From: Bob Sinclair (bsinclair@netmasterclass.net)
Date: Thu Dec 16 2004 - 18:37:26 GMT-3
Tim,
Thanks for the reference regarding Odom's new book! Have you seen Szigeti's
new book, End-to-End QoS Network Design, ISBN: 1-58705-176-1? It has less
depth on individual tools, but provides needed, up-to-date context. He
compares the single-rate and two-rate three-color markers.
The practical advantage to the two-bucket, 3-color markers is that you have
three actions available: conform, exceed and violate. The Bc bucket is the
dividing line between conform and exceed (in-profile, out-of-profile). The Be
bucket is the dividing line between exceed and violate
(a-little-out-of-profile vs. way-out-of-profile). You might transmit traffic
that conforms, mark down traffic that exceeds and drop traffic that violates.
So, have three colors (two buckets) allows you to break traffic down more
finely.
Once you decide you want three colors (conform, exceed, violate), you can then
choose between the singe-rate and two-rate methods of filling the second
bucket (Be). In the single-rate method tokens can only go into the Be bucket
if they have been saved (you were below CIR for a while). If one is paid $10
per week one can burst to $15 only by saving. In the two-rate method, tokens
are placed into the second bucket independent of saving from CIR. If you have
a job paying $10 per week, and a second job paying $5 per week, these are two
independent rates, and one can burst to $15 per week without having to save.
Szigeti says the latter method leads to more predictable bursts.
As regards the utility of the shape peak command, I am with you, and do not
see it.
HTH,
Bob Sinclair
CCIE #10427, CCSI 30427, CISSP
www.netmasterclass.net
----- Original Message -----
From: ccie2be
To: Bob Sinclair
Cc: Group Study
Sent: Thursday, December 16, 2004 3:35 PM
Subject: Re: Shape peak and police pir
Hi Bob,
Yes, They changed the name of the book from DQoS to just QOS. ISBN
1-58720-124-0.
Getting back to peak rate, I might be a bit thick skulled on this, but
what's the application of enforcing a peak rate separate from the cir? I feel
I'm missing the big picture with this.
For example, let's say you're pre-sales technical support and you're talking
to the buyer of ISP services for a mid-size firm and this buyer, as luck
should have it, doesn't know squat about traffic engineering or bc, be or Tc.
But, he does know that his firm likes to sell lots of widgets over the
internet and that if visitors to his web site find that response time is too
slow, they leave his web site and buy from someone else. He also knows that
his web site (which he created mostly by himself) has some nifty pages that
contain some large graphic files.
Also, occasionally, his firm runs specials which generate lots of traffic to
his site, but as you would expect, he also wants to minimize his monthly web
traffic expense while insuring his customers have a good web experience.
How would you help him decide whether it would be better for him to sign up
for cir with be service or your combo cir and pir service?
Also, do you have any comments on the shape cir versus shape peak issue?
TIA, Tim
----- Original Message -----
From: Bob Sinclair
To: ccie2be
Sent: Thursday, December 16, 2004 2:52 PM
Subject: Re: Shape peak and police pir
Hi Tim,
Are you referring a new edition of Cisco DQOS by Wendell Odom? I don't
see one at ciscopress.com. Could you post the ISBN?
Re policing with CIR/PIR: If you want a three-color marker (conform,
exceed, violate), then you can you can use either method (single rate or
two-rate). The RFCs (2697 and 2698) say that the single-rate is most useful
when it is the LENGTH of a burst, not its peak rate that determines service
eligibility. The two-rate is most useful when peak rate needs to be enforced
separately from a committed rate. The two-rate method allows for a sustained
excess rate, and seems easier to configure, to me, but it really depends on
the traffic profile you want to enforce.
HTH,
Bob Sinclair
CCIE #10427, CCSI 30427, CISSP
www.netmasterclass.net
----- Original Message -----
From: "ccie2be" <ccie2be@nyc.rr.com>
To: "Group Study" <ccielab@groupstudy.com>
Sent: Thursday, December 16, 2004 2:15 PM
Subject: Shape peak and police pir
> Hi guys,
>
> I've been going through the 2nd Edition of Cisco QOS and overall it's a
very
> good book. It fills some of the gaps that weren't covered in the first
> edition and is as you'd expect more up-to-date.
>
> I just finished going over shaping and policing for the umpteenth time,
but
> I'm still a bit fuzzy on the above commands.
>
> I understand the syntax and mechanics of the above commands, but I don't
get
> the point of these commands.
>
> What I mean is why use shape peak instead of shape average with
different
> values? With shape peak, (bc + be) are aent every Tc. If bc + be = X,
why
> not just use shape average X instead? What's the difference?
>
>
> For police pir, I have similar question: what's the practical benefit
of
> having 2 rates? Can't the same result be achieved, for example, by an
ISP, by
> charging one rate for packets (or bytes) that conform and and higher
rate for
> packets that exceed?
>
> Thanks, Tim
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html
This archive was generated by hypermail 2.1.4 : Mon Jan 03 2005 - 10:31:27 GMT-3