Hello to all,
Below is part of an email request from one of my users - I would
appreciate any comments - suggestions or resolutions.
thanks:
I have yet another request regarding the parallel environments in
elcap. The ompi_fillup PE has partially solved the problem of
shared-memory parallelized programs, but only partially. The new
issue is that if I request 4 slots and the queue assigns me a node
where there is, say, 1 slot already reserved, then the calculation is
going to be split in different nodes and fail.
My questions are:
i) is it possible to create a new PE to make a 4-slot calculation be
queued until a node is available with all 4 slots free?
ii) is there a way of doing this using some clever job script flag?
iii) can the queue system be made to accumulate 1-slot jobs in as
few nodes as possible to avoid the above problem?
--
Joseph Norris Applications Developer/Server Administrator
University of California Merced - School of Natural Science
209-201-3410
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
