This is Lite Plone Theme
You are here: Home System Status On-going woes on cosmos2 [UPDATED]

On-going woes on cosmos2 [UPDATED]

Dear users,

Cosmos2 is back on-line, but unfortunately while waiting for the final diagnosis of what has been happening this week, we have had to turn off the XPMEM acceleration and MEMMAP acceleration from MPI jobs. We know some of you have used these features explicitly (i.e. not relying on the defaults which we have now changed) in the past, and we kindly request you to refrain from using MPI_USE_XPMEM, MPI_XPMEM_ENABLED and MPI_MEMMAP_OFF variables.

This will have a small performance penalty on MPI jobs until SGI can sort out the root cause of the issue and fix it. The size of the penalty depends on the specifics of how the MPI ranks communicate with each other: some jobs might see no penalty at all.

We are sorry for any inconvenience caused.

Best regards, COSMOS Management

Original news item was:

Dear users,

There is a problem which has plagued cosmos2 this week and which seems to be repeating itself. We will need to reboot it once again, killing any jobs in the process.

We are sorry for the inconvenience caused.

Best regards, COSMOS Management