Hi Yinglong, It certainly is possible, however the ease of measurement and effort required depend on a few things. Are you planning on modifying the linux scheduler or running your own scheduler? If the prior then everything should work immediately, although it will be more difficult to modify and capturing statistics on all the processors will be more difficult as well. If you plan to use your own scheduler within M5, no one has done that, however someone recently got pthreads (where the number of threads < number of processors) working in M5 syscall emulation mode. I think their patches should make it into the repository soon.
Ali On Jan 27, 2009, at 2:40 AM, Yinglong Xia wrote: > Hi All, > > I am new to the M5. I am wondering if there are any algorithm people > using M5 for application level performance evaluation. I am working on > application level schedulers on multicore processor. In order to > understand the scalability of the scheduler, I intend to execute our > Pthreads code on a virtual processor simulated by M5, where we can > assign the number of cores in the processor (saying, 32 or 64). I am > wondering if the above idea is feasible on M5 and if it is possible to > accurately measure the execution time when various number of cores > (concurrent threads) are used in the simulated processor. Any comments > are highly appreciated. > > Best, > Yinglong. > _______________________________________________ > m5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
