[Gimle-users] Gimle login node rebooted
Kent Engström
kent at nsc.liu.se
Mon Mar 28 09:15:56 CEST 2011
Dear Gimle Users,
the Gimle login node had to be rebooted due to problems accessing
Lustre filesystem etc.
The symtoms do not look like the ones we had early this year; that
problem does indeed look like its solved.
The symtoms do look like they did for the crashes we hade in November
last year. Those crashes were caused by one or several processes
doing "du", "find" or similar operations that read file attributes
for vast portions of the Lustre filesystems, making the kernel run
out of slab memory.
We must advice you to use a compute node (allocate interactively using
"interactive -N1" or run as a batch job) if you need to do things like
that.
--
Kent Engström, National Supercomputer Centre
kent at nsc.liu.se, +46 13 28 4444
More information about the Gimle-users
mailing list