perfsonar development work

[Maciej Glowiak <>]

==========================================================================

 Minutes from Utrecht Meeting Wednesday, 28th of March 2007

==========================================================================

Initial, common session only.

Agenda:

 10:00 - 10:20: Introduction
 10:20 - 10:30: Passive - Sven Ubik
 10:30 - 11:00: Year 4 objectives - Nicolas Simar
 11:00 - 11:15: perfSONAR, JRA1, MDM and pilot.
 11:15 - 11:30: Hades Alarm System.


==========================================================================

1.Introduction by Nicolas Simar (NS)

First NS told about roles of the team and people involved to the
project. He presented new meeting format: less status update, more
discussion in smaller groups (6-8 people), people not interested in the
session should do something valuable for project. Then NS presented
detailed agenda and time schedule for the meeting and suggested who
should attend what session

NS explained that the software is Prototype but once released the
development become operational. Differences between prototype and
operational and relationship between MDM Pilot, MDM Prototype were
described. NS said also that development would be used as operational
and developers mustn't forget about it.

NS mentioned that there would be new editor for JRA1 deliverables and
documents, Ian Thomson.AL asked whether he was engeneer, but NS answered
he was not.

Mario asked who rewiew the deliverables. NS explained that first the 
document was reviewing by JRA1 then by Tomas Kalin and finally by the 
Executive.

2. Plan for Y4+6M, Nicolas Simar (NS)

Then NS was talking about plans for year 4. First he mentioned JRA1
sub-activities and tools and then said what was done (or was started)
during year 3 (pS 2.0, new developments: TopS,mLS,AA,flow base RRD MA,
TCMP, ABW was mentioned here). He explained also what was the aims for
the end of the Y3. Then he presented objectives for Y4: to provide
services through visualization and alarms, getting services operational,
to support other GN2 activities, to rationalize tools). Next he continue
talking what should be achieved during the year 4 (for instance using
AA, mLS, hello, resource protection, scheduler, push, chaining; easy
installation, performance, alarms, API for other tools). NS mentioned
also ideas for visualization such as single point of entry, improvements
of Look & Feel (it should be more uniform for all UI tools). Finally NS
told about resources for Y4+6M. Total number for that period of time is
203 PM. At the end he mentioned risks and opportunities of the project -
AA, cNIS, feedback from operational peple, decision process to define
what tools and services should be part of MDM - and presented main
questions to JRA1.
    Sven Ubik (SU) asked about rationalizing the developments, and NS 
explained
that we had too many visualization tools and should rationalize the
developments what means that not all of them were necessary and probably
some of them could be combined.
   Jerome Durand (JD) was curious who asked questions to JRA1 which was 
presented by
NS, and NS said that they were mostly given by Executive people and
other activities leaders.

3. Passive monitoring, Sven Ubik (SU)

First SU presented short summary of deployments in JRA1 and SA3. Then he
talked about ABW, the tool for passive capacity usage monitoring
distirubuted with DiMAPI. Next he presented the motivation for SA3
passive monitoring deployment and what information would be provided
(capacity, packet loss, connection health, protocol and application
distribution).
NS asked SU to send a summary of current deployments but after a short
discussion they decided to continue this topic later.
Then SU mentioned partners which would be deploying passive monitoring
software such as ABW, Tbwtools, Pktloss, Extended IPFIX. He also
introduced the problem of packet loss. Andreas Hanemann (AH) asked how 
to combine active and passive monitoring of packet loss. SU explained 
that they were, in fact, not the same metric, because active monitoring 
were sending test packets, different than in real internet and that it 
was very difficult to emulate real traffic. NS was wondering if passive 
monitoring would be done for specific case (to diagnose specific 
problem) or in general. SU said that in the future there should be one 
passive monitoring station per (? - not sure about it ?). Athanassios 
Liakopoulos (AL) asked about the amount of data for 10GBit interface. SU 
explained that it depended on traffic but above 300mbit/s hardware for 
passive monitoring is required. The discussion would be continued later on.

4. Hades alert system, Stephan Kraft (SK)

First SK briefly presented available metrics (OWD, OWL, traceroute
output, MP healt /ping, remote dist, remote ntp/). NS asked whether SK
was describing what they had that moment. SK confirmed and explained
that they were using Nagios standard plug-ins as an implementation, but
in future they wanted to do own development which would be more flexible
and would provide more informations and metrics. Then SK presented
working deployment of Nagios system which was accessible by web
interface. SK presented also alert criteria and suitable thresholds
needed. He mentioned that the objective is to rationalize notifications
(not to send all unnecessary) information. NS asked whether SK planned
to collaborate with administrators and other people involved to
deploying tools and SK was open for any feedback.
AH noticed that it'd be good if CNM showed highlights of problems,
visualized notifications (for instance as red or green colors of links)
and SK explained that the presented product was developed week before.

5.TCMP UI (tracefile capture MP User Interface). Uros Juvan (UJ)

UJ presented TCMP-UI tool, which was web based application. He showed
the example usage of the tool. While UJ was downloading huge result
files LK presented pS 2.0 release update

6. ps 20 release update, Loukik Kudarimoti (LK)

LK said version 2.0 RC1 is released. The documentation will be improved
over the next couple of weeks. The product contained 6 applications: rrd
ma, sql ma, ls, ssh/telnet mp, clmp, bwctl mp, all bundled together.
Five of them were in Java and one in Perl.
LK presented key features of the release such as new installation
concept, light weight bundle, multiple services can be installed on the
same tomcat, new installer, result codes, echo functionality, LS
registration, automatic downloading software. He mentioned the
requirements of the software - java, ant, exist, tomcat, perl, etc. Then
presented quick  review of the release management process that took 21
weeks including holidays. Handover process took 9 weeks. 19 people were
involved to the process (including developers, testers, release team).
Apart from it LK presented a lot of other statistics of the process. He
also mentioned about benefits of release process: much experience, new
documentation and many improvements. LK also thanked all people involved
to the process and asked people on the meeting to help the release team
by testing RC because the release process would need their feedback to
be finished.
AH asked how many release candidates would be there and LK said that
about three would be expected. AL was wondering how much time would it
take to install the whole package and LK explained that, for instance,
LS could take 15 min, but installing for first time might take even 1
hour (with reading documentation, following all steps in the
installation guide). NS was wondering how many bugs the release team
encountered. There were about a hundred in total software and 
documentation bugs.
Then there were the discussion on where to give feedback. Problems and
comments could be requested by bugzilla.
There were also a question about target Operating system. LK said that
we didn't limit perfSONAR to use only one system, because we couldn't
have forced administrators to install selected OS. Instead perfSONAR 2.0
was tested on various machines with various operating system. Basically
perfSONAR should work on any machine working with Java, but as UJ
noticed there might be other dependencies such as native libraries (such
as RRD tool) which may not work on some OS-es (such as Windows)

5b. At the end UJ finished his presentation on TCMP-UI and showed the
results
NS asked what tool could be used for viewing the results. UJ said that
any tool able to read pcap data, for instance: tcpdump, wireshark, ethereal.

==========================================================================

--

 --------------------------------------------------------------------
| Maciej Glowiak        Network Research and Development            ||
| 

     Poznan Supercomputing and Networking Center ||
| (+48 61) 858 2024          -- skype_id: maciej_psnc   GG: 4526858 ||
 ====================================================================