There is nothing in Garnet that restricts the no. of cores/nodes. If
you are having more than 32 cores/L2 caches/memories, the duplication
of packets happen due to a optimization in Set.h in ruby. You can
correct that. Take a look at https://lists.cs.wisc.edu/archive/gems-users/2008-May/msg00013.shtml
Cheers,
Niket
On Nov 18, 2009, at 9:56 AM, Jianchen Hu wrote:
Hi,
I use Garnet network tester to run with random traffic. And when I
try a large network (128-node), the network interface will generate
multiple transaction for the same message. I use MOESI_SMP_directory
protocol and I think all the messages should be unicast, like the
case of 32-node or 64-node network.
But in a 128-node network with CROSSBAR topology, the NI will
generate 2 transactions for the same message to the same
destination. (dest_nodes.size() == 2) So I got double flit count.
Is it a bug in Garnet or I miss something ?
Thanks.
Jianchen Hu
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/
" to your search.
|