The mesh in garnet is deadlock free since it uses xy routing (if u use mesh_xy topology) which eliminates cyclic dependences. And unlike Torus, u cannot don't have a ring in each dimension so there cannot be a cyclic dependence there either. If you are seeing a deadlock then something else is wrong.
Cheers, Tushar On Jul 28, 2017, at 11:28 PM, F. A. Faisal <dipu.7...@gmail.com<mailto:dipu.7...@gmail.com>> wrote: Dear Professor, Thanks for the reply. However, I think there is no deadlock prevention mechanism is also implemented in garnet. Such case, a simple mesh network can also create the deadlocks with the side by side nodes. It is required to make the Deadlock prevention like- hold and wait. To do this we definitely need to know the out VC condition for next router/node and check his status. If that node is free for next outport with correct VC number, only then We can send the packet to the outport of the current Node. However, I think garnet does not have this checking on the port status of the next node and its outport VC. This is a very important issue for garnet. Please help me. Thanks and regards, F. A. Faisal On Fri, Jul 28, 2017 at 11:28 PM, Krishna, Tushar <tus...@ece.gatech.edu<mailto:tus...@ece.gatech.edu>> wrote: Hi Faisal, I would try testing the design with one vnet (--inj-vnet=0). And within each vnet you need to vc partitioning scheme - you basically need to ensure that there is no potential cyclic dependence between the VCs. I'll reply to your query in your earlier email next week. Cheers, Tushar > On Jul 28, 2017, at 6:57 PM, F. A. Faisal > <dipu.7...@gmail.com<mailto:dipu.7...@gmail.com>> wrote: > > Dear All, > > I found that using high simulation cycle and large network like- 256 nodes > the flit injection rate for different Vnets are different. Hence, networks > like Torus are becoming deadlocked very soon, even I implement the VC > partitioning. > > Now, the question is can I use the different VC of Different Vnets for > reducing this conjection > or can I distribute the flits evenly between all the vnets to make sure the > network is not in deadlock. > > I will appreciate your earliest reply. > > Thanks and best regards, > > F. A. Faisal > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org<mailto:gem5-users@gem5.org> > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users _______________________________________________ gem5-users mailing list gem5-users@gem5.org<mailto:gem5-users@gem5.org> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users _______________________________________________ gem5-users mailing list gem5-users@gem5.org<mailto: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
