[c-nsp] Line-rate on catalyst swiches

[Peter Rathlev]

On Wed, 2013-07-24 at 12:49 +1000, Shanawaz Batcha wrote:
> Lets assume I have a WS-2960G-48 model switch. The advertised
> backplane capacity for this platform is 32G. That is 32Gbps for 48
> ports, so I assume straightaway that its a blocking (as in
> not-non-blocking) switch

Correct. All the C2k and C3k switches are (blocking) store-and-forward
switches.

> 1) Does traffic within the ASIC hit the backplane and count towards
> the backplane 32 gbps capacity? Does say traffic from port Gi0/29
> to Gi0/30 (both in Asic 8) reach the backplane or simply switched
> locally at ASIC level?

AFAIK the ASICs perform forwarding decisions themselves and will not
send traffic destined for a local port towards the switch fabric.

> 2) If the answer to the above is yes, and assuming all ports are
> transmitting at the same time at 1 Gbps,  then presumably we can
> call this a line-rate switch provided the right connections are in
> the right places? if the answer is no, and assuming all ports are
> transmitting at the same time at 1 Gbps, then there will be about
> ~16 gbps worth of drops?

Yes on both counts. Even with locally switched traffic you can easily
imagine a scenario where all traffic has to cross the switch fabric.

The 2960S has no ingress QoS but I don't know about "plain" 2960.
Without ingress QoS you can't work around an overloaded backplane with
e.g. priority queueing. If you need line-rate the 2960 (non-S) probably
doesn't fit your needs.

Take a look at BRKARC-3437 from 2011. It touches on 2960 a bit though it
does not answer your specific question in any exact way.

-- 
Peter