Hi,
We are trying to setup mpi runs of the molecular dynamics program CPMD on
our new xeon machines.
dual processor 3ghz 6gig ram. Our older machines are dual athlon 2ghz+ with
2gig ram.
We have been benchmarking performance to ensure that we have everything
setup corrrectly and have found some rather peculiar results.
1. 1 xeon - 1 processor - 23 seconds a step
2. 2 xeons - 1 processor each - 19seconds a step
3. 1 xeon - 2 processors - 19seconds a step
4. 3 xeons - 2 processors each - 13seconds a step
The fact that 2 and 3 are the same is worrying. lam-mpi is version 6.5.9
and was configured with rpi=usysv - ifc/icc was the base compiler. Our copy
of CPMD is linked against intel math libraries. It was also compiled with
ifc.
On the athlon's - with same code and same lam
1. 1 athlon - 1 processor - 13minutes total running time
2. 1 athlon - 2 processors - 8minutes total running time.
This shows that the code is capable of much higher scalability then is shown
on the xeon. Scalability of CPMD on the APAC SC, aproaches linear, but its
memory architecture is undoubtably superior to athlon and xeon. (Steps and
total running time is virtually proportional)
Using NetPipes - mpi performance betweeen 2 jobs on same machine:
xeon - latency 0.9usec - peak bandwidth 9GBit/sec
athlon - latency 1.2usec - peak bandwidth 4GBit/sec
bandwidth curves follow same shape - just xeon double the speed.
MPI performance between xeons, 100mbit connection, latency 90usec, bandwidth
peak of 85MBits/sec
All benchmarking was done with -O passed to mpirun.
The xeon machines mpi characteristics as measured by netpipes are superior -
but it gets a performance increase going from 1 to 2 cpus inferior to
athlon. In addition, the performance gain for 2 cpu's on the same machine
only matches that of 2 machines with 1 cpu each running across the network.
Any suggestions for how we can improve the mpi performance on our xeon
machines?
Regards,
Gareth
_________________________________________________________________
ninemsn Extra Storage is now available. Get larger attachments -
send/receive up to 2MB attachments (up to 100 percent more per e-mail). Go
to http://join.msn.com/?page=dept/home&pgmarket=en-au
|
