It's not impossible, but it's very challenging. You'd have to figure out which data the models are sharing, and use messages to keep that data in sync. If it's a significant amount of data (like you're trying to model different processors within one shared-memory machine) it gets very hard and would likely not perform well.
Also, if you have multiple machines, often the best and most efficient use of them is to run multiple independent simulation jobs concurrently (and that requires no programming overhead beyond some job scripts). I'd say there's enough work and challenge in getting things to run well in a multi-threaded single-machine environment that it's premature to worry about going beyond that. Steve On Tue, Oct 18, 2011 at 4:36 PM, Digant <[email protected]> wrote: > Hi Steve, > I was planning to work on parallel M5. > In addition to making it multi-threaded, I was wandering, if we can make it > run on multiple machines. > Can we do that ? e.g. Master on 1 machine and other slaves on different > machines communicating with > master over socket (slaves will run few models with their eventQ) ? > > Regards, > Digant Desai. > _______________________________________________ > gem5-dev mailing list > [email protected] > http://m5sim.org/mailman/listinfo/gem5-dev > _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
