Previous: ROMIO_MPI_IO
Next: Setting_Threads
View by Date; View by Name

Scheduler Change

Sept 15, 1998

The following is for SGI MP and OpenMP parallel codes (compiler options -pfa, 
-pca, and -mp). Other types of codes will not be affected.

SGI recommends the use of dynamic thread scheduling for best performance
with the MIPSpro 7.2.1 compilers (the default on the NCSA Origin Array since
May 25, 1998). With dynamic threads, the number of threads that are used for
executing parallel regions can be adjusted by the run time environment to 
best utilize system resources. The current default at NCSA is gang-scheduling, 
which is enabled globally via the environment variable MPC_GANG.

The default on the NCSA Origin Array has changed to dynamic thread scheduling
starting today. In general, users should see improved performance of their
codes. Executables that were compiled with older versions of the compiler 
(i.e., compiled before May 25, 1998) should be recompiled for optimal 
performance. 

See the pe_environ man page for information on various environment 
variables available on the Origin.

Silicon Graphics Origin2000:usr/news/Scheduler_Change
Last Modified: September 15, 1998

ncsa

Scheduler Change