That is correct. Atomic mode is as fast as it gets. You can still take a checkpoint in atomic, then subsequent atomic runs will be sped up (around an hour or so in my experience). You could also go into the disk image and modify the bbench source itself to execute fewer iterations, the default is 2 so you could make it run only 1. However, I wouldn't recommend this.
On Wed, Mar 21, 2012 at 12:46 PM, Anirudh Sivaraman <[email protected]>wrote: > Thanks for the extremely prompt replies. I am not extremely concerned > about timing number accuracy, but would rather have faster simulation > time. But from what I understand there is no way to reduce the 12 > hours time to anything lesser, because it's already running the > fastest CPU type (in terms of simulation speed). > > Anirudh > > On Wed, Mar 21, 2012 at 12:42 PM, Anthony Gutierrez <[email protected]> > wrote: > > I would suggest taking a checkpoint after the system boots and before > bbench > > launches. This can be inserted in the bbench.rcS script by adding > /sbin/m5 > > checkpoint right before bbench is launched. Then you can restore into O3. > > This should shave about a day off of simulation time since, im my > > experience, booting alone can take over a day in O3 mode. > > > > On Mar 21, 2012 12:36 PM, "Ali Saidi" <[email protected]> wrote: > >> > >> Hi Anirudh, > >> > >> > >> > >> By default gem5 uses the AtomicSimpleCPU. If you would like to get > timing > >> information you'll probably need to use the out-of-order CPU model which > >> will take longer to run. It's an open question how to best sample with a > >> multi-threaded browser type workload. It would be great to know the > answer, > >> but I don't think there is any consensus yet. It is possible to run the > >> workload to completion with a detailed CPU model, it just takes a > couple of > >> days of simulation and that is what some people do at the moment. > >> > >> > >> > >> Thanks, > >> > >> Ali > >> > >> > >> > >> On 21.03.2012 11:31, Anirudh Sivaraman wrote: > >> > >> Hi > >> > >> Running BBench takes me ~ 12 hours. I was wondering if there was > >> anyway to speed that up by sampling (and/or) checkpointing. BBench > >> runs using fs.py and I notice that by default the DriveCPUClass in > >> fs.py is set to AtomicSimpleCPU in the line : DriveCPUClass = > >> AtomicSimpleCPU > >> Does this mean that be default BBench runs the AtomicSimpleCPU model > >> that is purely functional ? This would, in turn, imply that no kind of > >> sampling would improve performance? Am I understanding things > >> correctly here ? > >> > >> Anirudh > >> _______________________________________________ > >> gem5-users mailing list > >> [email protected] > >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > >> > >> > >> > >> > >> > >> _______________________________________________ > >> gem5-users mailing list > >> [email protected] > >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > > > > _______________________________________________ > > gem5-users mailing list > > [email protected] > > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > _______________________________________________ > gem5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > -- > Thanks, > Tony > > <http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users> > > <http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users> >
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
