Minutes of the MPI working group

Thursday September 14, 2000 at 11:00AM - 12:00AM PDT

Next Meeting: Thursday September 21, 2000 at 11AM (CIT will call 414-229-6508).

Outline changes to wrapperAPI design:

The meeting started by outlining the changes that had been discussed at the meeting on Saturday, September 9, 2000. The following decisions were made:

• Since the searches are no longer indexed by the wrapperAPI, the output structure will need some modification. It was agreed that Kent and Patrick would coordinate on this and update the structure. For immediate progress, however, it was agreed that the notfinished flag on the slaves would be part of the output structure.
• If the ouput structure returned by applyFilters() is null, then result information will not be communicated to the master process. Otherwise, it will. This means that two models of reporting results are available: (i) the search master collects all results from the search slaves and reports it, or (ii) all slave nodes report information to the master process.
• Fraction of search complete will be reported by search master only.
• If load balancing is not being performed, then each search is repsonsible for cleanly finishing its work in the alloted time. This decision can now be made internally to the search codes.
• A mechanism must be put in place to communicate to the search code when the LBComm has changed, this could be a flag passed to the applyFilters function. The search code will then be reponsible to diseminate this information to all nodes running the code so that they can return control to the wrapper and receive the new LBComm.
• The lowest ranked node in the LBComm should always be preserved since it may be acting as the search master. This will be implemented in the wrapperAPI code and should be indicated in both the wrapperAPI baseline requirements document and an appendix to the LAL specification which will present the requirements for search code that will run in the LDAS environment.

A work plan

• Duncan and Masha: work together to redesign the wrapperAPI and the dso function calling format (by 21 September 2000).
• Jolien: change LALWrapperInterface to comply with the new design. This will take about 2-3 days once the new design has been figured out.
• Kent and Patrick: figure out the changes required to the output structure due to removing the indexing (by 21 September).

Any other business

• Add links to LAL, LDAS and wrapper dso's CVS archives to the web page.
• Important deadlines in the development of the MPI components of LDAS:
  1. 10 October 2000: New design including appendix to LAL specification to be presented to ASIS group at their next teleconference.
  2. 17 December 2000?: Possible "workshop" on implementing searh codes as dso's within the wrapperAPI paradigm.
  3. 15-22 January 2001: MDC