Its more likely that something is causing a backup of message processing
at the cache controllers. Do the controllers stop processing messages?
Also try increasing the NUMBER_OF_TBE parameters in the
rubyconfig.defaults to very high values.
--Mike
Liqun Cheng wrote:
Hi,
When debugging "Possible Deadlock detected" problem in the
MOESI_CMP_token protocol, I found lots of messages jammed in some
throttles. The debug filter in network/simple/Throttle.C:267 can print
out hundreds of stalled messages. I have increased the link bandwidth
to 100000 (100 bytes per cycle), g_DEADLOCK_THRESHOLD to 500000,
persistent request invoking threadshold to 50 average miss times and I
have also tried adaptive/non-adaptive routing algorithms. But all
above can't solve the problem. Any idea what else can be wrong?
thanks a lot!
Legion
------------------------------------------------------------------------
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
|