The IP over ATM Mailing List Archive by date

[bansal@ccrl.nj.nec.com (Vivek Bansal)]

> Vivek,
> 
> An FYI-
> 
> Please note that if this is done badly, it can be much worse.  If the
> egress to a slow link backs up to the ingress of a fast link that is
> also serving traffic to the egress of fast uncongested links, it can
> potentially starve the ingress buffering.  This can cause loss on the
> fast link to fast link paths due to congestion on the slow link.
> 
> Router vendors have managed to make this mistake and some still
> haven't entirely corrected it.  I'm hoping switch vendors will not
> have to make the same mistakes.
>
> Curtis

Curtis,

In most ATM switch designs that I have seen, the primary criteria of designing
is always no HOL blocking, so for the case that you mention that should not
happen b'coz if it is a input/output buffered switch, then you better have 
separate logical queues at the input port for each output port. 

You are right some switch/router vendors do not take this into account.

Vivek.