I have a CFX-5.7 model which ran fine from the original def file using only a fraction of the memory on the machine (MPICH Parallel on 2 CPU's). The restart fails with the message that there is not enough memory on the machine! I even tried running the original def file with the 001.res file as the initial guess; same result. It makes it through the partitioner, and then fails when starting the solver. I'm baffled.

Anybody seen this behavior or know what's causing it?

Thanks, Jeff

Restarted this problem today with PVM instead of MPICH. It ran with almost 0.5 Gb free memory left on the machine. It must be an MPICH issue.

Jeff

Hi Jeff,

This error can also be caused by CFX incorrectly guessing how much memory it needs and not allocating enough memory. Try giving a memory allocation factor of 1.2 (or maybe higher if necessary) and try again.

Glenn Horrocks

I did this, but it didn't work. The message was clearly stating that there "wasn't enough on the machine", not that the job didn't have enough to run. It looks like MPICH was trying to allocate quite a bit more than it actually needed to run the job and hitting the ceiling. Why it ran the initial case is still a mystery. The problem is now converged with PVM - any port in a storm.

Jeff

Hi Jeff,

You may have found a problem with MPICH in CFX5. Can you provide details to your CFX vendor so if it is a bug it can be fixed?

Glenn Horrocks