[Beowulf] hpcc MPIRandomAccess bug
joachim at ccrl-nece.de
Thu Aug 25 00:22:17 PDT 2005
Håkon Bugge wrote:
> Running hpcc in its latest version, I have from time to time been
> exposed to missing updates in the MPIRandomAccess benchmark. The
> benchmark tolerates upto 1% missing updates, hence the benchmarks
> "passes" as such. However, when running a pure MPI implementation of
> this benchmark, missing updates are unacceptable. Hence, I investigated
> the issue. Would be interesting to know if anyone else has been exposed
> to the same.
This effect might not show up with other MPI's due to the MPICH-way of testing.
MPI_Testany always processes the array of requests from 0...N-1. If the order in
the array relates to the "age" of a request (which is not unlikely, although I
didn't look in to HPCC), the problem you saw won't show up. And as most MPI's
(unlike ScaMPI) are derived from MPICH...
But I wonder why HPCC tolerates 1% missing updates at all! What's the
justification for tolerating (exactly 1%) data loss?
Joachim Worringen - NEC C&C research lab St.Augustin
fon +49-2241-9252.20 - fax .99 - http://www.ccrl-nece.de
More information about the Beowulf