Could it be that your workling set size fits in 16 caches but not in 4
caches? AKA what are your cache sizes and configurations and what is
the data set size?
Also, what mechanism are you using to start/stop simulation?
Regards,
Dan
On Thu, Jul 17, 2008 at 12:01 PM, Huan Fang <huanf@xxxxxx
<mailto:huanf@xxxxxx>> wrote:
Hi All,
I have met a real strange problem that concerns user misses. When
I run
a parallel application on 4cores, the misses are normal:
Total_misses: 3909948
total_misses: 3909948 [ 988120 975421 973131 973276 ]
user_misses: 3864534 [ 951449 972051 970646 970388 ]
supervisor_misses: 45414 [ 36671 3370 2485 2888 ]
98% of total misses are user misses.
But with 16 core configuration, the application finishes really fast
(only 5% ruby cycles compared to 4-core), and the user misses are
close
to 0:
Total_misses: 1597
total_misses: 1597 [ 174 14 1 4 52 34 3 11 80 10 37 36 1105 1 1 34 ]
user_misses: 25 [ 5 0 0 0 3 4 0 0 0 0 0 1 12 0 0 0 ]
supervisor_misses: 1572 [ 169 14 1 4 49 30 3 11 80 10 37 35 1093 1
1 34 ]
I am sure ruby records the same area of the application. But the
result
is simply unreasonable. Is there any thing I didn't notice?
-Huan
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx <mailto: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.
--
http://www.cs.wisc.edu/~gibson <http://www.cs.wisc.edu/%7Egibson>
[esc]:wq!
------------------------------------------------------------------------
_______________________________________________
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.