Currently, the -j option of mosrun randomly picks one of
the nodes in the -j list, regardless of the relative load of
the nodes in the -j list.
This proposal is to do one of the following:
1. The -j option to mosrun should take a one-time
snapshot of the load on the nodes provided in the -j list
and pick the least busy one.
2. A new option (perhaps -b, for "balance") should be
added to mosrun that takes a range of nodes like -j, but
chooses the least busy of the nodes.
mosrun -j2-7,16 /path/blah
randomly picks of the the 8 nodes.
mosrun -b2-7,16 /path/blah
picks the least loaded node at the time of the mosrun
request (after which point the process is locked on that
chosen node, just as with -j today).
Log in to post a comment.