[parsec-users] Simulating Swaptions

Iordan Alexandru airjrd at yahoo.com
Thu Apr 16 05:03:57 EDT 2015


Hello
I am running some experiments with several PARSEC benchmarks (Blackscholes, Bodytrack, Fluidanimate, Streamcluster and Swaptions) for core counts ranging from 2 to 32 cores. To account for the non-deterministic behavior of my simulator (Sniper), I run 10 simulations of each experiment and average the result. I also calculate the standard deviation for each set based on execution time to make sure I do not encounter some weird outliers. All is well with all my simulations except 1 set: Swaptions for 32 cores. For the 10 simulations in this configuration the standard deviation is about 15% while all the others are bellow 5%. Is there some sort of limitation for Swaption regarding the core count? Is Swaptions supposed to have large variation in its execution time?
I mention that I compiled all my benchmarks with the gcc-tbb option.
Any ideas would be highly appreciated.
Alexandru
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20150416/800167bd/attachment.html>


More information about the parsec-users mailing list