[c-nsp] to shape or not to shape

[Arie Vayner (avayner)]

Exactly, and this is why we need shaping for a sub rate link - the
router would not use the different class configuration for the different
traffic classes unless it knows that the link is congested.
If we use a 10Mbps link for a 5Mbps service (or even worst a 1GE link
for a 150Mbps service...) and do not shape, the router would not drop
anything (it thinks it has the line rate all the way to the other side,
so no congestion), and traffic would be tail-dropped (with no class
differentiation) on the SP's ingress policer.

When using a full-rate link (i.e. E1 etc), you do not need to shape it
explicitly, because the router "knows" what is the physical bandwidth.

Arie

-----Original Message-----
From: cisco-nsp-bounces at puck.nether.net
[mailto:cisco-nsp-bounces at puck.nether.net] On Behalf Of Mikael
Abrahamsson
Sent: Sunday, October 10, 2010 00:48
To: Roger Wiklund
Cc: c-nsp
Subject: Re: [c-nsp] to shape or not to shape

On Sat, 9 Oct 2010, Roger Wiklund wrote:

> Yeah that's what I belive also. This whole thing started with a person
> at my work telling me that we should shape a 1984 to 1984 just to
> delay packets instead of tail dropping.

I don't get it. Tail dropping is what you do when the queue is full, 
you're delaying a lot of packets and you don't want to fill the queue
any 
more. Saying "we should delay packets instead of tail dropping" just 
doesn't make any sense to me.

Tail dropping is what you do in a queue/buffer when it's full. It
happens 
whatever type of queue/buffer you're talking about. FIFO is one queue
per 
interface, CBWFQ has multiple. At some time if you try to push enough 
traffic thru, the buffers will fill up and you'll drop packets.

-- 
Mikael Abrahamsson    email: swmike at swm.pp.se
_______________________________________________
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/