Distribution of the task to different cluster nodes in the solar grid engine (SGE)

Question

I am trying to get sun gridending (sge) to start separate MPI job processes in all nodes of my cluster.

What happens is that each node has 12 processors, so SGE assigns 12 of my 60 processes to 5 separate nodes.

I would like it to assign 2 processes to each of the 30 available nodes, because with 12 processes (dna sequence alignment) running on each node, the nodes run out of memory.

So I'm wondering if it is possible to explicitly pass SGE to assign processes to a given node?

Thank,

Pavel.

Answer 1

, , 2 12 node.

 qconf -sq queuename

. , 5 4 () .

....
slots                 1,[master=4],[slave1=4],[slave2=4],[slave3=4],[slave4=4]
....

qconf -mq queuename

4 2.

Answer 2

"alloc_rule" ; , $pe_slots alloc_rule, -pe qsub , .

Answer 3

"qconf -msconf", . . "load_factor". "-" ( ).

, , . exec , . exec, , , , - , load_factor ( , ).

. , , , . , load_factor, , , . "qconf -me hostname", "complex_values", "slots = 16", 16 - , .

Answer 4

, . SGE script:

#$ -l nodes=30,ppn=2

Requests 2 MPI p rocesses p er n ode (ppn) and 30 nodes. I think there is no guarantee that this 30x2 layout will work in a 30 <node cluster if other users also run many jobs, but maybe you can try.