From: Filyurin, Yan (yan.filyurin@eds.com)
Date: Fri Jun 22 2007 - 22:03:08 ART
Actually I was never able to find a good document that described and
Odom's book while excellent did not quiet explain it for me to
understand it, so I continued asking around. In the end I saw the
following revelation, after talking to our ANS
I also would like to offer a shameless promotion for Michael Zuo's notes
that includes some details on that as well and goes into it.
The revelation was the different between shape average and shape peak.
In case of shape average, one can do the shaping rate, BC and BE. Often
people don't do BE and the goal of shape average is not to exceed the
rate. So in the course of a TC BC packets pass the shaper and move on.
If less than BC moves through the shaper, there are tokens left and
those will get transmitted in the next TC cycle. So token bucket is BC
+ BE and in the next TC intervals the tokens that were accumulated can
be used if anything needs it. So that way shape average complies with
CIR and only allows sending of BC+ bps if it accumulated credit from the
previous cycle, or cycles. The token bucket start with just BC. You
can configure BE, but that is so you can use bandwidth more efficiently
and what did not get sent in one interval can be send in the next one.
In case of shape peak the bucket starts out with BC + BE, gets filled
with BC every time interval and so if you want to transfer at the very
beginning BC + BE, you can do that, but to keep on doing that, you have
to accumulate credits, so in theory when doing shape peak, you suddenly
get to send up to peak rate, if enough credits got accumulated. In
other words, at every interval you can send up to a peak rate if
bandwidth is available and you have the credits and you already start
out with them in the token bucket. That is very similar to how frame
relay traffic shaping works. So if you are specifying BE and BC under
that map-class you are kind of saying that this PVC can burst up to that
peak rate. So the same formula of rate = CIR * (1 + be/bc) will apply
to frame relay traffic shaping.
Generic traffic shaping as I understand it is very similar to shape
average. Same goes for specifying traffic-rate is frame relay map
class. This part I am still researching.
-----Original Message-----
From: Mike Kraus (mikraus) [mailto:mikraus@cisco.com]
Sent: Wednesday, June 20, 2007 11:38 PM
To: Filyurin, Yan
Subject: RE: Help with concept explanation
Yan, what document helped the light bulb come on for you? I need to
read up on this myself as well.
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
Filyurin, Yan
Sent: Wednesday, June 20, 2007 10:27 PM
To: ccielab@groupstudy.com
Subject: RE: Help with concept explanation
Never mind, I figured it out, after reading up some more on it, but I
must say shape average vs. shape peak is the biggest QOS mystery!
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
Filyurin, Yan
Sent: Wednesday, June 20, 2007 4:19 PM
To: ccielab@groupstudy.com
Subject: Help with concept explanation
I know this comes in the discussions every month, but I think there is a
difference in how Cisco Documentation describes it vs. how some authors
describe it.
After reading a few examples in the practice labs, I saw that if I were
given a frame relay traffic shaping requirement that would be something
like this.
A circuit has access rate of 1024kbps and PVC provisioned at 512kbps and
configure the connection to not go above the CIR, but if enough credits
have been accumulated be able to burst up to the port speed.
Frame Relay traffic shaping is pretty straightforward, assuming TC of
1/8 of a second.
Frame-relay map-class shape
frame-relay cir 512000
frame-relay bc 64000
frame-relay be 64000
There could be numerous alternative solutions either using frame-relay
traffic, rate, generic traffic shaping or class based shaping. And the
one thing, I am not clear on is why in case of Class Based traffic
shaping the configuration involves shape peak.
Class class-default
shape peak 512000 64000 64000
If I understand how shape peak works, the token bucket will effectively
be BC + BE, which would be in case on frame-relay traffic shaping, but
in case of shape peak BC + BE will go into bucket every time interval
and won't it be different from frame relay shaping configuration
Would shape average be more appropriate.
I know everyone is sick of this discussion, as it keeps constantly
appearing and I thought I understood this well, but this seems to go
against it.
Also Cisco documents it differently that Wendell Odom's book.
Yan
This archive was generated by hypermail 2.1.4 : Sun Jul 01 2007 - 17:24:51 ART