January 14, 2010

Performance Meeting - 2010/01/14
 
Phone Number: (812) 856-3600
PIN:          002690#
 
Agenda:
  • Feedback on schedule
Attendees:  Shava (UCSD), Robert (IU), Raphael (ISI), Warren (TACC), Mauricio (UFL), Thomas (TU-D)

Notes:      
  • Review from last week
  • Robert updated benchmark section
  • Shava updated instrumentation section to be more generic based on feedback in last week's SW call - added selection of logging/timing library to schedule 
  • Number of databases?  All data in one?  This would be for FutureGrid infrastructure components only -- experiment data would be elsewhere.  Will determine number of databases once the FutureGrid architecture is more well defined and we can evaluate each component to see how much data would need to be collected and at what frequency.
  • Database schema?  Will budget time to look into that once we know more about the data we are collecting
  • Moab/XCAT logging - Shava asked Joe and Jenett -- Jenett replied they have not yet integrated Moab/XCAT and provided links and cc'ed Adaptive Computing guy -- waiting for more info; Warren suggested taking a look at Moab log on BigRed and to ask whether that might be a good comparison
  • Warren asked about use cases - how long does it take to instantiate an experiment on FG?
  • Performance tools - last week talked about doing a test install to make sure the tools can be installed well together; responsible for packaging but not installs
  • New iDataplex arrived at IU - Shava will email Dave about availability to get a node for testing
  • How will performance tools get integrated into images - asked on the sw call but Kate nor Renato on call
  • Usage information - talked last week about logging usage of performance tools
  • How do we do that when tools are APIs?  Could log when a performance tool is selected when image built.  Could also collect the information in annual user survey and collect suggestions for additional tools as well
Project Schedule:
  • We need to plan out for the year, project schedule attached lists out schedules for all components of our architecture with approximations for a first version of each component and then maintenance
  • Shava will sketch out a first draft of timing based on estimated dependencies and people's time commitment
  • Warren raised concern about match to PEP - Shava noted Inca, PAPI, and Vampir were the only components mentioned in the PEP and that the PEP didn't go into much detail about how performance would be collected and performance problems detected -- Shava will do another read thru and verify
  • Shava was using FastTrack (a Mac product) to create schedule - exported to XML - Robert can you verify you can import into project?
  • Robert suggested Open Project as free