Next Meeting: 8:00AM PDT, Thursday April 6th, 2000 (note the time change).
Hierarchical Report:
UWM posted an analysis of using the wrapperAPI within the context of a Hierarchical Report. This can be found at the URL: http://www.lsc-group.phys.uwm.edu/~patrick/work/projects/mpiGroup/local_notes/hierarchical.html
UWM went over the three bullets of the report. In summary the report finds that if sufficient amounts of data can be sent to the templateFilters() function (on order 5X (15 min segments) then the wrapperAPI would only add a 4% overhead to analysis time. This was based on code tests performed by Jolien.
There is one remaining issue associated with "when to perform" the dynamic load balancing. UWM would like the node load to stay constant for several cycles of the templateFilters() call. Kent proposed adding an addition loop to the wrapperAPI around the templateFilters() call but inside the load balancing loop which could be controlled from a command line option that specified the number of inner loop cycles which do not do load balancing before dropping back to the outer loop that does do load balancing. Patrick asked that UWM be given a couple of days to review this option before adding it to the requirements.
Kent inquired if there was enough effort on the hierarchical effort to make it the target analysis for this project. UWM explained that the analysis didn't carry far enough to warrant that big a change and we should still stick to the flat search previously outlined. However, UWM did express interest in starting a second project after completion of this project in order to develop the hierarchical search under the wrapperAPI. Warren added that we would have a two week head start on that project after this past two weeks' analysis :).
Changes to the Requirements:
Kent outline the changes to the wrapperAPI requirements which are available from the webpage. The changes made were to support MPI communications in the templateFilters() and the conditionData() routines. It was explained that the wrapperAPI would construct the needed MPI communication structures appropriate for the MPI communications within the dynamic library and pass this information into the functions instead of their local node number and max nodes available.
Schedule Update:
In summary of the effort that went into the analysis of using the wrapperAPI in a hierarchical search, it was accessed that the completion of a dynamically loaded library for the project would be delayed by approximately two weeks and the development of the wrapperAPI from CITs side by about one week due to the need to develop a release of LDAS to support the Engineering Run at Hanford.
Organizational Details:
Kent will be at Hanford next Thursday so the meeting time will change as reported above. Kent will initiate the call from Hanford. If that is not possible then Masha will initiate the call from CIT.
Minutes taken by Kent
Agenda for next Meeting: