[torqueusers] MPI and LOAD_BALANCING
cmesunoom at yahoo.cn
Wed Apr 7 19:59:44 MDT 2010
I have two problem,
First, TORQUE can run with any message passing library and provides limited integration with some MPI libraries. Does it have any advantage?
Second, I set the option load_balancing in sched_config as fllows:
load_balancing true prime
and, I have set all nodes type, time-sharing. Then, set the ideal_load and max_load in every mom_config.
However, when I submit a job, tracejob jodID said that:
Job Deleted because it would never run.
Not enough of the right type of nodes available.
If I set the nodes type cluster, the job can run, but this job run on the node which has heavier load, not the lower load node!
Why is it? Does the load_balancing not support load balancing? Does the load_balancing not support time-sharing nodes?
Any help will be appreciated!
More information about the torqueusers