Parallel Problem of CFX 14.5.7 on a Intel Xeon 2690V2 computer with Win7 x64 OS
 

Cpus of my computer is Intel Xeon 2690 V2, the OS is windows 7 x64 Ultimate, and the edition of CFX is 14.5.7, in which Platform MPI 8.2.1 is the default parallel soft for local and distributed parallel. When "Platform MPI Local Parallel" computation starts, there is always an mpid.exe error showed by the Os with a messagebox, and the computation is shut down. I find the error block is in libmpirun.dll from the detailed information of the error.

btw, CFX 12.1 doesn't have this problem.

Any advices? Thanks a lot.

Try a re-install of Platform MPI, make sure you have admin privileges and "run as administrator".

Make sure Firewall has exceptions for MPI.

Run the MPI test program to see if that works.

thanks for you reply, but that doesn't work.:mad:

I re-install Platform-MPI from Ansys 14.5.7 installation cd by right-clicking setup.exe and chose "run as administrator", then on the GUI click "Install MPI for Ansys Parallel Processing" and then "Install Platform-MPI"....

there is no Firewall installed in the computer.

Does the MPItest program work?

Local MPI Test:
Create a folder "temp" in C:

Open a Command Prompt and change to this newly created directory: type in:
cd\temp
Then press enter.

then type in

"C:\Program Files\ANSYS Inc\v150\ansys\bin\winx64\ansys145.exe" -np 2 -mpitest
Then press enter again.
Did this work?

Yes！ that works.

"MPI test has been completed successfully."

and then, what need I do? thanks.

Well, I'm not sure. This just tells me that MPI is working correctly. For some reason CFX doesn't use it correctly. Could have something to do with you having a previous version of MPI installed, so when it tries to run, it uses the older, incompatible version?

If It was me I would uninstall and erase all traces of the old software, then start fresh with only the new version.

Or try to use Intel MPI instead, then there shouldn't be any confusion between versions.

thanks, evcelica, you said "try to use Intel MPI instead", but how? although I installed Intel MPI 4.0.3, there are only "Serial","Platform MPI local parallel" and "Platform MPI distributed Parallel" items in the "Run Mode" selection list.

I'm using version 15, and Intel MPI is right there on the drop down list. Looking at my previous versions 14, 14.5, I see there is no Intel MPI option. So maybe its not possible on those versions.

Do you have the environmental variables set:
MPI_ROOT set to: C:\Program Files (x86)\Platform Computing\Platform-MPI

If I was you I would probably uninstall everything and clear all the environmental variables related to ANSYS.
Then Re-install the newest version and see if that works. Windows has a firewall, so make sure you disable it for de-bugging.
If it still doesn't work, I would do a clean re-install of windows.
I know, pain in the a$$, but it seems like something is set wrong on your system, possibly from the previous install, and a clean O.S. install may be the only hope if all else fails.
Why would you install 12.1 on such a new system anyways?
Or install version 15 and see if that works. That way you could try Intel MPI.

I re-installed the O.S and now am using the version 15, still the Platform-MPI doesn't work, fortunately, Intel MPI works.

I installed 12.1 after the problem of 14.5.7, to find if 12.1 can work on such a new system and some computation work need to be done in a short time, so I have to let the computer work.

And now, the O.S and Ansys are all new, Platform-MPI still doesn't work for CFX. So maybe, there is a compatibility problem with Platform-MPI, CFX15 and Intel Xeon E5?

thank you very much.

I've used platform MPI on XEON E5v1, Sandy Bridge E i7s, and Ivy bridge-E i7s. Never on E5v2 though. That is very strange that it didn't work for you?!? I'm glad you got Intel MPI working though, good job.

Hello!

Thank you very much for your help and support in this site!

I have Ansys CFX version 15 with Windows 7 Professional O.S (64bit) and the Intel MPI doesn't work e return with this message:

Host name lookup failed for host "namehost"

abort: Unable to connect to “RemoteUserName:8678"
sock error: Error = -1

…
Command on host hostname exited with return code 0.

It seems that the problem is communication with another machine. is it necessary to run Intel MPI to have rsh configured on Windows machines? The guidance provided by ansys is not exhaustive...

again, thank you!

Yes, you have to have remote desktop privileges to all machines.
Did you set your hosts.ans file to include the relevant computers and information?
Do you have your password cached?

I have partially solved the problem by installing the version of ansys 14.5 and using the Platform MPI (IBM) 8.2.1. Now parallelization distributed with this Platform MPI work, but I have found two new problems: when I start the .bat file setpcmpipassword.bat and I write one password the program return this message:

C:\Program Files\ANSYS Inc\v145\commonfiles\MPI\Platform\8.2.1\Windows>se tpcmpip
assword.bat

Password for MPI runs:
mpirun: Drive is not a network mapped - using local drive.
MPI Application rank 0 exited before MPI_Init() with status 0

Microsoft Windows [Versione 6.1.7601]
mpirun: ERR: Check Failed - cached password is invalid.

Ok...but by ignoring this, now when I start one distributed simulation with only two core I have found this another problem:

HPMPI:CPWB<3388>:<4492> <CYCLIC> AFFINITY method not applicable with single ldom.
HPMPI:CPWB<3388>:<4324> <CYCLIC> AFFINITY method not applicable with single ldom.

where the affinity CPU is not correct, each solver-pcmpi.exe was given on all cores.

Any ideas? Thank you evcelica for your support :-)

Sorry, I'm not sure.
Does distributed work when you use more than just 1+1 cores?
If not, do you have a password? When I run the set password script, I just type in the password and hit enter, I get no messages at all, not even a confirmation, then I just close the command window.

Hi, I have got same message: AFFINITY method not applicable with single ldom. Does it mean that the parralization is not working? Although this message appears, the four cores is involved.