Great! Thanks! Jun
On Sat, Sep 1, 2012 at 1:27 AM, Tushar Krishna <tus...@csail.mit.edu> wrote: > Hi Jun, > You are right. It will work with hammer. The LD, IFETCH and ST will not > make any physical sense as they are generated randomly, but that should be > good to stress the protocol and your topology. > > - Tushar > > On Sep 1, 2012, at 1:25 AM, Jun Pang wrote: > > Hi Tushar, > > Thanks for your reply. > > I feel like that the NetworkTest should also work with hammer protocol. As > I understand it, the network_test.cc will generate three events for cache, > which are LD, IFETCH and ST. Hammer also has those events, but hammer's > directory will not drop those messages right away, so there will be some > real transitions for hammer. I actually tried a few experiments with > hammer, and got some results I expected. I wonder if that makes sense. > Thanks. > > Jun > > On Sat, Sep 1, 2012 at 12:07 AM, Tushar Krishna <tus...@csail.mit.edu>wrote: > >> Hi Jun, >> I am not very familiar with the stats generated from simple network. >> You'll have to dig into the code to see what those histograms mean. >> >> The network tester is supposed to be run with the NetworkTest coherence >> protocol, not MI_example, or MOESI_hammer. The network tester allows you to >> inject synthetic traffic like uniform random, bit complement etc which has >> nothing to do with any protocol. >> You could start with these to test your topology. >> If you want to test out MOESI_hammer, you will have to use the ruby >> random tester. >> >> - Tushar >> >> >> On Aug 30, 2012, at 9:38 PM, Jun Pang wrote: >> >> Hi Tushar, >> >> I found "Message Delayed Cycles" histogram information in ruby.stats, I >> wonder if the "average" is the latency for each messages. Also, I want to >> run some synthetic benchmarks to evaluate my new topology with hammer >> protocol in ruby simple network. I read the wiki page and know that >> network tester has some synthetic benchmarks, but is supposed to be used >> with the dummy protocol MI_example, so I start with that protocol. However, >> in Message Delayed Cycle information part, I find that most messages have 0 >> delay. I wonder why. It is also like this when I switch to hammer >> protocol. And for two runs with the same maxpackets injected but very >> different link latency, the run time is almost the same. I wonder what is >> the correct way to use the tester with synthetic benchmarks. If this tester >> is not actual or does not work well with hammer, are there any other >> recommendation for a few simple benchmarks to run? Thanks. >> >> Jun >> >> On Wed, Aug 29, 2012 at 3:38 PM, Tushar Krishna <tus...@csail.mit.edu>wrote: >> >>> Yes simple network does not output network latency stats, only garnet >>> does. >>> In Garnet the start and end are easy to identify since they are network >>> interfaces, while all intermediate nodes are routers. In Simple Network on >>> the other hand all nodes are identical … Messages move between >>> MessageBuffers at different nodes. A node doesn't easily know if the >>> message came from another router or the cache controller. You will probably >>> need some way to identify that, after which calculating network latency >>> should be straight forward. >>> >>> - Tushar >>> >>> >>> On Aug 29, 2012, at 3:31 PM, Jun Pang wrote: >>> >>> > Hi all, >>> > >>> > I didn't see any network latency summary in ruby.stat when running >>> ruby simple network with some protocol. I wonder if that is implemented or >>> I need to do that by myself. If so, what's the simple way to do it? >>> > >>> > Thanks, >>> > >>> > Jun >>> > _______________________________________________ >>> > gem5-users mailing list >>> > gem5-users@gem5.org >>> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >>> >>> _______________________________________________ >>> gem5-users mailing list >>> gem5-users@gem5.org >>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >>> >>> >> _______________________________________________ >> gem5-users mailing list >> gem5-users@gem5.org >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> >> >> >> _______________________________________________ >> gem5-users mailing list >> gem5-users@gem5.org >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> > > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
