|
|
||||||||||||||||||||||||||||||||||||||||||
| > home > support > FAQ > differences | |||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||||||||||||||||||||||||||||||||||
Default differences |
|||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||||||||||||||||||||||||||||||||||
MP_SHARED_MEMORY=yesMP_POLLING_INTERVAL=1000000MP_TIMEOUT=300MP_SHARED_MEMORY=noMP_POLLING_INTERVAL=400000 (for US and 180000 for IP)MP_TIMEOUT=150We define our LPARs (Logical PARtions) to be the same as two MCMs (Multi-Chip Modules), which are tightly coupled, so that we may use all the switch channels to allow maximum communication through-put. For HPCx, each LPAR has 16 processors.
If your job is a parallel job, then you must specify job_type=parallel and node_usage=not_shared
If your job is a serial job, then you must specify job_type=serial and node_usage=shared
If your job is an interactive job, then you must specify job_type=parallel and
node_usage=shared
LoadLeveler imposes a maximum memory per task to be 14.4GB/(number-of-processes),
where number-of-processes is the number_of_tasks in a purely
MPI job, or number_of_tasks*OMP_NUM_THREADS, in a
mixed-mode code.
| http://www.hpcx.ac.uk/support/FAQ/differences.html | contact email - www@hpcx.ac.uk | © UoE HPCX Ltd |