| |
|
|
|
|
ncsa |
Running Dedicated Jobs on the NCSA Origin2000
- Jobs can be run in dedicated mode via the dedicated queues available
in the batch system. Jobs in these queues run one at a time on a dedicated
system. The queues currently available can be found in
the documentation on
Running Jobs on the Origin.
-
For best performance, it is recommended that you do not use all the
processors on the dedicated system. Leave 1-2 processors free for use by
the operating system and other system processes.
Also, you should never have more threads/processes than there are
processors.
-
Jobs that run in dedicated queues are charged for all the processors on the
dedicated system, no matter how many processors are actually used. Also, there
is a premium charge for running dedicated jobs.
See the
Charging Algorithm
for information.
The dedicated queues are recommended for benchmarking and for production
runs
that use more than 64 threads/processes. To be cost effective, production
runs should use as many processors as possible, leaving 1-2 processors for
use by the operating system.
Jobs using up to 64 threads/processes can be run in
timeshared queues.
- SGI recommends setting the following environment variables to lock each
thread/process to a processor. In the C-shell, this is:
setenv _DSM_MUSTRUN for parallel shared memory jobs
setenv MPI_DSM_MUSTRUN for MPI Jobs
- For parallel shared memory jobs, SGI recommends disabling dynamic threads
and gang-scheduling when running in dedicated mode..
In the C-shell, this is:
setenv OMP_DYNAMIC FALSE
setenv MPC_GANG OFF
- Users whose codes do not scale to the number of processors available in the
dedicated queues can still use the dedicated queues to run multiple jobs for
faster turnaround. See
Running Multiple Jobs in the Dedicated Queues on the NCSA Origin2000
for information.
|
|
|
|
|
