From: Derick Winkworth (dwinkworth@wi.rr.com)
Date: Fri Feb 23 2007 - 09:43:19 ART
(1) Are you saying that RED acts on each individual flow queue then? So we
could have 16 flows each with at least "minimum threshold" packets in the
queue? Suppose minimum threshold is 20, that means there could be 16x20
packets queued during congestion... wouldn't that result in huge amounts of
delay? Even if there were only two or three conversations and the link is
512k it would get choppy (depending on avg packet size obviously).
I'm thinking there is a second FIFO-like queue to which packets are
"round-robin'ed" in from the flow queues, and RED is performed on the second
queue.
(2) The different WRED categories (the default PREC ordered list), they are
all referring to the same queue I thought. If you change "maximum
threshold" for one, but not the others such that "prec 0's" maximum threshold
is 128, then the router will not tail drop "prec 0" packets until the queue is
128 packets deep. On the other hand, the remaining precedences will still be
tail dropped at the default maximum threshold (40 or whatever)... Wouldn't
all this more-or-less be pointless if WRED was being performed on each flow
queue in fair-queueing?
----- Original Message -----
From: Ash
To: Derick Winkworth
Cc: ccielab@groupstudy.com
Sent: Wednesday, February 21, 2007 8:53 PM
Subject: Re: Fair-queue in conjunction with RED
Hi Derick,
From a technical standpoint, WFQ+RED isnt the same as FRED. Reason is, in
WFQ if you dont have enuff queues allowed on an interface, multiple flows
having the same IPP/DSCP values will be enqueued in the same queue. When the
queue starts to fill up, RED will start dropping packets randomly/equally
regardless of the size of the individual flows.
FRED on the other hand in an above mentioned situation will still consider
the IPP/DSCP as well the flow size. Flows with more packets will be dropped
dropped more agressivley than the flows with less packets.
hope that helps
On 2/21/07, Derick Winkworth <dwinkworth@wi.rr.com > wrote:
All:
Anyone know exactly how RED and FQ work together? They are commonly
placed
together in the default class of a policy-map.
Each flow gets a queue through FQ right? A conversation hash I think its
called. Is RED performed on each individual flow queue? Or is there a
second
queue (like a FIFO queue) after the FQ mechanism?
One idea is that the two of them together is the same thing as Flow based
WRED...
Thoughts?
_______________________________________________________________________
Subscription information may be found at:
http://www.groupstudy.com/list/CCIELab.html
This archive was generated by hypermail 2.1.4 : Thu Mar 01 2007 - 07:38:48 ART