Re: [Gems-users] slow down using pthreads

[Dan Gibson <degibson@xxxxxxxx>]

It would seem that 63 of the processors are executing the OS idle loop very efficiently, while one is spawning your threads, executing your threads, and running the rest of your process setup and teardown. There is a big simulation cost to having large numbers of processors in your simulations.

This is why we usually tell folks to start ruby after all threads have been forked (pthread_creat()ed), via a magic call.

Regards,
Dan

Brad Moore wrote:

Hey all,

I've been going through a bunch of the "too slow" threads, and I haven't been able to figure out if people
solved their issues, or just dealt with it being slow.

I've been working with MSI_MOSI_CMP_directory, on Sarek, with 64 cores/processors on a single chip.

I have a benchmark I've run (using pthreads).  It spawns two threads, and then the main thread
waits for their completion.  Each thread executes an empty function.

The execution took 8hrs, here's a clip of the stats file:

Total_misses: 9769
total_misses: 9769 [ 1909 74 61 63 60 57 61 67 58 64 61 65 71 60 56 64 59 60 59
66 60 62 60 694 5595 5 5 5 6 5 5 4 5 5 5 6 5 6 6 5 5 6 5 5 5 5 6 5 6 4 6 4 5 6 6
 5 5 5 6 4 6 4 6 5 ]
user_misses: 862 [ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 147 715 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ]
supervisor_misses: 8907 [ 1909 74 61 63 60 57 61 67 58 64 61 65 71 60 56 64 59 6
0 59 66 60 62 60 547 4880 5 5 5 6 5 5 4 5 5 5 6 5 6 6 5 5 6 5 5 5 5 6 5 6 4 6 4
5 6 6 5 5 5 6 4 6 4 6 5 ]

instruction_executed: 1297255830 [ 742697 20746903 20744087 20750036 20753361 20
762925 20744119 20734758 20751076 20741291 20745643 20741295 20704298 20755676 2
0764321 20749043 20750176 20759319 20753849 20733527 20750740 20743099 20571572
17260854 435687 21096508 21096659 21095505 21094517 21096362 21096307 21097062 2
1094905 21096437 21095760 21094580 21097473 21096164 21095831 21095639 21095560
21094589 21095370 21095410 21096449 21095421 21095607 21096475 21094764 21096264
 21094548 21096134 21095489 21095448 21095393 21095571 21095687 21095840 2109483
6 21097020 21094194 21097636 21094171 20927893 ]

Thanks,
-Brad

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.