bug fix for issue https://github.com/GlobalArrays/ga/issues/371

[edoapra]

No description provided.

[bjpalmer]

I need to check this a bit. There is a possibility that g_state.node_size is modified when the number of progress ranks is increased.

[edoapra]

I need to check this a bit. There is a possibility that g_state.node_size is modified when the number of progress ranks is increased.

That's what I have initially thought, but after adding some debugging printouts, this is what I saw

GA_NUM_PROGRESS_RANKS_PER_NODE NPROC #."Compute procs"   g_state.nodes_size
2                                 4      2                  2
2                                 6      4                  3
3                                 9      6                  3
1                                 9      8                  9
2                                 4      3                  4

This shows that

# computer procs = GA_NUM_PROGRESS_RANKS_PER_NODE*(g_state.node_size -1)

Over the last few days, I have done some extensive testing on the system where I discovered this issue and the fix seem to be correct: it stops the execution when /dev/shm is getting full, while the code does not stop when this does not occur in a condition when GA_NUM_PROGRESS_RANKS_PER_NODE > 1