Qos: SRR buffer configuration . Do you know what we are doing ?

From: johngibson1541@yahoo.com
Date: Sun Apr 01 2007 - 01:53:31 ART


We have commands that changes queue length like this one,

mls qos srr-queue output buffers

Sure we know when the queue is very long it makes the packet less likely
to be lost. However, we know when a packet is in a very long queue, by the
time it is sent out, it may be useless (voice conversation packets delayed
over a second is useless when it arrives at the destination). So,
we might as well not bother to configure a long queue length. In fact we
should configure a short queue to reduce the switch's workload of taking
care of useless packets. However how short is short enough for
voice conversation? How long is long enough for a ftp file packets ?

The univercd doesn't bother to give a formula about how short a queue
should voice conversation packets go. Because the univercd is just
a configuration guide. What should be configured is up to a designer.

But I just can't accept that fact that we CCIEs don't bother to know.

I think we insult ourselves if we just want to configure the buffer length
after we are told to configure a certain number and we are just good at
looking up the manual to locate the command.

Seriously, if I make the exam, I will say "a voice packet should not
be delayed for more than 20ms in this switch but there should be less than
1% of voice packet lost rate, please configure the shortest possible queue
length accordingly".

How to convert packet lost rate to queue length ? I don't have the equation
in hand.

John



This archive was generated by hypermail 2.1.4 : Tue May 01 2007 - 08:28:34 ART