CERN-BARC video conference, 14/2/2007

Agenda:

Action List:

New/ongoing:

ID Status Due Date Who Description
1 Ongoing 21/2/07 Miro/Dennis/
Murthy/Salim		 Sit together and analyze the XML API client software, come up with list of items to correct OR if XML API needs to be rewritten SWIG interface now proposed to be at the C++ API level, as SWIG C++ problems resolved. API backwards compatibility not important as only Veronique uses API as external user. Salim to circulate Perl examples
6 Ongoing 21/2/07 Salim More precise performance measurements for XML API client. Miro requests even more details
7 NEW 21/2/07 Murthy/Sharma Prepare a report (and presentation? Check with Juergen) on the last quarter for the next CERN-BARC coordination videoconference, based on the weekly reports available at http://indico.cern.ch/categoryDisplay.py?categId=1346
8 NEW 21/2/07 Miro/Salim/
Dennis		 Define what C++ API versions need to be swigified in the first production deployment
9 NEW 21/2/07 Salim Implement comma-separated host list and cluster support in client API (cluster support will be done by Miro on server side)
10 NEW 21/2/07 Sharma Circulate code for foreign profile handling in CCM
11 NEW 21/2/07 Dinesh/Bill Address SSH tunneling problem

 

Closed:

4 Closed 7/2/07 Miro Warn Lemon users (particularly GridICE) about upcoming Lemon protocol changes for TCP
2 Closed 14/2/07 Salim/Miro Put XML API client/server code into CVS, maintain it from there.
5 Closed 14/2/07 Miro/Dennis send C code for local cache access to Salim.
3 Closed 14/2/07 Miro XML API server: Verify if there are any a) PHP configuration problems b) possibilities for OCI-level limitation enforcement c) possibilities of caching intermediate query results in a temporary Oracle table

Reports:

Murthy:

1. Lemon cache/transport re-engineering
Some code changes were required for new cache design to work properly. They are done and test is re-run. In the coming days test for transport design will be run to see the functionality

2. XML API
The main problem of contention of designing swig interface for factory type functions that return malloced structures in function arguments is resolved. The swig interface is being tested for memory leaks etc.

Review of xml api
As the main swig problem is solved (assuming other C++ subtilities of the api are easy to have swig interface), there is no need to have a C wrapper over the C++ api. We can start generating the swig perl interface for the C++ api and test it.
With Dennis observations and Salim's subsequent comments, a plan of action has to be arrived at and distribution of work has to be done.

 

Salim:

- I've put XML API Client code in CVS (elfms/lemon/xml_api/) after getting write access to repository.

- The biggest problem in Swig'fying the XML API was the typemaps for "structure returned by refernce", this has been solved and now I am, able to generate interface in Perl for functions involving such reference arguments. Also I've tested Perl interface to linked lists, Its working fine.

- For local cache access Dennis suggested newer implementation. He insists that there should be a clear separation between old SOAP based API and current XML API implementations of the same.

- First lag of performance measurements for XML API is done and sent to Miro.

Hi Miro,

I carried out tests to determine CPU and memory consumption of the xml API. Following are the results:-

Time(Sec) Size (MB) CPU(%) Memory(%)

1 2.58 4.5 0.1

2 5.14 18.2 0.9

2 7.54 28.7 1.0

2 10.02 40.5 1.2

3 12.48 48.7 1.3

These statistics involves time, cpu, and memory consumption while parsing the XML data in the library.(I've used get_all_metrics() method with different time intervals and fixed set of metric ids)... This does not include time to get XML data using http fetch, and time taken to display and or use parsed data.

This is just a sample, we can extend the reading to go to extreme end....

Sharma:

- added the support for comma separated list of clusters in wassh-ssm.
- there is no major change in the status of CCM developments as I was on leave. presently testing the fetch api and new ccm-fetch with capability to incorporate foreign profiles. First version of these will be in CVS by this week.
 

Dinesh:

The testing of new features linking of objects & power view is completed with dummy data.

I'm using the xml data (power and linking objects) as available currently. Once I get restructured xml data I will change the xml parsing code to improve it further.

The remote desktop connection is not working from BARC so I'm using the local cache feature of CCTracker to read the xml input data. For testing the updates initiated thro' CCTracker I will try ssh tunnel.
 

AOB: