[c-nsp] ingress vs egress queues

[Mack McBride]

-----Original Message-----
From: cisco-nsp-bounces at puck.nether.net [mailto:cisco-nsp-bounces at puck.nether.net] On Behalf Of Phil Mayers
Sent: Friday, March 04, 2011 5:29 AM
To: cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] ingress vs egress queues

On 04/03/11 10:59, Nick Hilliard wrote:
> On 04/03/2011 09:25, Phil Mayers wrote:
>> Genuinely curious to hear real-world use-cases for the ingress
>> queueing on
>> 6500/7600.
>
> Ingress queueing is used extensively for cut-thru fabrics.

Interesting, but not really what I had in mind - 6500/7600 are not cut-thru.

Now I'm curious though - how on earth can you queue on a cut-thru 
switch? If you queue, you're storing the packet and are no longer cut-thru?

Do you have a pointer to any docs on this?

I'm guessing that even a cut-thru switch has to receive "enough" of the 
header to switch/route the packet, and apply whatever QoS classification 
so it's sort of a "store & forward lite"; then once the output port is 
"known", if a packet is not currently being transmitted you jump to 
cut-thru mode. If a packet is being transmitted you have to queue, but 
how is that different to egress queueing?

Interesting...
_______________________________________________
cisco-nsp mailing list  cisco-nsp at puck.nether.net
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

------ reply ------
To answer the question about the 6500/7600:

Backplane congestion will cause the ingress queue to fill and even get
Queue drops often significant.  This can happen at the 16G level on these
platforms.  This is 16G to any egress fabric channel which generally covers half a card.
The fabric channel is 20G and like most Ethernet you start seeing congestion at
The 80% level due to micro bursts.  I have seen a number of companies run into this
problem.  Particularly on the 6708 blades where the port grouping/order is not what people
think it is so overload a fabric channel.

Mack McBride
Network Architect.