Re: [Gems-users] about the network link bandwidth

[Mike Marty <mikem@xxxxxxxxxxx>]

> I have some questions about the network bandwidth.
> I notice that in ruby default configuration, there is a variable
>
> g_endpoint_bandwidth:10000
> what does this value mean?
>

The bandwidth for each link is multiplied by this value.  The unit for
bandwidth is "thousandths of a byte".  Thus if the link specifies a
multiplier of "10" and the g_endpoint_bandwidth is 10000, then that link
can deliver 100000 thousandths of a byte per cycle...or 100 bytes.

> and when I dump the ruby stat file, in the network stats
> section:  it shows the links_utilized_percent_swich_X_link_X:
> the first value gets from "throttle_ptr->getUtilization()" function,
> what does this value mean?
>

The utilization of the link

All of the networking code is in $GEMS/ruby/simple/network


> Finally, in the slicc files (I am looking at
> "MSI_MOESI_CMP" architecture), I did not find how the
> NETWORK_LINK_LATENCY value is applied to the latency of
> L2->dir requests when they are put into
> "DirRequestIntraChipL2Network_out" queue,
> where can I find the code that dealing with adding latencies
> to requests according to the network topology.
>

NETWORK_LINK_LATENCY is used by some of the auto-generated networks in
$GEMS/ruby/network/simple/Topology.C