Minutes of the MPI working group

Thursday October 26, 2000 at 11:00AM - 12:00AM PDT

Attending: Warren Anderson, Masha Barnes, Kent Blackburn, Patrick Brady, Duncan Brown, Jolien Creighton, Phil Ehrens, Alan Wiseman.

Next Meeting: Thursday November 2, 2000 at 11AM

Review progress with inspiral dso
Duncan Brown reported that the init() and finalize() and conditionData() functions of the inspiral dso are completed and have been debugged with the wrapperAPI. applySearch() is under development; a normalization is being tracked down and corrected. The LAL communications packages needed to be modified in order to work using the communicator; that is done. It was also decided that, for the purposes of the MDC, the wrapperAPI will be used to distribute the data.

A telecon will be arranged by Jolien to discuss the output from the inspiral search codes. Kent reminded us that the database can only handle 100 events per second from all searches running in the LDAS environment. Each event is about 100 bytes.

Review of progress with wrapperAPI
Masha reported that the wrapperAPI is ready. She has tested data distribution with the trivial dso and is continuing to test with the inspiral dso. While she is waiting for the inspiral dso, she is writing documentation for the wrapperAPI. Communications with the mpiAPI have been tested, but the protocol has not.

Status of the mpiAPI
Phil Ehrens reported that no work has been done on the mpiAPI since he has been supporting the immenent release of LDAS. The remaining issue is creation of log files for the wrapperAPI. This is normally handled by the managerAPI, but since the wrapperAPI does not follow the standard model it must be worked out in mpiAPI. About 75% of code is complete still.

Any other business
It was generally agreed that we are on schedule for integration of the mpiAPI, wrapperAPI and inspiral dso by November 15. This is on schedule.

The LIGO Beowulf had a 50% failure rate associated with poor memory. They are working with the vendor to replace the memory with higher grade components. Only 8 nodes are working

List of errors that can be produced by codes has been sent by Masha and Jolien. Phil is to produce a similar list soon. Duncan will produce a similar list for the shared object as soon as he completes the applySearch() function.

It was decided that we need to gain some experience with LAM and encourage the software committee to make a decision about its use for search code development. This will impact preparations for the MDC in the mpiAPI. As a result, it was decided that a software committee meeting will be scheduled for Thursday November 2, 2000 to discuss this issue further. Warren commented that LAM was very poorly supported for alpha linux when the UWM beowulf was built 2 years ago. The most serious issue raised by people was support of the product in the long term.


Patrick Brady