Problems during ABAQUS installation

[Indenter]

Hi everybody!

I hope anyone can help me with this problem.

In my group, we have recently bought a little computing cluster composed
by a master node with 1 cpu and three computing nodes, each one with 2 QuadCore processors. This is the first time I deal with this kind of installation, until now I was used to run ABAQUS in Windows work stations, and who sold us the cluster had no idea about how to install ABAQUS.

The OS installed is CentOS, the cluster is administrated through RocksCluster
and the ABAQUS version installed is the 6.5 (I know it's really old... but the acquisition of new software isn't my privilege :( )

So we have all this stuff installed in the master node, we can submit jobs using batch files (through SGE) and also through the command line... but just with 1 cpu.

Each time we try to use various cpu's for one job (just using THREADS, not MPI) we get the crazy message

The number of cpus (X) exceeds the number of cpus available (1).

Being X less than 8. No matter if we try that with a batch file (the job is assaigned to one of the computing nodes) or from the command line of one of the computing nodes.

The problem seems to be that, once we installed ABAQUS in the head node, it recognaized just 1 cpu so now, each time we submit a job, it runs the executable ABQsysconf.exe that tell ABAQUS that just 1 cpu is available.

Could the problem be relate with the installation? Should I edit any file? Should I threaten the cluster with a hammer?

Any comment will be really appreciated

[Jorgen]

The hammer idea might be interesting ;)

I believe you need to modify your abaqus_v6.env file to specify the names of each node and the number of CPUs each node has. Then Abaqus should be able to distribute the analysis appropriately.

Note that it is typically recommended to use MPI instead of threads. The MPI library is usually installed at the same time as you install Abaqus.

-Jorgen

[Indenter]

Hi Jorgen,

Thanks for your help. I have already tryed to modify the .env file (being that at the same folder from which I'm submiting the job) adding the name of my computing nodes and the available cpu's... and I get the same message! I'm trying to use just THREADS to skip problems wit shared libraries and so one. Once I manage to run with THREADS I would like to change to MPI.

Any other idea?

[Jorgen]

Hmm, it just hit me that THREADS might not work when you have multiple computers that are connected by ethernet. It might be that THREADS only work on a single computer with multiple CPUs.

I recommend that you try using MPI, which works when you use multiple computers...

-Jorgen