In order to achieve better user level execution in each core, you have to set affinity of each application to a core. Also make sure that you start simulation after each application has started. If you can modify source code of these benchmarks then try to use 'sysv-semaphores' to synchronize these applications ( http://marssandbeyond.blogspot.com/2011/09/creating-barrier-between-processes-with.html )
- Avadh On Thu, Oct 25, 2012 at 12:06 PM, ali ahmadi <[email protected]>wrote: > Hi All, > > I tried to ran some simulation using multi-core. Benchkmarks that I am > using are single thread code. > I ran one of them with 4 cores and I got following stats for user: > > OOO_0_0: cycles 40156294 > OOO_1_1: cycles 19384 > OOO_2_2: cycles 101236 > OOO_2_2: cycles 0 > > > For the Kernel, load distributed among cores in a good way. Is this > because my code is single thread? > How can I fix that without changing code to multi-thread? > > How can I distribute the load in a fine grain way? > > > Thanks in advance for your help and comment. > > > Regards, > Ali Ahmadi > > > > > _______________________________________________ > http://www.marss86.org > Marss86-Devel mailing list > [email protected] > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > >
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
