[c-nsp] ingress vs egress queues

[Mack McBride]

For the record not trying to start a holy war.

-----Original Message-----
From: cisco-nsp-bounces at puck.nether.net [mailto:cisco-nsp-bounces at puck.nether.net] On Behalf Of Phil Mayers
Sent: Thursday, March 03, 2011 3:34 AM
To: cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] ingress vs egress queues

On 03/03/2011 03:30 AM, Aaron Riemer wrote:
> Hey guys,
>
>
>
> Can someone please explain to me the difference between ingress and egress
> queues on catalyst switches?

Ingress queues are more or less useless[1], since most Catalyst switches 
have non-blocking (or nearly so) fabric/interconnects.

Egress queues are useful, for all the "normal" reasons.

>
>
>
> I always thought Ingress was traffic coming 'in' to the switch port from a
> device and egress traffic coming out of the port but I don't think this is
> correct?

You were right. Ingress/Egress are from the perspective of the switch.

>
>
>
> Does a packet first go through the ingress queue and then to the egress
> queue before it is 'switched'.

Yes

>
>
>
> I have been trying to find some good documentation to aid my understanding
> in this area. If anyone could help me out with this it would be greatly
> appreciated.

The QoS SRND is useful for background info on Catalyst QoS. Other links 
have been posted to the list fairly recently - have a look at the archives.

Regards,
Phil

[1] If you believe the explanation in the QoS SRND, which I find 
reasonably convincing.
_______________________________________________
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/


--------------- response ----------------------

Ingress queues can be useful/essential on some platforms:
6500, 7600, ASR 9K, CSR, ASR 1K (the last two are technically routers, 
ASR 9K shares a LOT with the first two but is classified as a router)

But the design of other platforms ingress queues aren't very useful:
2960
Mostly because of the backplane design.

Packets on some platforms enter the ingress queue then have some switching before
entering the egress queue structure (6500/7600/9K).  Other platforms use a theoretical egress queue 
on the switching engine, so technically switching and egress queuing happen simultaneously.
This usually involves a complex buffering mechanism with classification and accounting
happening as part of the switching process (ASR 1K being a great example).  And there is usually
a single real output queue on the port itself in this model.

And still other use port group queuing where a group of ports has a shared queue,
So some switching occurs before the data enters the queue and some afterwards.
I am unaware of any platform where egress queuing happens strictly before switching.

I guess to really answer the OP's question we have to ask which Catalyst switch platform?
They don't all have the same queuing model or handle it quite the same way except at a theoretical
level.

Ie. Theoretical high level design

ingress queue -> classify -> mark -> switch -> egress queue

Mack McBride
Network Architect