Thanks a lot for the reply,
Basically, I like to obtain the same result which is available in the
reference paper of Garnet.
In case of 8x8 mesh network with same configuration of four VCs with having
each 4-flit buffers_per_data_vc (16 flits/port), we could obtain similar
results as the paper till 0.175 injection rate, but if we consider the
injection rate is 0.20, this simulation result (garnet 1.0 and fixed
pipeline router) shows the average latency more than 300cycles. However,
the paper shows about 60 cycles even for 0.30 injection rate.
In case of Garnet 2.0 >> 0.20 injection rate shows 130.76cyc
I also make sure the router stage is equal to 5 by considering the
flexible option (garnet 1.0) and reduce the number of data flits with
block_size_bytes=16, which also shows the average latency as above the 300
clock cycles.
Please let me know if there is any more configuration to be done to obtain
the referenced paper result.
./build/ALPHA_Network_test/gem5.debug configs/example/ruby_network_test.py
--num-cpus=64 --num-dirs=64 --topology=Mesh --mesh-rows=8
--sim-cycles=1 --injectionrate=0.15 --synthetic=0
--garnet-network=fixed
Thanks and best regards,
F. A. Faisal
..
On Thu, May 25, 2017 at 2:47 AM, Krishna, Tushar
wrote:
> Average Network Latency has just been renamed to Average Network Flit
> Latency since Garnet2.0 also reports the Average Network Packet Latency.
>
>
> Please take a look at the gem5 wiki for a detailed description of
> garnet1.0 vs 2.0.
> http://www.gem5.org/Garnet
> http://www.gem5.org/Garnet2.0
>
> I’m not sure what validating one against the other means.
>
> Garnet2.0 uses a 1-cycle router by default so the network latency is not
> expected to match that from the original garnet1.0 paper which assumed a
> 4-cycle router as the default.
> You can mimic a 4-cycle router in garnet2.0 by setting --router-latency=4
> (look at configs/network/Network.py) but it will not match exactly.
>
> Cheers,
> Tushar
>
>
>
> On May 24, 2017, at 6:11 AM, F. A. Faisal wrote:
>
> Dear All,
>
> I wanted to validate the garnet 2.0 with garnet 1.0.
> Hence, I like to get the average network latency or the average latency.
> However, this parameter is missing from the stats.txt file in garnet 2.0.
> And even I compare the average packet latency of garnet 2.0 (21.487374)
> with the garnet 1.0 average latency (40.403994) for 8x8 mesh with same
> configuration as referenced paper of garnet, which is completely different.
>
> Please let me know how to validate garnet 2.0 as of garnet 1.0.
>
> Thanks in advance.
>
> Best regards,
>
> F. A. Faisal
> ___
> 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