I see your point, but the cons make sink the idea for me. How about a compromise -- write up a scripty-foo to automatically download and build some of the more common benchmarks? This still makes it a trivial exercise for the user, but it avoids us needing to bundle already-popular benchmarks in OMPI (plus, they release at different schedules than us).
For extra bonus points, you could make the scripty-foo be a dumb client that downloads an XML file from www.open-mpi.org that indicates where it should *really* download and build a given benchmark from. This would allow us to "release" new benchmarks independent of Open MPI releases (e.g., if NetPIPE releases a new version, we can just update the XML file on www.open-mpi.org). On Apr 1, 2010, at 7:14 PM, Eugene Loh wrote: > I wanted to know what folks thought about adding a ping-pong performance > test to the examples directory? > > Pros: This would facilitate performance sanity testing by OMPI users -- > particularly MPI neophytes. It would add something to the examples > directory with a performance orientation. It would give us > (devel@ompiorg) a known quantity when trouble shooting performance with > users. It could conceivably raise OMPI visibility in the MPI world. It > could be a stepping stone to developing a more complete set of MPI > performance sanity tests with time. > > Cons: There are already many performance tests. We shouldn't be > replicating what others do, but should be leveraging what they do. > Other existing tests are relatively easy to use and already familiar to > many users. > _______________________________________________ > devel mailing list > de...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/devel > -- Jeff Squyres jsquy...@cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
