Minutes for FTB conference call - 2010 May 26th
- Argonne National Lab: Rinku
- Oak Ridge National Lab: David, Aniruddha, Scott
- Ohio State University: Sonya
- Indiana University: Josh
- University of Tennessee: Aurelien
- Lawrence Berkeley National Lab: Absent
- FTB 0.6.2a1 has been released. There is one pending bug (Trac ticket #92). We discussed possible solutions to this approach. Scott suggested an approach consisting of "components sending periodic heartbeat to agents" or "component connecting and disconnecting periodically from the agent". Aurelien suggested having a component connect to multiple agent and moving to a different topology like a mesh network. The solution to this needs to be worked on..
- ANL wanted feedback on policy management framework. Details can be found at: http://wiki.mcs.anl.gov/cifts/index.php/Policy_mgmt .
- Rinku to upload the publications/presentations on the web page, wiki and svn. Everyone to send their publications/presentations" to Rinku this week. Please include the date and the event where the presentation was given.
- Carry forward from last week - Possible publication with Cobalt and Application scenario. This was not discussed today but keep this in place for next conference call.
- Decision to be made on what posters we plan to submit to upcoming conferences like SC'10..Rinku to initiate email conversation.
- Thomas sent an updated on tests scripts prior to conference call. These test scripts will be finalized this week and will be used for the alpha+ testing
- ORNL to test code on CRAY
- Aurlien is testing the FTB 0.6.2a1 code
- Carry forward from last week - Aurelien to look at the Cobalt wiki (http://wiki.mcs.anl.gov/cifts/index.php/Cobalt) and send list of FTB events for Cobalt that his library might need.
- OSU to test FTB 0.6.2a1 code with FTB-IB, FTB MVAPICH and give a heads-up
- MPICH and OpenMPI folks need to look at the events sent by OSU for MVAPICH Migration and work together to standardize them
- Carry forward from last week - Abhishek sent a document outlining the standard MPI events and the standardized payloads. MPICH2 folks have responded to the document and Abhishek needs to send the updated document.MVAPICH folks need to provide feedback.
- Carry forward from last week - Abhishek/Josh proposed an idea of having FTB itself serialize/de-serialize the event payload. i.e the user calls an FTB API routine, which will figure out the right structure and return the information to the user. This is in constrast to the current FTB where the user itself has to figure out what the FTB event payload is. We discussed the pros and cons of this approach. Josh/Abhishek need to start an email thread on this topic to get views from the broader cifts team.