From: Jon Carmichael (jonc@xxxxxxxxxxx)
Date: Sun Feb 10 2002 - 06:41:41 GMT-3
I can't configure it without the bandwidth command, I tried. It won't let
me put any random detect parameters in the policy-map without putting
bandwidth in first, and that in turn makes it CBWFQ instead of WRED. I
think it's possible that the example on page 134 is wrong, if not wrong then
certainly incomplete. Where the book does not show "bandwidth" as a
parameter, my router is telling me differently.
By turning the min-threshold to 1 packet, and the max-threshold to 2
packets, I should be able to force drops. The book does a fair job of
explaining the formulas to packet-drop probability, but it has some really
lousy examples. According to the book's explanation of
"exponential-weighting-constant" if I set that to a very low number (1 or
2), it would make the average-bandwidth very volatile and force drops at
even the slightest burst of packets.
JONC
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com]On Behalf Of
Kelly M. Knowles
Sent: Sunday, February 10, 2002 1:04 AM
To: Jon Carmichael; ccielab@groupstudy.com
Subject: RE: Playing with WRED
It's been my experience that WRED only drops packets if you have
congestion on the link. If you have available bandwidth it will forward
the packet. Instead of the bandwidth command, try using a small clock
rate on your serial interface to create the congestion.
Kelly
-- Kelly Michael Knowles Network Engineer P & J Computers, Inc. 1 Highland Ave, Albany NY 12205-----Original Message----- From: Jon Carmichael [mailto:jonc@pacbell.net] Sent: Sunday, February 10, 2002 1:43 AM To: ccielab@groupstudy.com Subject: Playing with WRED
I'm having lots of fun here trying to make WRED work. Following the book IP Quality of Service, by Srinivas Vegesna for those of you who might have the book or are interested in making this work. I'm struggling with the relationship between WRED and CBWFQ, and why I'm being forced by the IOS to treat those two as the same, perhaps WRED is a subset of CBWFQ?
I'm working an example similar to his on page 134, but not getting the results I would like. I've done this on a 2500 and a 4500 with the same result, and that is that it keeps thinking that I'm doing CBWFQ, when I want to make WRED drop some packets. I'm setting my minimum-threshold to 2, my maximim threshold to 4 for all precedence levels, where I think I can generate enough tcp packets so force some drops, but the problem is that my interfaces think it's CBWFQ as soon as I put a service-policy on them, when what I thot I wanted is WRED.
Of course RED, is not WRED, and if you do "random-detect" on the interface, you don't get to mess with weights, but a service-policy, to follow the example in the book is coming up as CBWFQ. Even after I put "random-detect" in my policy-map, I am still required to put a "bandwidth" before I can set things like the "exponential-weighting-constant" or the "random-detect precedence..." Of course once I apply the policy-map to the interface it pops up with a queueing strategy of CBWFQ, where I was hoping to see WRED.
The relevant parts of my configuration are below..
* * * class-map non-critical match access-group 101 ! ! policy-map SET-WRED class non-critical bandwidth 8 random-detect random-detect exponential-weighting-constant 3 random-detect precedence 0 2 4 1 random-detect precedence 1 2 4 1 random-detect precedence 2 2 4 1 random-detect precedence 3 2 4 1 random-detect precedence 4 2 4 1 random-detect precedence 5 2 4 1 random-detect precedence 6 2 4 1 random-detect precedence 7 2 4 1 ! * * *
JONC
This archive was generated by hypermail 2.1.4 : Thu Jun 20 2002 - 13:46:18 GMT-3