[comp.sys.encore] barrier synchronization

blj@ravel.csg.uiuc.edu (Bob Janssens) (12/18/89)

We are running UMAX-4.2 R3.3.0 on our 8 processor Multimax 510.  I am having
some trouble with the barrier synchronization.  About 20% of the time my
simple gaussian elimination program deadlocks at one of the barriers.
However, print statements show that every process reaches the barrier.  I
have tried the fbarrier and the regular barrier with all three types of
blocking, and get similar results with all.  I also have an account on the
Argonne National labs encore which is running the beta version of UMAX-4.3.
Here my program runs fine without any deadlock.  Other users on our system
have had similar problems with parallel programming.  Could this be a 
problem with our OS?  Have other people had similar problems?

Thanks,
Bob Janssens				b-janssens@uiuc.edu
Center for Reliable and 
High Performance Computing