Soumyaroop, No, it's actually more rudimentary than that, but it works. It's basically a script that takes N checkpoints from single-cpu workloads and aggregates them into another checkpoint with N single-threaded cores so that it looks like a multiprogrammed checkpoint. So when you do a checkpoint restore, you restore from the aggregated checkpoint.
Being this simple has the added side effect of taking up lots of space, unfortunately, because you need a new checkpoint for every benchmark combination, but it was pretty quick, dirty and simple. It may be possible to hack more into the inner guts so that M5 knows to read from multiple checkpoints and fix up things like asns and page table entries at runtime, but I havne't done that. If you are interested, that could be a better solution if you want to run a lot of workloads with a lot of benchmarks in each workload and you're concerned about storage. I hope to push the patch in the next few days, it's not that complicated but it's been in my tree and unused a while and I want to make sure the test spin works and have it cleaned up first. Lisa On Tue, Jan 12, 2010 at 5:23 PM, soumyaroop roy <[email protected]> wrote: > Thanks for the reply, Lisa. > > So, if I understand correctly, the patch that you have enables > simulation of 'n' workloads on 'n' single-threaded cores after > restoring them from 'n' checkpoints, right? > > By when do you plan to push the patch to the dev tree? And, I think I > could perform some testing on it. I have some testing infrastructure > set up. > > regards, > Soumyaroop > > > On Tue, Jan 12, 2010 at 4:48 PM, Lisa Hsu <[email protected]> wrote: > > Soumyaroop, > > > > I have a patch in my tree which does something similar. I am cleaning it > up > > now for public distribution, but even then it won't be totally tested, > > particularly on anything but ALPHA. Can you wait until I push it? You > can > > use it as a reference point to modify for SMT. Currently, it's only good > > for multiprogrammed workloads, but has never been used for > SMT-situations. > > Shoudl be trivial to fix up though. > > > > Lisa > > > > On Sun, Jan 10, 2010 at 4:25 PM, soumyaroop roy <[email protected]> > wrote: > >> > >> Hi, > >> > >> Is it possible to simulate workloads in a multiprocessor system with > >> 'n' cores each with multithreading capability of 'm' threads such that > >> checkpoint restoration points (each checkpoint dump created using a > >> uniprocessor configuration as demonstrated in the ASPLOS tutorial) for > >> each of those workloads can be specified? > >> > >> For e.g., consider a system with 2 cores (each with SMT of 2) running > >> 4 workloads (w0-w3) > >> core0 - [w0 , w1] > >> core1 - [w2 , w3] > >> > >> But those workloads should be restored from separate checkpoints > >> (which may be created using simpoint values). > >> > >> I am working in SE mode. I searched through some past posts on m5 but > >> could not find any discussion that talked about this. > >> > >> regards, > >> Soumyaroop > >> > >> -- > >> Soumyaroop Roy > >> Ph.D. Candidate > >> Department of Computer Science and Engineering > >> University of South Florida, Tampa > >> http://www.csee.usf.edu/~sroy <http://www.csee.usf.edu/%7Esroy> > >> _______________________________________________ > >> 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 > > > > > > -- > Soumyaroop Roy > Ph.D. Candidate > Department of Computer Science and Engineering > University of South Florida, Tampa > http://www.csee.usf.edu/~sroy <http://www.csee.usf.edu/%7Esroy> > _______________________________________________ > 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
