Software Coordinating Committee Conference Call January 18, 2007 1:00 PM EST Recorder: C. DeTar Present: Brower, DeTar, Levkova, Joo, Zhang, Efstathiadis, Clark, Watson, Renner, Pochinski, Fowler, Gottlieb, Holmgren, Simone, Khoriaty, Jung, Osborn, Edwards Absent: Mawhinney, Scholz, ===================================================================== ** Action items Note: New webpage for SciDAC-2 Software Agenda & Minutes & Related Documents: http://super.bu.edu/~brower/scc.html (Linked from http://physics.bu.edu/~brower/) Balint's Level 3 Design Considerations document is posted. ============= Agenda ================= 0.1 Any news of usqcd.org ? (Don, Chip ????) Brower: Can we have the web page ready by early Feb? ** Watson: No problem. Holmgren: The FNAL folks should be ready soon. Joo: Then I could do a redirect to FNAL. Holmgren: Some links are broken. Also the links to CVS heads are empty. Edwards: QIO, QLA, QDP/C and LIME are circular links. Here are the responsibilities: QDP/C++ Edwards/Joo QOPQDP Osborn QDP/C Osborn QLA Osborn QIO DeTar C-LIME DeTar CGDWF Pochinski ADAT Edwards/Joo BAGELQDP Edwards/Joo ENSBC Edwards/Joo Joo: Here are the CVS node names so maintainers can work on their pages: usqcd-docs HTML usqcd-software Tar balls Osborn: It would be nice to update the web page from the doc site. Edwards: The current separation was done by previous request. Osborn: Can we do server-side includes? Then we can update a version number in the source package and automatically get it into the web page. Edwards: Clever idea. Joo: It is probably turned off. I'll check on it. 0.2 Visualization -- Does anyone have a PhD candidate for Massimo? (Massimo if he can is able to make the call) Brower: Please send candidate names to Massimo. 1. Light weight solution to web based data base on software info (Rob, Ying, ...) Fowler: The PERI group had some discussion. They wanted to know what are the requirements. Brower: Some of our BU classes are using wikis. They look quite versatile for our needs. I can look into it further. Joo: We aren't really looking for a real database -- just a way to collect wisdom. 2. Workflop document (prototype?) (Jim, Don) Holmgren: Please see my email circulated before the meeting. The existing workflow packages are very grid-centric We wrote up an example of a typical LQCD project and produced a list of requirements. We have a Wiki for this project. The IIT folks would like to take code and set up a pilot project at NCSA. Once we have software for connecting VDS to our PBS, we should be ready to test -- say February. LQCD_Workflow_FuncReq.pdf Joo: Can you include code builds in the dependencies with these tools? Simone: I brought this up. It doesn't seem that that was in the plan. Fowler: We have been using Diverna, which is not that ambitious. It is more oriented toward biology. We can run flood model stuff with it. We could use it to control benchmarking experiments to play with compiler options on several different machines. The process is simple. We just pass in the parameters. Workflow just manages parameter input and not any complex dependencies. ** Brower: Please look at this document and comment next week. Holmgren: I'll forward information about the Wiki. 3. Progress on comparison of performace for DW Level 3 codes for BG/L (Andrew, Chulwoo, James,Balint) Brower: I circulated Balint's e-mail this morning. Joo: We are still working out standards for making comparisons. Brower: We have a variety of lattice sizes, so need a variety of benchmarks. DeTar: Shouldn't we have MFlops for a variety of local volumes and then also reporting time to completion for a standard problem. Fowler: How about time for iteration and number of iterations? Jung: To allow for different stopping conditions, people could take time per iteration.and estimate for their own case. Brower: We are trying to determine the best choice for molecular dynamics. So my main concern is being able to choose the best on our INCITE and leadership class machines. Brower: Why are there differences? Joo: Different preconditioning, different flop counts. DeTar: What if we use time to completion for a given stopping condition? Joo: The stopping condition depends on preconditioning? DeTar: So, what if we define it in terms of the unpreconditioned problem? Joo; The MD may not need the unpreconditioned solution. Pochinski: How about two numbers, one for unprecond. and one for precond. Brower: I would start with unpreconditioned, as needed for spectroscopy. Is that reasonable? Joo: The benchmarks I have been running so far have been for spectroscopy. Would that be useful for your MD on leadership machines? Khoriaty: I would be interested in implementing the benchmarks. Joo: I would be happy to provide them. Brower: It would be good for Joy to take over some of this. Brower: On the BG/P we will be encouraged to run on many processors, so small local volumes. QLA for BG/L Osborn: The test suite seems to work. Joy, you were getting more errors than I. We are probably ready for a pseudorelease. BGL_APP_L1_SWOA Joo: What is the BGL_APP_L1_SWOA store-gathering flag? This would be good to include in our "collected wisdom Wiki". Jung: On BG/L Bagel runs faster. Fowler: On Opterons there are similar strategies. It helps with cache management. 4. Plan for miminal CRE implementation (Chip/Balint, Don, Statos) Watson: This has been held up because our procurement is still not finished. We are struggling with the continuing resolution. We are hoping we can award the procurement by next week. 5. When do we start to prototype and design API for multi-core? (Robert,...) Topic postponed. 6. Level 3 standard DeTar: Progress on the Asqtad RHMC has been held up by the desire to have a Level 3 standard first. It would be good to have the Asqtad multimass inverter and vectorized fermion force very soon in advance of the USQCD proposal. Jung: I didn't want to have to write and then rewrite the wrapper. Brower: It isn't good to hold this up. Jung: If we are not going to be all inclusive about the API, I can do it. Brower: I suggest we proceed with James's current standard and we can work on the standard in parallel. Brower: Let's have the interface published and then proceed. Jung: I'd need a test case for validation. DeTar: We just run a full test problem. We can switch back and forth between Level 3 and non Level 3 and compare. I'll be glad to help. 6. Software committee meeting at All Hands? Brower: I thought there were too many meetings around the Allhands meeting, so not good to have yet another. Committee conference concluded at 2:30 PM EST. Next call Jan 25 at 1:00 PM EST ======================================================================