Parallel with Windoze, speed difference between PV

March 9, 2005, 08:43

I am running CFX 5.7.1 parallel on a mixed cluster of machines, all with Windows. The cluster consists of a dual Xeon 3.6, a quad Opteron 248 and 5 3.2 GHz P4's. Some of the tests that I have run have shown a dramatic improvement in wallclock time when going from PVM to MPI. Can anybody throw some light on this? I vaguely recall somebody once mentioning that PVM on Windows is not nearly as quick as on Linux?

March 9, 2005, 17:24

Hi,

On windows machines MPICH for windows is usually much faster than PVM. In my benchmarking running MPICH on windows is roughly the same speed as running PVM on Linux. I have no idea why the difference, but that's the end result.

Glenn Horrocks

March 9, 2005, 22:42

The PVM implementation really sucks on Windows and is especially bad when you run on dual Xeons. MPICH is better but is less stable.

On a quad opteron you should be seeing superlinear performance if your case is big enough (PVM or MPICH)

March 10, 2005, 02:25

Thanks guys, your comments are very much in line with what I've been observing.

March 9, 2005, 08:43	Parallel with Windoze, speed difference between PV	#1
Charles Guest Posts: n/a	I am running CFX 5.7.1 parallel on a mixed cluster of machines, all with Windows. The cluster consists of a dual Xeon 3.6, a quad Opteron 248 and 5 3.2 GHz P4's. Some of the tests that I have run have shown a dramatic improvement in wallclock time when going from PVM to MPI. Can anybody throw some light on this? I vaguely recall somebody once mentioning that PVM on Windows is not nearly as quick as on Linux?

March 9, 2005, 17:24	Re: Parallel with Windoze, speed difference betwee	#2
Glenn Horrocks Guest Posts: n/a	Hi, On windows machines MPICH for windows is usually much faster than PVM. In my benchmarking running MPICH on windows is roughly the same speed as running PVM on Linux. I have no idea why the difference, but that's the end result. Glenn Horrocks

March 9, 2005, 22:42	Re: Parallel with Windoze, speed difference betwee	#3
Neale Guest Posts: n/a	The PVM implementation really sucks on Windows and is especially bad when you run on dual Xeons. MPICH is better but is less stable. On a quad opteron you should be seeing superlinear performance if your case is big enough (PVM or MPICH)

March 10, 2005, 02:25	Re: Parallel with Windoze, speed difference betwee	#4
Charles Guest Posts: n/a	Thanks guys, your comments are very much in line with what I've been observing.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
operation speed Difference between 32bit & 64bit	IUIU	FLUENT	1	February 9, 2011 23:06
Increase speed of parallel computation	Purushothama	Siemens	2	November 30, 2010 15:51
Parallel speed up	Soren	CFX	18	May 31, 2002 13:26
Parallel speed up for CFX 5 on PC's	Roued	CFX	6	November 28, 2001 19:02
speed up ratio at parallel processing	Kim hak-gyu	Main CFD Forum	1	October 25, 2000 10:57