with LSControl
message with http://perfsonar.net/services/LS/summary eventType to all of the
LS instances it learns of. This is of course "out of turn" as LS1 doesn't
have the token yet but this reasoning is twofold: <dl class="empty">
-<dd>This expedites all ring members knowing the new LS, the ring will grow
instantly by inserting the new member into the correct position.</dd>
-<dd>This allows the information it provides to the leaders to become
available for the next upper ring summary as soon as possible. Worst case
scenarios would place this knowledge cycles away from being recognized.</dd>
-</dl>
-</dd>
-<dd>6. All members of the ring process the summaries, save the necessary
information, and recognize the new peer. Responses are sent to LS1.</dd>
-</dl>
-<h3 id="rfc.section.2.3.2">
-<a href="#rfc.section.2.3.2">2.3.2</a> <a id="tokens"
href="#tokens">Token Passing</a>
-</h3>
-<p id="rfc.section.2.3.2.p.1">The "token" is an LSControlMessage (see <a
href="#LSControl-Token" title="LS Token Message">Section 4.5</a>) meant
to be passed around an LSRing to the various members in some order. There are
various criterion that can be used in deciding how to order the ring so that
everyone can predict where the token is, when they might expect to get it,
and whom they should get it from/ pass it to next. It is important that we
choose a sound method that is simple to calculate, and should use as much
"knowledge" of the ring as possible without burdening the LS instances too
much with complex calculations.</p>
-<p id="rfc.section.2.3.2.p.2">A common method used in P2P systems such as
Gnutella when forming "ultrapeers" is to consider the size of the data that a
node is serving. The principle, as described <a
href="http://rakjar.de/gnufu/index.php/GnuFU_en#Network_model:_Change_who_calls_whom:_Ultrapeers_and_Leafs";>here</a>,
alludes to the fact that nodes with less content to look after (i.e. less
services, or services with a smaller amount of data) can spend more time and
effort helping the enterprise as a whole by taking on additional roles (such
as serving as leaders). As such we will record the number of metadata
elements that register with each LS and share this with our friends in the
form of the "contentSize" parameter.</p>
-<p id="rfc.section.2.3.2.p.3">Token passing is directly related to the
concept of leader election (see <a href="#Leader_Election" title="Leader
election">Section 2.3.4</a>), so more explanation of this approach will
follow. For now we are justified in saying that the "contentSize" forms a
good criterion for "ordering" the members of the ring. With all members of
the ring aware of everyone's data size, we can easily know who we should pass
the token to, and receive it from at any point in time. We assume passing
order between the LSs from smallest to greatest "contentSize" (with wrap
around at the ends).</p>
-<p id="rfc.section.2.3.2.p.4">The token can be viewed as "permission to
talk" and permits the holding LS to send it's summary information to all
other available LS instances (see <a href="#LSControl-Summary-lower"
title="LS Summary Message (Lower)">Section 4.6</a> and <a
href="#LSControl-Summary-upper" title="LS Summary Message
(Upper)">Section 4.7</a>). The responses will be parsed to get any
useful updated information about current dLS cloud state.</p>
-<p id="rfc.section.2.3.2.p.5">The holder of the token, after completing
summarization, will wait some pre-determined amount of time before sending
the token to the next LS instance. In general the LS instances should not be
overly sensitive to the progression of the token. If each LS instance is
monitoring the progress, and for some reason we have lost the token it may
start a flurry of retransmits and drops that will take cycles to calm down
again. Thus we leave the decisions regarding tokens up to a single node,
namely the designated leader of a scope. We build functionality into leader
nodes to be the only "maker" and "executioner" of tokens. Extra tokens are
dropped/created only by a single node in the ring. All strange thrashing
behavior is avoided and if something bad happens it is eliminated in a single
passing. The leader node will have knowledge of the size of the ring (even if
the ring has grown our join algorithm will inform all interested parties
instantly) and t
he token "wait" period (should be a standard value) thus calculating the
expected time is not an issue.</p>
-<h4 id="rfc.section.2.3.2.1">
-<a href="#rfc.section.2.3.2.1">2.3.2.1</a> <a
id="token_passing_algorithm" href="#token_passing_algorithm">Token Passing
Algorithm</a>
-</h4>
-<div id="join-example">
-</div>
-<div id="rfc.figure.5">
-</div>
-<p>Illustration of Token Passing Algorithm</p>
-<pre>
+ </pre><p>LS2, LS3 and LS4 are in the ring. LS1 wants to join
the dLS cloud.</p><p class="figure">Figure 4</p><p
id="rfc.section.2.3.1.1.p.2">The algorithm for joining the ring works as
follows:</p><p id="rfc.section.2.3.1.1.p.3"> </p><dl class="empty"><dd>1. LS1
- Candidate sends LSControlRequest message with the
http://perfsonar.net/services/LS/join eventType to selected LS2 which is
already a member of the ring. LS2 was found in LSRing initial file during
bootstrapping process.</dd><dd>2. LS2 receives join message from L1 and
decides whether to accept it or not. Application of security policy may occur
here.</dd><dd>3. LS2 sends the LSControlResponse answer indicating success
joining (result code: http://perfsonar.net/result/success/LS/join) or failure
(error code).</dd><dd>4. After receiving the response from LS2, LS1 parses
the results of LSControlResponse to discover all members of its
scope.</dd><dd>5. LS1 broadcast its summary info (lower or upper) with L
SControl message with http://perfsonar.net/services/LS/summary eventType to
all of the LS instances it learns of. This is of course "out of turn" as LS1
doesn't have the token yet but this reasoning is twofold: <dl
class="empty"><dd>This expedites all ring members knowing the new LS, the
ring will grow instantly by inserting the new member into the correct
position.</dd><dd>This allows the information it provides to the leaders to
become available for the next upper ring summary as soon as possible. Worst
case scenarios would place this knowledge cycles away from being
recognized.</dd></dl> </dd><dd>6. All members of the ring process the
summaries, save the necessary information, and recognize the new peer.
Responses are sent to LS1.</dd></dl><h3 id="rfc.section.2.3.2"><a
href="#rfc.section.2.3.2">2.3.2</a> <a id="tokens" href="#tokens">Token
Passing</a></h3><p id="rfc.section.2.3.2.p.1">The "token" is an
LSControlMessage (see <a href="#LSControl-Token" title="LS Token
Message">Section 4.5</a>) meant to be passed around an !
LSRing t
o the various members in some order. There are various criteria that can be
used in deciding how to order the ring so that everyone can predict where the
token is, when they might expect to get it, and whom they should get it from/
pass it to next. It is important that we choose a sound method that is simple
to calculate, and should use as much "knowledge" of the ring as possible
without burdening the LS instances too much with complex calculations.</p><p
id="rfc.section.2.3.2.p.2">The essential idea in the token passing mechanism
for leader election is that some identifier is chosen for each node and that
the node with the highest (or lowest) identifier win the election and becomes
the leader. The basic mechanism of leader election is that participants form
a logical ring and initiate an election. An election can be initiated when a
new machine joins, at system start time, or when a host feels that the leader
may have failed based on failure to receive a periodic token. When
an election is initiated, the initiating host sends an election message to
its counter-clockwise neighbor and changes its state to
“ELECTING”. It places its identifier inside the message. The
ulimate goal is for the host with the highest identifier to be chosen. When a
host receives an election message, it compares its identifier with that in
the message. It forwards the higher of the identifiers. When a node receives
a message with its own identifier, it knows that it has been selected and the
election terminates.</p><p id="rfc.section.2.3.2.p.3">The next question is
how to choose the identifier for a given node. There still needs to be some
discussion here. The first proposal was to use the IP address of the node as
the lower-order 32-bits of a 64-bit number and to allow the higher-order bits
to be set as a "priority" field. This would effectively allow a system
administrator to make sure that her most powerful or well-connected nodes
became the leader when th
ey were available. In the absence of a priority, the nodes e!
ssential
ly are randomly ordered.</p><p id="rfc.section.2.3.2.p.4">Another possibility
is to use common method used in P2P systems such as Gnutella when forming
"ultrapeers" is to consider the size of the data that a node is serving. The
principle, as described <a
href="http://rakjar.de/gnufu/index.php/GnuFU_en#Network_model:_Change_who_calls_whom:_Ultrapeers_and_Leafs";>here</a>,
alludes to the fact that nodes with less content to look after (i.e. less
services, or services with a smaller amount of data) can spend more time and
effort helping the enterprise as a whole by taking on additional roles (such
as serving as leaders). As such we will record the number of metadata
elements that register with each LS and share this with our friends in the
form of the "contentSize" parameter.</p><p id="rfc.section.2.3.2.p.5">Token
passing is directly related to the concept of leader election (see <a
href="#Leader_Election" title="Leader election">Section 2.3.4</a>), so
more explanation of t
his approach will follow. For now we are justified in saying that the
"contentSize" forms another good criterion for "ordering" the members of the
ring. With all members of the ring aware of everyone's data size, we can
easily know who we should pass the token to, and receive it from at any point
in time. We assume passing order between the LSs from smallest to greatest
"contentSize" (with wrap around at the ends). Of course, this metric leaves
out the relative power of the machines in question, so futher modifications
are possible.</p><p id="rfc.section.2.3.2.p.6">The key is that as long as the
identifier is chosen consistently within a given scope, the choice of
identifier doesn't affect the operation of the protocol.</p><p
id="rfc.section.2.3.2.p.7">The token can be viewed as "permission to talk"
and permits the holding LS to send its summary information to all other
available LS instances (see <a href="#LSControl-Summary-lower" title="LS
Summary Message (Lower)">Section&
nbsp;4.6</a> and <a href="#LSControl-Summary-upper" title="L!
S Summar
y Message (Upper)">Section 4.7</a>). The responses will be parsed to get
any useful updated information about current dLS cloud state.</p><h4
id="rfc.section.2.3.2.1"><a href="#rfc.section.2.3.2.1">2.3.2.1</a> <a
id="token_passing_algorithm" href="#token_passing_algorithm">Token Passing
Algorithm</a></h4><div id="join-example"></div><div
id="rfc.figure.5"></div><p>Illustration of Token Passing Algorithm</p><pre>

_____________(3)____________
| |
@@ -580,106 +359,7 @@

- </pre>
-<p>LS1, LS2 and LS3 are in the ring. LS1 receives a token.</p>
-<p class="figure">Figure 5</p>
-<p id="rfc.section.2.3.2.1.p.2">The algorithm for token passing works as
follows:</p>
-<dl class="empty">
-<dd>1. LS1 receives the token i.e. LSControlRequest message with the
http://perfsonar.net/services/LS/token/ eventType from its predecessor
L3.</dd>
-<dd>2. LS1 updates its local peer list based on token content.</dd>
-<dd>2.x (More description needed)</dd>
-<dd>2.1 If token contains entry that is not present in the local LSRing,
copy this entry to local LSRing</dd>
-<dd>2.2 If token doesn't contains entry that is present in the local
LSRing... (to be discussed: there are two cases: new LS just joined to LS
that has got the token right now OR local LSRing contains obsolete LS entry
-- I guess we'll base on active parameter, but it's not elaborated how it
works</dd>
-<dd>3. LS1 sends LSControlRequest message with the
http://perfsonar.net/services/LS/summary/ eventType to all peers in the lease
(excluding itself).</dd>
-<dd>3. LS2 receiving this message checks its collection and updates it if
necessary with service info including "contentSize".</dd>
-<dd>4. LS1 waits for some amount of time.</dd>
-<dd>5. LS1 sends token to next LS (LS2) from the LSRing lower scope (if it
fails, try next one).</dd>
-</dl>
-<h4 id="rfc.section.2.3.2.2">
-<a href="#rfc.section.2.3.2.2">2.3.2.2</a> <a
id="rotation-time-computing" href="#rotation-time-computing">Token rotation
time computing</a>
-</h4>
-<h3 id="rfc.section.2.3.3">
-<a href="#rfc.section.2.3.3">2.3.3</a> <a id="summary-blast"
href="#summary-blast">Summarization Notification</a>
-</h3>
-<p id="rfc.section.2.3.3.p.1">As discussed in the prior two sections there
are two acceptable instances to send your summary to the LSRing: </p>
-<ol>
-<li>When first joining</li>
-<li>When holding the token</li>
-</ol>
-<p id="rfc.section.2.3.3.p.2">In the first case we are explicitly entering
ourselves into the ring when we get our first message from a peer. This
ensures we show up in the token rotation instantly. The second case is the
routine exchange started when a token arrives from a peer.</p>
-<p id="rfc.section.2.3.3.p.3">
-<a href="#LSControl-Summary-lower" title="LS Summary Message
(Lower)">Section 4.6</a> and <a href="#LSControl-Summary-upper"
title="LS Summary Message (Upper)">Section 4.7</a> contain examples of
the message format for this exchange. It is left up to the implementation
when the summarization occurs (i.e. at message send time, or also as a
periodic event).</p>
-<h3 id="rfc.section.2.3.4">
-<a href="#rfc.section.2.3.4">2.3.4</a> <a id="Leader_Election"
href="#Leader_Election">Leader election</a>
-</h3>
-<p id="rfc.section.2.3.4.p.1">The most important role of the lower scope
nodes is electing a leader to serve in the upper levels. This logical ring
should consist of one representative LS from each of the lower scopes. This
representative will be selected based on the "contentSize" parameter, which
as we have discussed previously is a count of the number of metadata elements
the LS is responsible for. We choose this parameter primarily because it
imposes a very simple ordering on the nodes, and allows us to choose the
"least busy" node for performing more important work.</p>
-<p id="rfc.section.2.3.4.p.2">The theory behind this choice of leader is
that unladen LS instances will not be as loaded processing requests and
responses from clients and other services, thus the choice to name them as
leaders is natural. An added feature of this criteria is that it becomes very
simple to designate a leader LS for a domain; simply deploy an LS instance
that does not accept registrations. This service will only serve in the role
of liaison to the upper level.</p>
-<p id="rfc.section.2.3.4.p.3">All LS instances will have a complete view at
any given time (and updated over time) of the values of the "contentSize"
parameter for all peer LS instances in an LSRing. This ensures that the
"Leader" and "Vice-Leader" are always known. Explicit election is therefore
not required, and succession can pass between the two executives quickly.</p>
-<p id="rfc.section.2.3.4.p.4">The Leader and Vice-Leader LS instances should
exchange messages (see <a href="#LSControl-Leader" title="LS Leader
Message">Section 4.8</a>) periodically to ensure that in the event of a
failure the lower level will still have a link to the upper level. A
Vice-Leader will be monitoring the time between successive communications
from the Leader to be sure it has not failed. In the event that it has, the
"Join" procedure will start to the upper level to keep the hierarchy
complete.</p>
-<h3 id="rfc.section.2.3.5">
-<a href="#rfc.section.2.3.5">2.3.5</a> <a id="scopes"
href="#scopes">Scopes</a>
-</h3>
-<p id="rfc.section.2.3.5.p.1">Scopes are named based on URIs. The top-level
domain provides a natural basis for the formation of these URIs. These URIs
may be constructed to allow internal differentiation. In the future, we may
need a mechanism to provide another level of hierarchy above the domain level
and below the root, but that is left for future work. Note: I would like to
see this be something like: <a
href="http://lsls.perfsonar.net/master/internet2.edu/";>http://lsls.perfsonar.net/master/internet2.edu/</a>.
Actual syntax doesn't matter that much but I would like the following
components: </p>
-<ol>
-<li>Well-known hostname to get the current root.hints from (lsls)</li>
-<li>A place holder for where we can break the scope between organization and
upper levels (master)</li>
-<li>A zero-conf default name for the organization (internet2.edu)</li>
-<li>A way to sub-divide the organization (everything after trailing / )</li>
-</ol>
-<h2 id="rfc.section.2.4">
-<a href="#rfc.section.2.4">2.4</a> <a id="search"
href="#search">Search</a>
-</h2>
-<p id="rfc.section.2.4.p.1">The search operation is obviously critical to
the Lookup Service's function. It is envisioned that searching could take one
of two major forms, iterative and recursive, analogous to those used in DNS.
This design will focus exclusively on iterative initially as the only method
in the first versions of the dLS. The key act when searching is to find what
eventTypes exist for a particular topology element or set of topology
elements.</p>
-<p id="rfc.section.2.4.p.2">As outlined above, the full data that services
register to an LS is not expected to leave the scope of that LS. The
information is summarized before wider distribution. Therefore, a client
needs to find an LS in the scope of the HLS to make queries about the
complete metadata. Specifically, a client wishing to locate information might
specify a topology element in a query to locate the LS instance (or
instances) that contain the relevant details. This separation of full data
and summary data means the overall act of searching is broken down into two
distinct phases - Discovery and Metadata Query.</p>
-<h3 id="rfc.section.2.4.1">
-<a href="#rfc.section.2.4.1">2.4.1</a> <a id="discovery"
href="#discovery">Discovery Phase</a>
-</h3>
-<p id="rfc.section.2.4.1.p.1">The discovery phase is used to locate the set
of Authoritative LS (or LSes) for a given Subject/eventType tuple. This
requires a query to be constructed over the Discovery information set (which
is not described yet, but which must consist of the 3-tuple of Subject
Summary, eventType and Authoritative LS.) Either a specific API call and a
pre-prepared query, or some automatic mechanism, must map the desired query
into a query of the Discovery infoset (see <a href="#LSControl-Discovery"
title="LS Discovery Message">Section 4.9</a>).</p>
-<h4 id="rfc.section.2.4.1.1">
-<a href="#rfc.section.2.4.1.1">2.4.1.1</a> <a id="discovery-alg"
href="#discovery-alg">Discovery Algorithm</a>
-</h4>
-<p id="rfc.section.2.4.1.1.p.1">The discovery algorithm is as follows.</p>
-<ol>
-<li>A client locates an LS of some sort (this may be known beforehand via a
configuration value, or from bootstrapping).</li>
-<li>The client should start by making a discovery query (possibly using an
API call) to locate an LS that contains the data it is interested in. The
results of this query will be: <ol style="list-style-type: lower-alpha">
-<li>Failure: Returned if there is no LS at a higher scope than the current
one, and nothing was found in the summary infoset that matches the query.</li>
-<li>Referral: This is returned when there is no match other than a "global
wildcard" <ol style="list-style-type: lower-alpha">
-<li>If this LS is not participating in the highest (upper) scope, then it
returns the leader of its current scope (or a direct referral to an instance
of the next-higher scope.) This is effectively a wildcard match saying "I
don't know the answer, but I know who might." This is how the Metadata
registered to an LS in another scope (domain) is found.</li>
-</ol>
-</li>
-<li>Success: We define success to mean at least one matching LS has been
returned. The LS must return the following: <ol style="list-style-type:
lower-alpha">
-<li>If this LS is an HLS for the discovery query, it returns itself.</li>
-<li>This LS also returns any other HLS instances it has found that match. An
LS instance will have summary information from other domains when it is
participating in a higher-level scope (such as upper).</li>
-<li>Note: this is where recursive searches would be added into the discovery
phase. The trail up the scope hierarchy would be followed by the LS itself
instead of returning the leader LS. Ideally, this list would be iterated on
by the LS so that only leaf LS instances are returned</li>
-</ol>
-</li>
-</ol>
-</li>
-<li>The client will need to iterate through the list of returned LS
instances. If the LS returns itself, this LS can be used in the following
Metadata query phase. If the returned LS is different, a discovery query
should be made to it.</li>
-</ol>
-<h3 id="rfc.section.2.4.2">
-<a href="#rfc.section.2.4.2">2.4.2</a> <a id="metadata-query"
href="#metadata-query">Metadata Query Phase</a>
-</h3>
-<p id="rfc.section.2.4.2.p.1">The Metadata Query Phase with an individual LS
is the same as the query mechanism that is in place with the current LS
implementations.</p>
-<p id="rfc.section.2.4.2.p.2">Once we have found the HLS (or Home LSes) that
contain data in the range of our discovery query, we can pose Metadata
Queries to each of them. The results will be failure or success.</p>
-<hr class="noprint">
-<h1 id="rfc.section.3" class="np">
-<a href="#rfc.section.3">3.</a> <a id="bootstrapping"
href="#bootstrapping">Bootstrapping</a>
-</h1>
-<p id="rfc.section.3.p.1">A distributed information system such as the LS
needs to address bootstrapping. In this system, an LS instance needs to find
other members of its scope (for each scope in which it participates.) To
accomplish this we will use a similar solution to what DNS uses
(root.hints).</p>
-<p id="rfc.section.3.p.2">We will maintain a service that maintains a list
of currently known LS instances. These known instances should preferably be
at the upper scope. All clients can cache this list. The service will be
accessed via a well-known hostname, and could be requested via UDP messages.
(We can also use TCP here for some sorts of anycast.)</p>
-<p id="rfc.section.3.p.3">Initially this will be deployed on one server. We
can extend this to handle redundancy and load balancing in the future by
using multiple DNS records and implementing ANYCAST with routing tricks for
this well known hostname. (Additionally, we can distribute an initial file
with a list of well known LS instances that are supported by the primary
perfSONAR participants.)</p>
-<p id="rfc.section.3.p.4">The above discovery algorithm is used to find an
LS within a given scope. Therefore, the only piece of information an LS
should need to be pre-configured with is the scope it belongs to. And as
stated above, that can be assumed to be "global:organization-dns-name". Note:
Need to define the specific syntax above.</p>
-<hr class="noprint">
-<h1 id="rfc.section.4" class="np">
-<a href="#rfc.section.4">4.</a> <a id="structures-and-messages"
href="#structures-and-messages">Structures and Messages</a>
-</h1>
-<h2 id="rfc.section.4.1">
-<a href="#rfc.section.4.1">4.1</a> <a id="service-metadata"
href="#service-metadata">Service metadata example</a>
-</h2>
-<p id="rfc.section.4.1.p.1">Example of metadata describing information
collected and stored in Measurement Archive service</p>
-<p id="rfc.section.4.1.p.2">
-<pre>
+ </pre><p>LS1, LS2 and LS3 are in the ring. LS1 receives a
token.</p><p class="figure">Figure 5</p><p id="rfc.section.2.3.2.1.p.2">The
algorithm for token passing works as follows:</p><p
id="rfc.section.2.3.2.1.p.3"> </p><dl class="empty"><dd>1. LS1 receives the
token i.e. LSControlRequest message with the
http://perfsonar.net/services/LS/token/ eventType from its predecessor
L3.</dd><dd>2. LS1 updates its local peer list based on token
content.</dd><dd>2.x (More description needed)</dd><dd>2.1 If token contains
entry that is not present in the local LSRing, copy this entry to local
LSRing</dd><dd>2.2 If token doesn't contains entry that is present in the
local LSRing... (to be discussed: there are two cases: new LS just joined to
LS that has got the token right now OR local LSRing contains obsolete LS
entry -- I guess we'll base on active parameter, but it's not elaborated how
it works</dd><dd>3. LS1 sends LSControlRequest message with the
http://perfsonar.net/s
ervices/LS/summary/ eventType to all peers in the lease (excluding
itself).</dd><dd>3. LS2 receiving this message checks its collection and
updates it if necessary with service info including "contentSize".</dd><dd>4.
LS1 waits for some amount of time.</dd><dd>5. LS1 sends token to next LS
(LS2) from the LSRing lower scope (if it fails, try next one).</dd></dl><h4
id="rfc.section.2.3.2.2"><a href="#rfc.section.2.3.2.2">2.3.2.2</a> <a
id="rotation-time-computing" href="#rotation-time-computing">Token rotation
time computing</a></h4><p id="rfc.section.2.3.2.2.p.1">The token rotation
time is the time of passing and serving token by all nodes in the LS ring.
This time should be computed by the leader basing on some knowledge about the
time of serving token by all particular nodes. The time may be based on times
saving in token message by all nodes. Initially, this can be very simple as
in "2 minutes X the number of nodes in the ring." (To be discussed)</p><p
id="rfc.section
.2.3.2.2.p.2">The key is that after the timeout has exceeded!
, it can
be inferred that the leader has failed and another election should be
initiated.</p><h3 id="rfc.section.2.3.3"><a
href="#rfc.section.2.3.3">2.3.3</a> <a id="summary-blast"
href="#summary-blast">Summarization Notification</a></h3><p
id="rfc.section.2.3.3.p.1">As discussed in the prior two sections there are
two acceptable instances to send your summary to the LSRing: </p><ol><li>When
first joining</li><li>When holding the token</li></ol><p
id="rfc.section.2.3.3.p.2">In the first case we are explicitly entering
ourselves into the ring when we get our first message from a peer. This
ensures we show up in the token rotation instantly. The second case is the
routine exchange started when a token arrives from a peer.</p><p
id="rfc.section.2.3.3.p.3"> <a href="#LSControl-Summary-lower" title="LS
Summary Message (Lower)">Section 4.6</a> and <a
href="#LSControl-Summary-upper" title="LS Summary Message
(Upper)">Section 4.7</a> contain examples of the message format for
this exchange. It is left up to the implementation when the summarization
occurs (i.e. at message send time, or also as a periodic event).</p><h3
id="rfc.section.2.3.4"><a href="#rfc.section.2.3.4">2.3.4</a> <a
id="Leader_Election" href="#Leader_Election">Leader election</a></h3><p
id="rfc.section.2.3.4.p.1">The most important role of the lower scope nodes
is electing a leader to serve in the upper levels. This logical ring should
consist of one representative LS from each of the lower scopes. </p><p
id="rfc.section.2.3.4.p.2">The Leader and Vice-Leader LS instances should
exchange messages (see <a href="#LSControl-Leader" title="LS Leader
Message">Section 4.8</a>) periodically to ensure that in the event of a
failure the lower level will still have a link to the upper level. A
Vice-Leader will be monitoring the time between successive communications
from the Leader to be sure it has not failed. In the event that it has, the
"Join" procedure will start to the upper
level to keep the hierarchy complete.</p><h2 id="rfc.sectio!
n.2.4"><
a href="#rfc.section.2.4">2.4</a> <a id="search"
href="#search">Search</a></h2><p id="rfc.section.2.4.p.1">The search
operation is obviously critical to the Lookup Service's function. It is
envisioned that searching could take one of two major forms, iterative and
recursive, analogous to those used in DNS. This design will focus exclusively
on iterative initially as the only method in the first versions of the dLS.
The key act when searching is to find what eventTypes exist for a particular
topology element or set of topology elements.</p><p
id="rfc.section.2.4.p.2">As outlined above, the full data that services
register to an LS is not expected to leave the scope of that LS. The
information is summarized before wider distribution. Therefore, a client
needs to find an LS in the scope of the HLS to make queries about the
complete metadata. Specifically, a client wishing to locate information might
specify a topology element in a query to locate the LS instance (or instanc
es) that contain the relevant details. This separation of full data and
summary data means the overall act of searching is broken down into two
distinct phases - Discovery and Metadata Query.</p><h3
id="rfc.section.2.4.1"><a href="#rfc.section.2.4.1">2.4.1</a> <a
id="discovery" href="#discovery">Discovery Phase</a></h3><p
id="rfc.section.2.4.1.p.1">The discovery phase is used to locate the set of
Authoritative LS (or LSes) for a given Subject/eventType tuple. This requires
a query to be constructed over the Discovery information set (which is not
described yet, but which must consist of the 3-tuple of Subject Summary,
eventType and Authoritative LS.) Either a specific API call and a
pre-prepared query, or some automatic mechanism, must map the desired query
into a query of the Discovery infoset (see <a href="#LSControl-Discovery"
title="LS Discovery Message">Section 4.9</a>).</p><h4
id="rfc.section.2.4.1.1"><a href="#rfc.section.2.4.1.1">2.4.1.1</a> <a
id="dis
covery-alg" href="#discovery-alg">Discovery Algorithm</a></h!
4><p id=
"rfc.section.2.4.1.1.p.1">The discovery algorithm is as follows.</p><ol><li>A
client locates an LS of some sort (this may be known beforehand via a
configuration value, or from bootstrapping).</li><li>The client should start
by making a discovery query (possibly using an API call) to locate an LS that
contains the data it is interested in. The results of this query will be: <ol
style="list-style-type: lower-alpha"><li>Failure: Returned if there is no LS
at a higher scope than the current one, and nothing was found in the summary
infoset that matches the query.</li><li>Referral: This is returned when there
is no match other than a "global wildcard" <ol style="list-style-type:
lower-alpha"><li>If this LS is not participating in the highest (upper)
scope, then it returns the leader of its current scope (or a direct referral
to an instance of the next-higher scope.) This is effectively a wildcard
match saying "I don't know the answer, but I know who might." This is how the
Metada
ta registered to an LS in another scope (domain) is found.</li></ol>
</li><li>Success: We define success to mean at least one matching LS has been
returned. The LS must return the following: <ol style="list-style-type:
lower-alpha"><li>If this LS is an HLS for the discovery query, it returns
itself.</li><li>This LS also returns any other HLS instances it has found
that match. An LS instance will have summary information from other domains
when it is participating in a higher-level scope (such as
upper).</li><li>Note: this is where recursive searches would be added into
the discovery phase. The trail up the scope hierarchy would be followed by
the LS itself instead of returning the leader LS. Ideally, this list would be
iterated on by the LS so that only leaf LS instances are returned</li></ol>
</li></ol> </li><li>The client will need to iterate through the list of
returned LS instances. If the LS returns itself, this LS can be used in the
following Metadata query phase. If t
he returned LS is different, a discovery query should be mad!
e to it.
</li></ol><h3 id="rfc.section.2.4.2"><a
href="#rfc.section.2.4.2">2.4.2</a> <a id="metadata-query"
href="#metadata-query">Metadata Query Phase</a></h3><p
id="rfc.section.2.4.2.p.1">The Metadata Query Phase with an individual LS is
the same as the query mechanism that is in place with the current LS
implementations.</p><p id="rfc.section.2.4.2.p.2">Once we have found the HLS
(or Home LSes) that contain data in the range of our discovery query, we can
pose Metadata Queries to each of them. The results will be failure or
success.</p><hr class="noprint"><h1 id="rfc.section.3" class="np"><a
href="#rfc.section.3">3.</a> <a id="bootstrapping"
href="#bootstrapping">Bootstrapping</a></h1><p id="rfc.section.3.p.1">A
distributed information system such as the LS needs to address bootstrapping.
In this system, an LS instance needs to find other members of its scope (for
each scope in which it participates.) To accomplish this we will use a
similar solution to what DNS uses (roo
t.hints).</p><p id="rfc.section.3.p.2">We will maintain a service that
maintains a list of currently known LS instances. These known instances
should preferably be at the upper scope. All clients can cache this list. The
service will be accessed via a well-known hostname, and could be requested
via UDP messages. (We can also use TCP here for some sorts of anycast.)</p><p
id="rfc.section.3.p.3">Initially this will be deployed on one server. We can
extend this to handle redundancy and load balancing in the future by using
multiple DNS records and implementing ANYCAST with routing tricks for this
well known hostname. (Additionally, we can distribute an initial file with a
list of well known LS instances that are supported by the primary perfSONAR
participants.)</p><p id="rfc.section.3.p.4">The above discovery algorithm is
used to find an LS within a given scope. Therefore, the only piece of
information an LS should need to be pre-configured with is the scope it
belongs to. And
as stated above, that can be assumed to be "global:organizat!
ion-dns-
name". Note: Need to define the specific syntax above.</p><hr
class="noprint"><h1 id="rfc.section.4" class="np"><a
href="#rfc.section.4">4.</a> <a id="structures-and-messages"
href="#structures-and-messages">Structures and Messages</a></h1><h2
id="rfc.section.4.1"><a href="#rfc.section.4.1">4.1</a> <a
id="service-metadata" href="#service-metadata">Service metadata
example</a></h2><p id="rfc.section.4.1.p.1">Example of metadata describing
information collected and stored in Measurement Archive service</p><p
id="rfc.section.4.1.p.2"> <pre>

<nmwg:metadata xmlns:nmwg="http://ggf.org/ns/nmwg/base/2.0/";
id="m_ale-netutil-1">
<netutil:subject
xmlns:netutil="http://ggf.org/ns/nmwg/characteristic/utilization/2.0/";
id="s_ale-netutil-1">
@@ -700,14 +380,7 @@
</nmwg:parameters>
</nmwg:metadata>

- </pre>
-</p>
-<h2 id="rfc.section.4.2">
-<a href="#rfc.section.4.2">4.2</a> <a id="lookup-info"
href="#lookup-info">Lookup Information</a>
-</h2>
-<p id="rfc.section.4.2.p.1">Example Lookup Information of Measurement
Archive. The metadata block contains basic service information and data
elements containing the metadata from the MA.</p>
-<p id="rfc.section.4.2.p.2">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.2"><a
href="#rfc.section.4.2">4.2</a> <a id="lookup-info"
href="#lookup-info">Lookup Information</a></h2><p
id="rfc.section.4.2.p.1">Example Lookup Information of Measurement Archive.
The metadata block contains basic service information and data elements
containing the metadata from the MA.</p><p id="rfc.section.4.2.p.2"> <pre>

<nmwg:metadata
id="http://newcastle.pc.cis.udel.edu:6767/perfSONAR_PS/services/snmpMA">
<perfsonar:subject id="subject.15977808">
@@ -743,18 +416,7 @@
</nmwg:data>

- </pre>
-</p>
-<h2 id="rfc.section.4.3">
-<a href="#rfc.section.4.3">4.3</a> <a id="LSRing" href="#LSRing">LS
Ring File Structure</a>
-</h2>
-<p id="rfc.section.4.3.p.1">The LSRing file represents the "state" of the LS
cloud at either level of hierarchy (we avoid using the terms "global" and
"local" here since the hierarchy may be much larger). This file must start
with some static values, and will be added to/deleted from as time goes on.
As such implementations must ensure that this file is under database control
of some sort.</p>
-<p id="rfc.section.4.3.p.2">Each metadata description contains the usual
service level info, as well as a "voting" parameter (labeled as
"contentSize") used to decide who the leader is, and a flag indicating if the
current service thinks the remote service is active. Both of these should be
updated frequently as reality changes.</p>
-<h3 id="rfc.section.4.3.1">
-<a href="#rfc.section.4.3.1">4.3.1</a> <a id="LSRingLower"
href="#LSRingLower">LS Ring lower level</a>
-</h3>
-<p id="rfc.section.4.3.1.p.1">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.3"><a
href="#rfc.section.4.3">4.3</a> <a id="LSRing" href="#LSRing">LS Ring
File Structure</a></h2><p id="rfc.section.4.3.p.1">The LSRing file represents
the "state" of the LS cloud at either level of hierarchy (we avoid using the
terms "global" and "local" here since the hierarchy may be much larger). This
file must start with some static values, and will be added to/deleted from as
time goes on. As such implementations must ensure that this file is under
database control of some sort.</p><h3 id="rfc.section.4.3.1"><a
href="#rfc.section.4.3.1">4.3.1</a> <a id="LSRingLower"
href="#LSRingLower">LS Ring lower level</a></h3><p
id="rfc.section.4.3.1.p.1"> <pre>

<nmwg:store type="LSRing-lower">

@@ -822,13 +484,7 @@

</nmwg:store>

- </pre>
-</p>
-<h3 id="rfc.section.4.3.2">
-<a href="#rfc.section.4.3.2">4.3.2</a> <a id="LSRingUpper"
href="#LSRingUpper">LS Ring upper level</a>
-</h3>
-<p id="rfc.section.4.3.2.p.1">
-<pre>
+ </pre> </p><h3 id="rfc.section.4.3.2"><a
href="#rfc.section.4.3.2">4.3.2</a> <a id="LSRingUpper"
href="#LSRingUpper">LS Ring upper level</a></h3><p
id="rfc.section.4.3.2.p.1"> <pre>

<nmwg:store type="LSRing-upper">

@@ -866,18 +522,7 @@

</nmwg:store>

- </pre>
-</p>
-<h2 id="rfc.section.4.4">
-<a href="#rfc.section.4.4">4.4</a> <a id="LSControl-Join"
href="#LSControl-Join">LS Joining Message for Joining a Ring</a>
-</h2>
-<p id="rfc.section.4.4.p.1">This message exchange represents when a "new" LS
instance comes online. The LS will send these messages to its "list" of known
LS instances until it gets a hit. The message consists of metadata/data
pair(s) that contain service information and a parameter indicating "size" of
the data set the LS manages. This will be used for leader voting purposes
later.</p>
-<p id="rfc.section.4.4.p.2">The response message should indicate success or
failure via the eventType, and will contain metadata/data pair(s). The
metadata should indicate who the service is, and its "size" for voting
purposes. The data section is an enumeration of all of the current members of
the ring and their votes. This information gives the new member a snapshot of
the ring.</p>
-<h3 id="rfc.section.4.4.1">
-<a href="#rfc.section.4.4.1">4.4.1</a> <a id="LSControl-JoinRequest"
href="#LSControl-JoinRequest">Request</a>
-</h3>
-<p id="rfc.section.4.4.1.p.1">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.4"><a
href="#rfc.section.4.4">4.4</a> <a id="LSControl-Join"
href="#LSControl-Join">LS Joining Message for Joining a Ring</a></h2><p
id="rfc.section.4.4.p.1">This message exchange represents when a "new" LS
instance comes online. The LS will send these messages to its "list" of known
LS instances until it gets a hit. The message consists of metadata/data
pair(s) that contain service information and a parameter indicating "size" of
the data set the LS manages. This will be used for leader voting purposes
later.</p><p id="rfc.section.4.4.p.2">The response message should indicate
success or failure via the eventType, and will contain metadata/data pair(s).
The metadata should indicate who the service is, and its "size" for voting
purposes. The data section is an enumeration of all of the current members of
the ring and their votes. This information gives the new member a snapshot of
the ring.</p><h3 id="rfc.section.4.4.1"><a href="#rf
c.section.4.4.1">4.4.1</a> <a id="LSControl-JoinRequest"
href="#LSControl-JoinRequest">Request</a></h3><p id="rfc.section.4.4.1.p.1">
<pre>

<nmwg:message type="LSControlRequest">

@@ -900,13 +545,7 @@

</nmwg:message>

- </pre>
-</p>
-<h3 id="rfc.section.4.4.2">
-<a href="#rfc.section.4.4.2">4.4.2</a> <a id="LSControl-JoinResponse"
href="#LSControl-JoinResponse">Response</a>
-</h3>
-<p id="rfc.section.4.4.2.p.1">
-<pre>
+ </pre> </p><h3 id="rfc.section.4.4.2"><a
href="#rfc.section.4.4.2">4.4.2</a> <a id="LSControl-JoinResponse"
href="#LSControl-JoinResponse">Response</a></h3><p
id="rfc.section.4.4.2.p.1"> <pre>

<nmwg:message type="LSControlResponse">

@@ -957,18 +596,7 @@

</nmwg:message>

- </pre>
-</p>
-<h2 id="rfc.section.4.5">
-<a href="#rfc.section.4.5">4.5</a> <a id="LSControl-Token"
href="#LSControl-Token">LS Token Message</a>
-</h2>
-<p id="rfc.section.4.5.p.1">This message exchange represents the token that
is passed between LS instances in a cloud. The message contains metadata/data
pair(s) wherein the Metadata is the sending LS's info, and the data contains
the contents of the LSRing file (lower or upper depending on the token we are
exchanging).</p>
-<p id="rfc.section.4.5.p.2">The response to this message should indicate
success or failure. Failure and timeouts should trigger a resend.</p>
-<h3 id="rfc.section.4.5.1">
-<a href="#rfc.section.4.5.1">4.5.1</a> <a id="LSControl-TokenRequest"
href="#LSControl-TokenRequest">Request</a>
-</h3>
-<p id="rfc.section.4.5.1.p.1">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.5"><a
href="#rfc.section.4.5">4.5</a> <a id="LSControl-Token"
href="#LSControl-Token">LS Token Message</a></h2><p
id="rfc.section.4.5.p.1">This message exchange represents the token that is
passed between LS instances in a cloud. The message contains metadata/data
pair(s) wherein the Metadata is the sending LS's info, and the data contains
the contents of the LSRing file (lower or upper depending on the token we are
exchanging).</p><p id="rfc.section.4.5.p.2">The response to this message
should indicate success or failure. Failure and timeouts should trigger a
resend.</p><h3 id="rfc.section.4.5.1"><a
href="#rfc.section.4.5.1">4.5.1</a> <a id="LSControl-TokenRequest"
href="#LSControl-TokenRequest">Request</a></h3><p id="rfc.section.4.5.1.p.1">
<pre>

<nmwg:message type="LSControlRequest">

@@ -1016,13 +644,7 @@

</nmwg:message>

- </pre>
-</p>
-<h3 id="rfc.section.4.5.2">
-<a href="#rfc.section.4.5.2">4.5.2</a> <a id="LSControl-TokenResponse"
href="#LSControl-TokenResponse">Response</a>
-</h3>
-<p id="rfc.section.4.5.2.p.1">
-<pre>
+ </pre> </p><h3 id="rfc.section.4.5.2"><a
href="#rfc.section.4.5.2">4.5.2</a> <a id="LSControl-TokenResponse"
href="#LSControl-TokenResponse">Response</a></h3><p
id="rfc.section.4.5.2.p.1"> <pre>

<nmwg:message type="LSControlResponse">

@@ -1048,24 +670,7 @@

</nmwg:message>

- </pre>
-</p>
-<h2 id="rfc.section.4.6">
-<a href="#rfc.section.4.6">4.6</a> <a id="LSControl-Summary-lower"
href="#LSControl-Summary-lower">LS Summary Message (Lower)</a>
-</h2>
-<p id="rfc.section.4.6.p.1">This message exchange represents when an LS
instance is holding the token and sharing summary information (lower scope).
The message consists of metadata/data pair(s) that contain service
information and a parameter indicating "size" of the data set the LS manages
as well as the minimal (without parameters) summary.</p>
-<p id="rfc.section.4.6.p.2">The response message should indicate success or
failure via the eventType, and will contain metadata/data pair(s). The
metadata should indicate who the service is, and its "size" for leader voting
purposes. The data section is message that can be used for logging.</p>
-<p id="rfc.section.4.6.p.3">When receiving the message, check your local
list and update it as needed for: </p>
-<dl class="empty">
-<dd>Do you know of this service? If so make sure the vote and other info is
ok.</dd>
-<dd>Update the summary info in your collection</dd>
-<dd>If you don"t know of them, add them!</dd>
-</dl>
-<h3 id="rfc.section.4.6.1">
-<a href="#rfc.section.4.6.1">4.6.1</a> <a
id="LSControl-Summary2Request" href="#LSControl-Summary2Request">Request</a>
-</h3>
-<p id="rfc.section.4.6.1.p.1">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.6"><a
href="#rfc.section.4.6">4.6</a> <a id="LSControl-Summary-lower"
href="#LSControl-Summary-lower">LS Summary Message (Lower)</a></h2><p
id="rfc.section.4.6.p.1">This message exchange represents when an LS instance
is holding the token and sharing summary information (lower scope). The
message consists of metadata/data pair(s) that contain service information
and a parameter indicating "size" of the data set the LS manages as well as
the minimal (without parameters) summary.</p><p id="rfc.section.4.6.p.2">The
response message should indicate success or failure via the eventType, and
will contain metadata/data pair(s). The metadata should indicate who the
service is, and its "size" for leader voting purposes. The data section is
message that can be used for logging.</p><p id="rfc.section.4.6.p.3">When
receiving the message, check your local list and update it as needed for:
</p><dl class="empty"><dd>Do you know of this service? I
f so make sure the vote and other info is ok.</dd><dd>Update the summary
info in your collection</dd><dd>If you don"t know of them, add
them!</dd></dl><h3 id="rfc.section.4.6.1"><a
href="#rfc.section.4.6.1">4.6.1</a> <a id="LSControl-Summary2Request"
href="#LSControl-Summary2Request">Request</a></h3><p
id="rfc.section.4.6.1.p.1"> <pre>

<nmwg:message type="LSControlRequest">

@@ -1099,13 +704,7 @@
</nmwg:message>

- </pre>
-</p>
-<h3 id="rfc.section.4.6.2">
-<a href="#rfc.section.4.6.2">4.6.2</a> <a
id="LSControl-Summary2Response"
href="#LSControl-Summary2Response">Response</a>
-</h3>
-<p id="rfc.section.4.6.2.p.1">
-<pre>
+ </pre> </p><h3 id="rfc.section.4.6.2"><a
href="#rfc.section.4.6.2">4.6.2</a> <a id="LSControl-Summary2Response"
href="#LSControl-Summary2Response">Response</a></h3><p
id="rfc.section.4.6.2.p.1"> <pre>

<nmwg:message type="LSControlResponse">

@@ -1131,24 +730,7 @@

</nmwg:message>

- </pre>
-</p>
-<h2 id="rfc.section.4.7">
-<a href="#rfc.section.4.7">4.7</a> <a id="LSControl-Summary-upper"
href="#LSControl-Summary-upper">LS Summary Message (Upper)</a>
-</h2>
-<p id="rfc.section.4.7.p.1">This message exchange represents when an LS
instance is holding the token and sharing summary information. The message
consists of metadata/data pair(s) that contain service information and a
parameter indicating "size" of the data set the LS manages. The "data"
portion is the summary info (FORMAT TBD!!!)</p>
-<p id="rfc.section.4.7.p.2">The response message should indicate success or
failure via the eventType, and will contain metadata/data pair(s). The
metadata should indicate who the service is, and its "size" for leader voting
purposes. The data section is message that can be used for logging.</p>
-<p id="rfc.section.4.7.p.3">When receiving the message, check your local
list and update it as needed for: </p>
-<dl class="empty">
-<dd>Do you know of this service? If so make sure the vote and other info is
ok.</dd>
-<dd>Update the summary info in your collection</dd>
-<dd>If you don't know of them, add them!</dd>
-</dl>
-<h3 id="rfc.section.4.7.1">
-<a href="#rfc.section.4.7.1">4.7.1</a> <a id="LSControl-SummaryRequest"
href="#LSControl-SummaryRequest">Request</a>
-</h3>
-<p id="rfc.section.4.7.1.p.1">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.7"><a
href="#rfc.section.4.7">4.7</a> <a id="LSControl-Summary-upper"
href="#LSControl-Summary-upper">LS Summary Message (Upper)</a></h2><p
id="rfc.section.4.7.p.1">This message exchange represents when an LS instance
is holding the token and sharing summary information. The message consists of
metadata/data pair(s) that contain service information and a parameter
indicating "size" of the data set the LS manages. The "data" portion is the
summary info (FORMAT TBD!!!)</p><p id="rfc.section.4.7.p.2">The response
message should indicate success or failure via the eventType, and will
contain metadata/data pair(s). The metadata should indicate who the service
is, and its "size" for leader voting purposes. The data section is message
that can be used for logging.</p><p id="rfc.section.4.7.p.3">When receiving
the message, check your local list and update it as needed for: </p><dl
class="empty"><dd>Do you know of this service? If so make s
ure the vote and other info is ok.</dd><dd>Update the summary info in your
collection</dd><dd>If you don't know of them, add them!</dd></dl><h3
id="rfc.section.4.7.1"><a href="#rfc.section.4.7.1">4.7.1</a> <a
id="LSControl-SummaryRequest"
href="#LSControl-SummaryRequest">Request</a></h3><p
id="rfc.section.4.7.1.p.1"> <pre>

<nmwg:message type="LSControlRequest">

@@ -1197,13 +779,7 @@

</nmwg:message>

- </pre>
-</p>
-<h3 id="rfc.section.4.7.2">
-<a href="#rfc.section.4.7.2">4.7.2</a> <a
id="LSControl-SummaryResponse" href="#LSControl-SummaryResponse">Response</a>
-</h3>
-<p id="rfc.section.4.7.2.p.1">
-<pre>
+ </pre> </p><h3 id="rfc.section.4.7.2"><a
href="#rfc.section.4.7.2">4.7.2</a> <a id="LSControl-SummaryResponse"
href="#LSControl-SummaryResponse">Response</a></h3><p
id="rfc.section.4.7.2.p.1"> <pre>

<nmwg:message type="LSControlResponse">

@@ -1229,18 +805,7 @@

</nmwg:message>

- </pre>
-</p>
-<h2 id="rfc.section.4.8">
-<a href="#rfc.section.4.8">4.8</a> <a id="LSControl-Leader"
href="#LSControl-Leader">LS Leader Message</a>
-</h2>
-<p id="rfc.section.4.8.p.1">This message exchange will be conducted between
the Leader and Vice-Leader on some (frequent) interval. It may even become a
part of the Leader's token exchange with the Upper Level.</p>
-<p id="rfc.section.4.8.p.2">The leader identifies itself, and sends down the
summaries from the upper level for the Vice-Leader to store. If the leader
should die, the vice leader will have a summary of the upper level and be
able to continue answering lower level queries and obtaining information from
the higher levels.</p>
-<h3 id="rfc.section.4.8.1">
-<a href="#rfc.section.4.8.1">4.8.1</a> <a id="LSControl-LeaderRequest"
href="#LSControl-LeaderRequest">Request</a>
-</h3>
-<p id="rfc.section.4.8.1.p.1">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.8"><a
href="#rfc.section.4.8">4.8</a> <a id="LSControl-Leader"
href="#LSControl-Leader">LS Leader Message</a></h2><p
id="rfc.section.4.8.p.1">This message exchange will be conducted between the
Leader and Vice-Leader on some (frequent) interval. It may even become a part
of the Leader's token exchange with the Upper Level.</p><p
id="rfc.section.4.8.p.2">The leader identifies itself, and sends down the
summaries from the upper level for the Vice-Leader to store. If the leader
should die, the vice leader will have a summary of the upper level and be
able to continue answering lower level queries and obtaining information from
the higher levels.</p><h3 id="rfc.section.4.8.1"><a
href="#rfc.section.4.8.1">4.8.1</a> <a id="LSControl-LeaderRequest"
href="#LSControl-LeaderRequest">Request</a></h3><p
id="rfc.section.4.8.1.p.1"> <pre>

<nmwg:message type="LSControlRequest">

@@ -1314,13 +879,7 @@

</nmwg:message>

- </pre>
-</p>
-<h3 id="rfc.section.4.8.2">
-<a href="#rfc.section.4.8.2">4.8.2</a> <a id="LSControl-LeaderResponse"
href="#LSControl-LeaderResponse">Response</a>
-</h3>
-<p id="rfc.section.4.8.2.p.1">
-<pre>
+ </pre> </p><h3 id="rfc.section.4.8.2"><a
href="#rfc.section.4.8.2">4.8.2</a> <a id="LSControl-LeaderResponse"
href="#LSControl-LeaderResponse">Response</a></h3><p
id="rfc.section.4.8.2.p.1"> <pre>

<nmwg:message type="LSControlRequest">

@@ -1328,17 +887,7 @@

</nmwg:message>

- </pre>
-</p>
-<h2 id="rfc.section.4.9">
-<a href="#rfc.section.4.9">4.9</a> <a id="LSControl-Discovery"
href="#LSControl-Discovery">LS Discovery Message</a>
-</h2>
-<p id="rfc.section.4.9.p.1">Structure of the LSDiscovery Message used to
locate info-sets. (FORMAT TBD!!!)</p>
-<h3 id="rfc.section.4.9.1">
-<a href="#rfc.section.4.9.1">4.9.1</a> <a id="LSDiscoveryRequest"
href="#LSDiscoveryRequest">Request</a>
-</h3>
-<p id="rfc.section.4.9.1.p.1">
-<pre>
+ </pre> </p><h2 id="rfc.section.4.9"><a
href="#rfc.section.4.9">4.9</a> <a id="LSControl-Discovery"
href="#LSControl-Discovery">LS Discovery Message</a></h2><p
id="rfc.section.4.9.p.1">Structure of the LSDiscovery Message used to locate
info-sets. (FORMAT TBD!!!)</p><h3 id="rfc.section.4.9.1"><a
href="#rfc.section.4.9.1">4.9.1</a> <a id="LSDiscoveryRequest"
href="#LSDiscoveryRequest">Request</a></h3><p id="rfc.section.4.9.1.p.1">
<pre>

<nmwg:message type="LSDiscoveryRequest">

@@ -1352,13 +901,7 @@

</nmwg:message>

- </pre>
-</p>
-<h3 id="rfc.section.4.9.2">
-<a href="#rfc.section.4.9.2">4.9.2</a> <a id="LSDiscoveryResponse"
href="#LSDiscoveryResponse">Response</a>
-</h3>
-<p id="rfc.section.4.9.2.p.1">
-<pre>
+ </pre> </p><h3 id="rfc.section.4.9.2"><a
href="#rfc.section.4.9.2">4.9.2</a> <a id="LSDiscoveryResponse"
href="#LSDiscoveryResponse">Response</a></h3><p id="rfc.section.4.9.2.p.1">
<pre>

<nmwg:message type="LSDiscoveryResponse">

@@ -1392,106 +935,16 @@

</nmwg:message>

- </pre>
-</p>
-<hr class="noprint">
-<h1 id="rfc.section.5" class="np">
-<a href="#rfc.section.5">5.</a> <a id="codes" href="#codes">Result
codes</a>
-</h1>
-<ul>
-<li>error.ls.foo -</li>
-<li>success.ls.foo -</li>
-<li>TBD</li>
-</ul>
-<hr class="noprint">
-<h1 id="rfc.section.6" class="np">
-<a href="#rfc.section.6">6.</a> <a id="apdx" href="#apdx">Appendices</a>
-</h1>
-<h2 id="rfc.section.6.1">
-<a href="#rfc.section.6.1">6.1</a> <a id="glossary"
href="#glossary">Glossary</a>
-</h2>
-<ul>
-<li>AuthoritativeLS - LS that is an authority for the perfSONAR services in
question. AuthoritativeLS is a result of discovery phase and can be used in
the metadata query phase.</li>
-<li>Berkeley DB XML - Oracle Berkeley DB XML is an open source, embeddable
XML database with XQuery-based access to documents stored in containers and
indexed based on their content.</li>
-<li>Bootstraping - It refers to the process of automatically finding at
service startup other LS instances of the scope utilizing a previously
configured registry.</li>
-<li>eXist XML DB - eXist is an Open Source native XML database featuring
efficient, index-based XQuery processing, automatic indexing, extensions for
full-text search, XUpdate support, XQuery update extensions and tight
integration with existing XML development tools.</li>
-<li>Home LS (HLS) - The Home LS of a Service is the LS where the Service
registers its Lookup Information</li>
-<li>Lookup Service (LS) - The Lookup Service is a key element of the
perfSONAR framework because it allows every independent service to be a
visible part of the system. New services may identify themselves to the
community and provide their detailed capabilities description. Other services
are able to communicate to the LS in order to get this data called Lookup
Information. Basic Lookup Service supports registration, query, keepalives
and de-registration actions (additionally updates?).</li>
-<li>Lookup Information - information registered by a Service in the Lookup
Service</li>
-<li>Lower Scope - The scoping paradigm meant to indicate inter-domain
relationships.</li>
-<li>LSRing - Represents the state of the LS cloud listing available LS
instances</li>
-<li>Multidomain / Distributed Lookup Information (mLS) - Lookup Service
which supports summarization and communication with other Lookup Services
(which might be in the same domain...)</li>
-<li>P2P - network infrastucture that does not have fixed clients and
servers, but a number of peer nodes that function as both clients and servers
to the other nodes on the network. This model is contrasted with the
client-server model.</li>
-<li>Service - A Service is an application that communicates with other
perfSONAR Services via standardized protocol set (SOAP+XML+NMWGv2)</li>
-<li>Summary Information - aggregated information from Lookup Information
that is sent by one LS to another</li>
-<li>Token Ring - A ring network in which the network topology features nodes
connected to exactly two other nodes, forming a circular pathway for signals:
a ring. Data travels from node to node, with each node handling every packet.
We use a logical ring in which a "token" message is used to synchronize the
communication among the nodes.</li>
-<li>Upper (Global) Scope - The scoping paradigm meant to indicate
intra-domain relationships.</li>
-<li>XSLT - Extensible Stylesheet Language Transformations is an XML-based
language used for the transformation of XML documents into other XML or
"human-readable" documents. The original document is not changed; rather, a
new document is created based on the content of an existing one.</li>
-<li>XQuery - A query language (with some programming language features) that
is designed to query collections of XML data. It is semantically similar to
SQL.</li>
-</ul>
-<h1 class="np" id="rfc.references">
-<a href="#rfc.section.7">7.</a> References</h1>
-<table summary="References" border="0" cellpadding="2">
-<tr>
-<td class="topnowrap">
-<b id="tridentcom">[1]</b>
-</td>
-<td class="top">Zurawski, J., Swany, M., and D. Gunter, “A Scalable
Framework for Representation and Exchange of Network Measurements”, In
Proceedings of 2nd International IEEE/Create-Net
+ </pre> </p><hr class="noprint"><h1 id="rfc.section.5" class="np"><a
href="#rfc.section.5">5.</a> <a id="codes" href="#codes">Result
codes</a></h1><ul><li>error.ls.foo -</li><li>success.ls.foo
-</li><li>TBD</li></ul><hr class="noprint"><h1 id="rfc.section.6"
class="np"><a href="#rfc.section.6">6.</a> <a id="apdx"
href="#apdx">Appendices</a></h1><h2 id="rfc.section.6.1"><a
href="#rfc.section.6.1">6.1</a> <a id="glossary"
href="#glossary">Glossary</a></h2><ul><li>AuthoritativeLS - LS that is an
authority for the perfSONAR services in question. AuthoritativeLS is a result
of discovery phase and can be used in the metadata query
phase.</li><li>Berkeley DB XML - Oracle Berkeley DB XML is an open source,
embeddable XML database with XQuery-based access to documents stored in
containers and indexed based on their content.</li><li>Bootstraping - It
refers to the process of automatically finding at service startup other LS
instances of the scope utilizing a previo
usly configured registry.</li><li>eXist XML DB - eXist is an Open Source
native XML database featuring efficient, index-based XQuery processing,
automatic indexing, extensions for full-text search, XUpdate support, XQuery
update extensions and tight integration with existing XML development
tools.</li><li>Home LS (HLS) - The Home LS of a Service is the LS where the
Service registers its Lookup Information</li><li>Lookup Service (LS) - The
Lookup Service is a key element of the perfSONAR framework because it allows
every independent service to be a visible part of the system. New services
may identify themselves to the community and provide their detailed
capabilities description. Other services are able to communicate to the LS in
order to get this data called Lookup Information. Basic Lookup Service
supports registration, query, keepalives and de-registration actions
(additionally updates?).</li><li>Lookup Information - information registered
by a Service in the Lookup Serv
ice</li><li>Lower Scope - The scoping paradigm meant to indi!
cate int
er-domain relationships.</li><li>LSRing - Represents the state of the LS
cloud listing available LS instances</li><li>Multidomain / Distributed Lookup
Information (mLS) - Lookup Service which supports summarization and
communication with other Lookup Services (which might be in the same
domain...)</li><li>P2P - network infrastucture that does not have fixed
clients and servers, but a number of peer nodes that function as both clients
and servers to the other nodes on the network. This model is contrasted with
the client-server model.</li><li>Service - A Service is an application that
communicates with other perfSONAR Services via standardized protocol set
(SOAP+XML+NMWGv2)</li><li>Summary Information - aggregated information from
Lookup Information that is sent by one LS to another</li><li>Token Ring - A
ring network in which the network topology features nodes connected to
exactly two other nodes, forming a circular pathway for signals: a ring. Data
travels from node to node
, with each node handling every packet. We use a logical ring in which a
"token" message is used to synchronize the communication among the
nodes.</li><li>Upper (Global) Scope - The scoping paradigm meant to indicate
intra-domain relationships.</li><li>XSLT - Extensible Stylesheet Language
Transformations is an XML-based language used for the transformation of XML
documents into other XML or "human-readable" documents. The original document
is not changed; rather, a new document is created based on the content of an
existing one.</li><li>XQuery - A query language (with some programming
language features) that is designed to query collections of XML data. It is
semantically similar to SQL.</li></ul><h1 class="np" id="rfc.references"><a
href="#rfc.section.7">7.</a> References</h1><table summary="References"
border="0" cellpadding="2"><tr><td class="topnowrap"><b
id="tridentcom">[1]</b></td><td class="top">Zurawski, J., Swany, M., and D.
Gunter, “A Scalable Framework for
Representation and Exchange of Network Measurements”, !
In Proce
edings of 2nd International IEEE/Create-Net
Conference on Testbeds and Research Infrastructures
for the Development of Networks and Communities
- (Tridentcom 2006).</td>
-</tr>
-</table>
-<hr class="noprint">
-<h1 id="rfc.authors" class="np">Authors' Addresses</h1>
-<address class="vcard">
-<span class="vcardline">
-<span class="fn">Jeff Boote</span>
-<span class="n hidden">
-<span class="family-name">Boote</span>
-<span class="given-name">Jeff</span>
-</span>
-</span>
-<span class="org vcardline">Internet2
+ (Tridentcom 2006).</td></tr></table><hr
class="noprint"><h1 id="rfc.authors" class="np">Authors'
Addresses</h1><address class="vcard"><span class="vcardline"><span
class="fn">Jeff Boote</span><span class="n hidden"><span
class="family-name">Boote</span><span
class="given-name">Jeff</span></span></span><span class="org
vcardline">Internet2
1000 Oakbrook Drive Suite 300
- Ann Arbor MI 48104 </span>
-</address>
-<address class="vcard">
-<span class="vcardline">
-<span class="fn">Maciej Glowiak</span>
-<span class="n hidden">
-<span class="family-name">Glowiak</span>
-<span class="given-name">Maciej</span>
-</span>
-</span>
-<span class="org vcardline">Poznan Supercomputing and Networking Center
+ Ann Arbor MI 48104
</span></address><address class="vcard"><span class="vcardline"><span
class="fn">Maciej Glowiak</span><span class="n hidden"><span
class="family-name">Glowiak</span><span
class="given-name">Maciej</span></span></span><span class="org
vcardline">Poznan Supercomputing and Networking Center
ul. Noskowskiego 10
61-704 Poznan
- Poland</span>
-</address>
-<address class="vcard">
-<span class="vcardline">
-<span class="fn">D. Martin Swany</span>
-<span class="n hidden">
-<span class="family-name">Swany</span>
-<span class="given-name">D. Martin</span>
-</span>
-</span>
-<span class="org vcardline">University of Delaware
+ Poland</span></address><address
class="vcard"><span class="vcardline"><span class="fn">D. Martin
Swany</span><span class="n hidden"><span
class="family-name">Swany</span><span class="given-name">D.
Martin</span></span></span><span class="org vcardline">University of Delaware
Department of Computer and
Information Sciences
- Newark, DE 19716</span>
-</address>
-<address class="vcard">
-<span class="vcardline">
-<span class="fn">Jason Zurawski</span>
-<span class="n hidden">
-<span class="family-name">Zurawski</span>
-<span class="given-name">Jason</span>
-</span>
-</span>
-<span class="org vcardline">Internet2
+ Newark, DE
19716</span></address><address class="vcard"><span class="vcardline"><span
class="fn">Jason Zurawski</span><span class="n hidden"><span
class="family-name">Zurawski</span><span
class="given-name">Jason</span></span></span><span class="org
vcardline">Internet2
1000 Oakbrook Drive Suite 300
- Ann Arbor MI 48104 </span>
-</address>
-</body>
-</html
+ Ann Arbor MI 48104
</span></address></body></html>
\ No newline at end of file

Modified: trunk/nmwg/doc/dLS/dLS.pdf
===================================================================
(Binary files differ)

Modified: trunk/nmwg/doc/dLS/dLS.xml
===================================================================
--- trunk/nmwg/doc/dLS/dLS.xml 2007-10-17 14:03:15 UTC (rev 292)
+++ trunk/nmwg/doc/dLS/dLS.xml 2007-10-29 03:51:36 UTC (rev 293)
@@ -47,11 +47,11 @@
particular pS service instance.
</t>
<t>
- From clients' perspective, Lookup Service operation involves
registration,
+ From the clients' perspective, Lookup Service operation involves
registration,
deregistration, querying and obtaining query results. Clients want
to discover
- the services that are running in the network. LS enables this by
gathering
+ the services that are running in the network. The LS enables this
by gathering
information from the services and then using it to fulfill client
queries.
- Figure below presents basic LS interactions.
+ The following figure presents basic LS interactions.
</t>
<figure anchor="ls-op">
<preamble>LS Operation</preamble>
@@ -70,28 +70,31 @@
</figure>
<t>
This document describes the support
- necessary to extend basic Lookup Service to a distributed mode of
operation.
- dLS reduces overhead in a single service and maintains consistent
information.
+ necessary to extend the basic Lookup Service to a distributed mode
of operation.
+ Distributing LS functionality is necessary to scale the system up in
+ terms of the amount of information that can be stored and searched.
There are a few key facets of this mode of operation:
- </t>
- <list style="symbols">
- <t>Summarization - to reduce the amount of information sent over the
+ </t>
+
+ <t><list style="symbols">
+ <t>Summarization - to reduce the amount of information sent over the
+
network or to anonymize sensitive data, some form of data
reduction
must take place.
</t>
<t>Scope - to enable a hierarchy of systems, some form of scoping
- must exist that forms logical communication and data exchange
- channels.
+ must exist that defines local and remote communication groups.
</t>
<t>Search - information location is key and the way in which
distributed location and search is handled is the crux of this
service.
</t>
- </list>
- <t>
- Additionally we present solutions to issues related to allow
seamless
- operation of this service including bootstrapping (i.e. how service
find
- other parts of the system) and domain specific concerns.
+ </list></t>
+
+ <t>
+ Additionally we present solutions to issues necessary to allow
effective
+ operation of this service including bootstrapping (i.e. how service
finds
+ other parts of the system) and domain-specific concerns.
</t>
</section>
<section anchor="system" title="System Specific Operation">
@@ -109,21 +112,31 @@
<xref target="lookup-info" />).
</t>
<t>
- The idea is to move the metadata from a local XML data store to a
+ The idea is to move the metadata from a service-local XML data
store to a
specialized LS with additional searching capabilities. While a
service instance may support limited searching, this is not
necessary as they should be focused on storing or gathering data
and leave the lookup functionality to the LS. Possible
exceptions
- are rapidly changing metadata like the most recent timestamp and
+ when a client may need to contact a service directly are when
metadata
+ rapidly changes, like the most recent data's timestamp and
full details of data stored in long-term archival MAs.
</t>
<t>The architecture of the dLS protocol assumes the existence of logical
- rings of LS instances. The current proposal involves two
levels of rings:
- lower scope and upper scope. Lower scope hides
inter-domain relationships
- between LS instances while upper scope hides intra-domain
relationships.
+ rings of LS instances. The architecture should allow for
multiple levels.
+ The current proposal involves two levels of rings:
+ lower scope and upper scope. Lower scope hides inter-domain
relationships
+ between LS instances while upper scope hides intra-domain
relationships.
+ 
+ 
</t>
- </section>
-
+
+ </section>
+
+ 
<section anchor="summary" title="Summarization">
<t>
The LS that a service contacts to register becomes the "Home LS"
@@ -133,36 +146,39 @@
enterprise and to draw relevant queries to itself.
</t>
<t>
- The construction of such a summary is important to the overall
- success of this service; summaries prevents other LS instances
- from being overloaded by information, must be general enough to
allow
+ Summarization is important to the overall
+ success of this service as summaries prevents other LS instances
+ from being overloaded by information. They must be general
enough to allow
for easy creation and exchange but also must retain enough
information to provide a rich query interface able to locate the
- distributed information. That means service metadata information
must be filtered
- (summarized) as it propagates through the LS cloud. We start by
making an
- observation that summarization is best based on scope (see also
- <xref target="scope" /> for forming scope), simply put this
- means that we should attempt to summarize the "most" the
"further" away from the
+ distributed information. That means service metadata information
+ must be reduced (summarized) as it propagates through the LS
cloud. We start by making an
+ observation that summarization is best based on scope (see also
+ <xref target="scope" /> for forming scope). Simply put, this
+ means that we should attempt to summarize "more" the "farther"
away from the
source that we get. This creates a smaller data set that travels
- the furthest away while keeping the larger and more informative
data
+ the farthest away while keeping the larger and more informative
data
sets closer to the source. We present the strategies as such:

<list style="symbols">
- <t>Summarization for the "lower scope" (formerly known as
+ 
+ <t>Summarization for the "lower scope" (formerly known as
"local scope")</t>
<t>Summarization for the "upper scope" (formerly known as
"global scope")</t>
</list>

- We limit the discussion in this case to these two scopes,
although
- extension to "n" scopes is possible. As the number of scopes
+ We limit much of the discussion in this document to two scopes,
although
+ an arbitrary number of scopes is possible without extension to the
+ model. As the number of scopes
increases additional "aggregation" will be necessary to combine
information thus reducing the size of the data sets further.
+ 
</t>
- <section anchor="lower_scope_summarization" title="Lower Scope
Summarization">
- <t>
+ <section anchor="lower_scope_summarization" title="Lower Scope
Summarization">
+ <t>
This first level of summarization, between HLS instances
internal
- to a domain, consists of simply dropping superfluous information
+ to a domain, consists of simply eliding detailed information
from the metadata descriptions provided by registered services.

For now we define this to be simply removing additional
"parameter"
elements from the metadata. Special consideration must be
given to
@@ -239,27 +255,30 @@
information, in the possession of non HLS instances, is
provided
as a convenience and should be treated in the same way that
directly registered information is (i.e. purged on
expiration).
- When requested, credit must also be attributed to the HLS as
being
- (non)authoritative for each piece of information.
+ When responding to queries for this information, the LS must
+ indicate whether or not it is authoritative.
</t>
</section>
<section anchor="upper_scope_summarization" title="Upper Scope
Summarization">
<t>
A designated member of each lower scope will be required to
interact
with the upper scope. The mechanics of how we learn who is the
- designated leader is discussed in <xref target="tokens" />. The
+ designated leader are discussed in <xref target="tokens" />.
The
leader of each lower scope (and the designated backup) will be
responsible
for examining each member's summary information and building a
- summarization/aggregation that best describes the contents of the
- ring.
+ summarization/aggregation that describes the contents of a
+ given scope. This summary will serve as input to the
higher-level
+ scope.
</t>
<t>
The most natural summarization is based on the topology of the
network (like in network routing). Thus, topology-based
- summarization will include this information as well as eventType
- information from the other LSs. Summarization will be performed
+ summarization will reduce available service instances in the same
+ way that IP addresses are summarized into network numbers. They will
+ indicate the eventTypes that a service has and ones that can it can
generate.
+ Summarization will be performed
using specialized summary algorithm. Topology information such
- as IP addresses will be summarized using algorithms basing on
+ as IP addresses will be summarized using algorithms based on
Radix Tree (see <xref target="IP-summary" />).
</t>
<t>
@@ -322,8 +341,7 @@
</t>
<t>
Once constructed, it is possible to consult the structure in
- creating IP summaries as well as constructing information
- regarding netmasks.
+ creating IP network summaries.
</t>
</section>
</section>
@@ -358,7 +376,7 @@
<t>Token Passing</t>
<t>Summarization Notification</t>
</list>
- Each of these procedures is paramount to keeping members of the
+ Each of these procedures is important to keeping members of the
distributed "service" functioning correctly. The algorithms will
be
presented in the frame of the lower scope, but will be used in the
same manner for the upper scope as well.
@@ -373,8 +391,8 @@
to start a ring that may not exist yet.
</t>
<t>
- A candidate LS will continuously search it's LSRing information
and
- send an LSControl mesage to its know LS instances with a "join"
eventType
+ A candidate LS will continuously search its LSRing information
and
+ send an LSControl mesage to its known LS instances with a "join"
eventType
(see <xref target="LSControl-Join" />) until a successful
response
is seen. The LS candidate will then search through
the successful LSControlResponse to this message and update its
LSRing with
@@ -393,7 +411,7 @@
<xref target="LSControl-Summary-upper" />) to all of the
"active"
LS instances from its LSRing. Again the responses will be
parsed
to get any useful updated information. At the end of this
process
- joining LS will possess an LSRing file reflecting the state of
the
+ the joining LS will possess an LSRing file reflecting the state
of the
dLS cloud. Each of the recipient LS instances which hasn't
heard
anything from this joining LS previously will do the same,
including adding this new member to their own lists (as they
@@ -430,8 +448,10 @@
<postamble>LS2, LS3 and LS4 are in the ring. LS1 wants to join
the dLS cloud.</postamble>
</figure>
+ </t>
<t>The algorithm for joining the ring works as follows:
- </t>
+ </t>
+ <t>
<list type="symbols">
<t>1. LS1 - Candidate sends LSControlRequest message with
the http://perfsonar.net/services/LS/join eventType to
@@ -439,7 +459,7 @@
LS2 was found in LSRing initial file during
bootstrapping
process.</t>
<t>2. LS2 receives join message from L1 and decides whether
- to accept it or not. A security policy may occur
here.</t>
+ to accept it or not. Application of security policy
may occur here.</t>
<t>3. LS2 sends the LSControlResponse answer indicating success
joining (result code:
http://perfsonar.net/result/success/LS/join)
or failure (error code).</t>
@@ -486,16 +506,47 @@
<t>
The "token" is an LSControlMessage (see <xref
target="LSControl-Token" />)
meant to be passed around an LSRing to the various members in
some
- order. There are various criterion that can be used in deciding
how
+ order. There are various criteria that can be used in deciding
how
to order the ring so that everyone can predict where the token
is,
when they might expect to get it, and whom they should get it
from/
pass it to next. It is important that we choose a sound method
that
is simple to calculate, and should use as much "knowledge" of
the
ring as possible without burdening the LS instances too much with
complex calculations.
- </t>
+ </t>
+
+ <t>
+ The essential idea in the token passing mechanism for leader election
is
+ that some identifier is chosen for each node and that the node with the
+ highest (or lowest) identifier win the election and becomes the leader.
+ The basic mechanism of leader election is that participants form
+ a logical ring and initiate an election. An election can be
+ initiated when a new machine joins, at system start time, or when
+ a host feels that the leader may have failed based on failure to
+ receive a periodic token. When an election is initiated, the
+ initiating host sends an election message to its
+ counter-clockwise neighbor and changes its state to “ELECTING”.
+ It places its identifier inside the message. The ulimate goal is
+ for the host with the highest identifier to be chosen. When a
+ host receives an election message, it compares its identifier
+ with that in the message. It forwards the higher of the
+ identifiers. When a node receives a message with its own
+ identifier, it knows that it has been selected and the election
+ terminates.
+ </t>
+
+ <t>
+ The next question is how to choose the identifier for a given node.
+ There still needs to be some discussion here. The first proposal was
to
+ use the IP address of the node as the lower-order 32-bits of a 64-bit
+ number and to allow the higher-order bits to be set as a "priority"
+ field. This would effectively allow a system administrator to make
sure
+ that her most powerful or well-connected nodes became the leader when
+ they were available. In the absence of a priority, the nodes
essentially
+ are randomly ordered.
+ </t>
<t>
- A common method used in P2P systems such as Gnutella when
forming
+ Another possibility is to use common method used in P2P systems
such as Gnutella when forming
"ultrapeers" is to consider the size of the data that a node is
serving. The principle, as described
<eref
target="http://rakjar.de/gnufu/index.php/GnuFU_en#Network_model:_Change_who_calls_whom:_Ultrapeers_and_Leafs";>here</eref>,

@@ -511,22 +562,32 @@
Token passing is directly related to the concept of leader
election (see <xref target="Leader_Election" />), so more
explanation of this approach will follow. For now we are
justified
- in saying that the "contentSize" forms a good criterion for
+ in saying that the "contentSize" forms another good criterion
for
"ordering" the members of the ring. With all members of the ring
aware of everyone's data size, we can easily know who we should
pass the token to, and receive it from at any point in time.
We assume passing order between the LSs from smallest to
greatest
- "contentSize" (with wrap around at the ends).
- </t>
+ "contentSize" (with wrap around at the ends). Of course, this
+ metric leaves out the relative power of the machines in question, so
+ futher modifications are possible.
+ </t>
+
<t>
+ The key is that as long as the identifier is chosen consistently
within a
+ given scope, the choice of identifier doesn't affect the operation of
the
+ protocol.
+ </t>
+ <t>
The token can be viewed as "permission to talk" and permits the
- holding LS to send it's summary information to all other
available LS
+ holding LS to send its summary information to all other
available LS
instances (see <xref target="LSControl-Summary-lower" /> and
<xref target="LSControl-Summary-upper" />). The responses
will be parsed to get any useful updated information about
current
dLS cloud state.
</t>
- <t>
+
+ 
<section anchor="token_passing_algorithm" title="Token Passing
Algorithm">

+ </t>
+ 
<t>
The Leader and Vice-Leader LS instances should exchange messages
(see <xref target="LSControl-Leader" />)
@@ -696,9 +778,16 @@
local queries yet chosen anyway based on separate criteria) is very
appealing.
<vspace blankLines="1" />END_NOTE:<vspace blankLines="2" />
</t>
+
+ I don't think that we can debate this in comments. I agree with
+ Maciej. There's no way this should be the only option. Again,
+ we describe the generic protocol based on an ID, and discuss, in
+ one place in the doc, how to choose that ID.
-->

</section>
+
+ 
</section>

<section anchor="search" title="Search">
@@ -883,7 +973,8 @@
with some static values, and will be added to/deleted from as
time
goes on. As such implementations must ensure that this file
is under
database control of some sort.
- </t>
+ </t>
+ 
<section anchor="LSRingLower" title="LS Ring lower level">
<t>
<artwork>

nmwg: r293 - trunk/nmwg/doc/dLS, svnlog, 10/28/2007

List archive

nmwg: r293 - trunk/nmwg/doc/dLS