perfSONAR User Q&A and Other Discussion

[shiraz memon <>]

Dear Jason;

Many thanks for your reply, it is indeed very insightful in getting closer to the decision of PerfSONAR's LS endorsement. However, few comments from my side are inline below:

On Mon, Nov 15, 2010 at 3:06 PM, Jason Zurawski <> wrote:

Hi Shiraz;

The perfSONAR LS is a very general framework when it comes to accepting information registration - any sort of 'service' that you can represent in XML formats can be be registered using some of the schemata we have defined for use in the project.  We normally break information into 2 categories:

 - Metadata e.g. information about the service such as how to contact it
 - Data e.g. capabilities of the service such as the form of data it controls

From the desired perspective, both type of info. will be published/discovered.

  

As an example we use the LS to register traditional perfSONAR services (e.g. Measurement Points, Measurement Archives) but have also added the ability to insert other entities that are not directly related to perfSONAR (e.g. GridFTP servers, Pheobus Gateways, REDDnet depots). This is still fledgling functionality and we do not have many more uses cases other than 'finding' these types of services.  Parsing over registered information is possible, just not fully explored.  Extending to some of the ideas you list below may require you to add extensions to the LS.

I would encourage you to look at the 'LSRegistrationDaemon' package (http://anonsvn.internet2.edu/svn/perfSONAR-PS/trunk/perfSONAR_PS-LSRegistrationDaemon/) as a first step to see the process we have taken to implement registration functionality.  It is written in perl and acts as a broker between the LS and other forms of services (usually by doing simple TCP connects to check availability of a target service).

 

Sorry for my dearth of knowledge on Perl, however, Isn't it written entirely on Java (according to the mdm admin guide)?

  

The only real complexity that I can see currently with your ideas is that while it is very simple to get the XML information into your hLS, the gLS would need to know a little bit more about the data types you would be registering for summarization purposes.  Currently this is something that the perfSONAR developers would need to be aware of to adjust the algorithms, but we are working on ways to completely automate this step.

 

Are the developers already working on the extensible summarisation process/plug-ability? 

Finally, which license is being endorsed to distribute the PerfSONAR's LS (or the other components)?

Hope this helps, thanks;

-jason

Cheers,

Shiraz

 

On 11/15/10 3:25 AM, shiraz memon wrote:

Hi Folks,

I am investigating LS currently and trying to find the potential
feasibility whether it could be used to discover the Grid Services such
as Job Management, Storage Management, Computing etc... on "a large
scale" deployment. The aforementioned services are exposed as Web or
WS-RF compliant services, thus metadata defined would be in XML format.

Here are some of the high level factors, that I would use to asses the
viability of LS within Grids
1. enable the service providers (any, but some are mentioned above) to
specify the metadata easily, that can be pulled/pushed from/into LS
      ** How easy it would be to extend the LS, so as to support the
info. publishing services, which are not a part of PerfSONAR's network?
2. acquaintance of new services to LS
      ** Can the LS simply allows the foreign services to publish their
metadata
3. querying the stored information at hLS or gLS using standardised
language, such as XPath or XQuery
      ** I think this is there, but how would the LS handles the large
amount of query results?
      ** Does it provide the iterating capabilities, if the large set
of xml documents are returned from an XMLDB?
4. to publish and retrieve the information in GLUE2 format
      ** INFO: GLUE2 is a standard from OGF, the model encapsulates the
Grid entities. the Glue-WG also have released the renderings in SQL, XML
and LDAP format.
      ** How challenging and realistic is to add support for GLUE2 model?
5. Is there any integration or extension available that can be endorsed
out-of-the box?

Thanks in advance,
Shiraz