[parsec-users] Fw: Running PARSEC benchmarks
cbienia at CS.Princeton.EDU
Thu Jan 14 10:50:45 EST 2010
The only way I can come up with to test how many kernels you have executed is by inserting magic instructions or something like that into the benchmark program. Then again, you’d have to run the whole program anyway to complete all kernels, so why bother?
The slow execution speed of detailed simulators is a real issue, but there is nothing you can do about it as a benchmark user. We need more innovation in simulation technology. Sampling might be a way out, but most simulators don’t support it (yet).
From: parsec-users-bounces at lists.cs.princeton.edu [mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of Muhammad abid Mughal
Sent: Thursday, January 14, 2010 10:37 AM
To: Parsec group
Subject: [parsec-users] Fw: Running PARSEC benchmarks
Thank you very much and thanks for your guidence. How will i come to know that all kernels are running?
Actually i am using GEMS and its too slow.If i run each core for 1 Billion instruction then its gonna be take seven or more days.
----- Forwarded Message ----
From: Christian Bienia <cbienia at CS.Princeton.EDU>
To: Muhammad abid Mughal <mabidm_pieas at yahoo.com>
Sent: Thursday, January 14, 2010 22:42:52
Subject: RE: Running PARSEC benchmarks
It’s not a sane methodology. Most PARSEC workloads have multiple kernels and unless you run all of them you won’t get the representative program behavior. If you want to run only a subset of the instructions you should use sampling, some simulators such as Flexus offer sampling by default.
It would make my life easier if you send your questions to the parsec-users mailing list because I usually get the same ones on and on.
From: Muhammad abid Mughal [mailto:mabidm_pieas at yahoo.com]
Sent: Thursday, January 14, 2010 8:07 AM
To: Christian Bienia
Subject: Running PARSEC benchmarks
Hope you doing good. Am i bothering you by directing questions directly to your email?.
My question is that:
I am using 8-Core CMP with Out-Of-Order Superscalar processor.
I am running PARSEC benchmarks for a total of 2 Billion inst . I use first One billion inst to warm caches and 2nd billion inst to perform detailed simulation.
I am using "simsmall" data set.
Is this methodology is correct to measure performance of PARSEC benchmarks?
Thanks in advance
Get your preferred Email name! <http://sg.rd.yahoo.com/aa/mail/domainchoice/mail/signature/*http:/mail.promotions.yahoo.com/newdomains/aa/>
Now you can @ymail.com and @rocketmail.com.
New <http://sg.rd.yahoo.com/aa/mail/domainchoice/mail/signature/*http:/mail.promotions.yahoo.com/newdomains/aa/> Email addresses available on Yahoo!
Get the Email name you've always wanted on the new @ymail and @rocketmail.
Hurry before someone else does!
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the parsec-users