Hi Tushar,

I tried running a simple 5x4 cpus in Mesh_XY topology using the command:

idris@idris-HP-EliteBook-8460p:~/Desktop/gem5$ build/Garnet_standalone/gem5.opt 
configs/example/garnet_synth_traffic.py --network=garnet2.0 --num-cpus=20 
--num-dirs=20 --topology=Mesh_XY --mesh-rows=5 --sim-cycles=3000 
--injectionrate=0.12 --synthetic=uniform_random --routing-algorithm=1

but received the following memory interleaving error.  
warn: DRAM device capacity (8192 Mbytes) does not match the address range 
assigned (32 Mbytes)
warn: rounding error > tolerance
    1.250000 rounded to 1
warn: DRAM device capacity (8192 Mbytes) does not match the address range 
assigned (32 Mbytes)
fatal: fatal condition intlvBits && intlvMatch >= ULL(1) << intlvBits occurred: 
Match value 16 does not fit in 4 interleaving bits
Memory Usage: 94504 KBytes

Hacking into this creates a chain reaction.

This is the main reason why I find it difficult to create the pyramid topology. 
It seems the simulator can only run a square matrix of (n to power 2) cpus. 

In the decoupling techniques I used, I declared the base layer and the first 
upper hierarchy as follows:

1.      routers = [Router(router_id=i, latency=router_latency)\
            for i in xrange(num_routers)]    # num_routers = 64
        network.routers = routers

2.      hier_routers     = [Router(router_id=i+j+1, latency=router_latency) \
            for j in range(num_h_routers)]   # num_h_routers = 16
        # Update the network routers here
        network.routers         = routers      

Logs in config.ini shows that all routers were created, all internal links and 
external links were also created. But the upper layers router could not wakeup.
This led me to the first mail I wrote: Inquiry on router wakeup in garnet2.0. 
Debug flags cannot see beyond the base layer (64) routers.

That is why I want to replace routers with switches at the upper layers. Do you 
thinks that is okay? or is there something wrong with the technique I deploy?


Kimiya F.

gem5-users mailing list

Reply via email to