[Bioclusters] limiting job memory size on G5s

Barry J Mcinnes Barry.J.Mcinnes at noaa.gov
Wed Dec 15 16:27:22 EST 2004

We have a test cluster of G5s running 10.3.6 and GE 6.
We want to be able to limit the physical memory usage, that works on SGE 
under Solaris.
The current limits for the queue are set via qmon, and are printed out 
for each job as

cputime         unlimited
filesize        unlimited
datasize        2097152 kbytes
stacksize       65536 kbytes
coredumpsize    0 kbytes
memoryuse       1048576 kbytes
descriptors     1024
memorylocked    unlimited
maxproc         100

When we run our memory allocating test job, grabbing n times 4 bytes, 
the following
happens, 2, 2.5, 2.75 x 10**8 all run fine, for
300x10**8 GE tries running the job on a node then puts the node in E state,
tries another node, puts it in E state, then actually runs on a third node.
All G5s are identical hardware. By my calculations the 2.75x4x10**8 
should fail.

Anyway is there a way to physically limit a job to use only 1GB of 
memory and the job fail
with a error without putting the node in Error and locking the node for 
other jobs ?

thanks barry

More information about the Bioclusters mailing list