From: ccie2be (ccie2be@nyc.rr.com)
Date: Sun Aug 28 2005 - 18:25:44 GMT-3
Stephen,
I agree with everything just said but I think the first decision - to use
shaping or policing - comes down to this.
What should happen to packets that exceed the specified rate?
If the packets should be dropped, then use some form of policing.
If the packets should be delayed, then use some form of shaping.
If you can't tell from the wording what should happen to the packets, ask
the proctor.
In general, for data traffic that isn't delay sensitive eg voip, shaping is
used for the outbound direction and policing is used on the inbound
direction.
Once you've decided to shape or police, then you need to figure out which
way to implement the policy. For that, I recommend reading Wendall Odom's
book, DQoS, 2nd edition.
HTH, Tim
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
gladston@br.ibm.com
Sent: Sunday, August 28, 2005 4:09 PM
To: ccielab@groupstudy.com
Subject: Re: QOS important question (GTS and 2rate policer) and
interpretation
Hi Stefan,
I will try this. Hope you get more feedback.
=====================
Traffic should be limited to 16 Mbps for example. Which solution should be
choosed.
=====================
First thing is to know if there is some word leading to police or shaping.
For police, the options are CAR or CB policing (if the task is on 3550,
there is no CAR option).
For shaping, the options are GTS, CB shaping and FRTS (considering DTS is
not an option for the CCIE lab).
To decide when to use CAR or CB policing, look out for words on the task
that leads to one solution.
These compares CAR and CB policing:
Classification:
CB policing can perform classification based on anything that CAR does and
more:
CoS, Destination MAC, input interface, RTP, NBAR.
These options are not available for CAR (well, it is possible to associate
an ACL with CAR and choose RTP ports)
Actions:
CB policing can perform any actions that CAR does (well, it does not have
'continue' keyword) and more:
set DE bit, set CLP bit
CAR does not do these specific actions
To decide when to use GTS, CB shaping or FRTS, look out for words on the
task that leads to one solution.
These compares GTS, CB shaping and FRTS (just the more important, on my
point of view):
FRTS has the most flexible support for queues. It supports fifo, wfq, cbwfq,
llq, pq and cq.
GTS only supports wfq.
All of them supports adaptive shaping.
All of them can be enabled on subinterfaces. (it is a good exercise to
implement cbwfq under subinterface, serial and ethernet).
Only FRTS supports FRF-12.
But memorize a 'comparisson of traffic table' is not good as implementing
each tool and comparing for yourself. If possible, implement each of them on
a big scenario at the same time, so you can go to router A, where you
configured GTS, and compare to router B, where you configured CB shaping.
======================
What is the difference in action of GTS and two rate policer??? Sure this
are different features?? But do they differ when the task is to limit the
traffic to some Mbps.
=====================
On is a shaper and the other policer. One does queue, other just drop.
=======================
HOw differ the following commands??
police cir 16000
and shape average 250000.
==========================
In fact you are not using two rate policer with the command 'police cir
16000'. If you just type 'police 16000', IOS complement with 'police cir
16000'.
You use two rate policer when you specify two rates, for example, 'police
cir 32000 pir 64000'. This would be a complete example of using it:
class-map match-all telnet-ping
!
policy-map two-rate-three-color-rfc2698
class telnet-ping
police cir 32000 pir 64000
conform-action transmit
exceed-action set-prec-transmit 1
violate-action drop
!
interface Serial0/2
bandwidth 128
service-policy output two-rate-three-color-rfc2698
Traffic up to 32kbps conform and are transmitted with original labels.
Traffic from 32kbps up to 64kpbs are marked with precedence 1 and
transmitted.
Traffic above 64kbps are dropped.
Two rate can be a litte confusing. It is necessary to remember how a single
rate model works. On single rate model, the Be bucket receive tokens only
when the Bc bucket fills out. That is, if there is no traffic, bucket Bc
will get full, extra/spilled tokens goes to bucket Be.
On two rate model, the bucket Be has its own 'replenish system', independent
of Bc bucket.
The basic here is that to transmit above Bc it is necessary a period of no
traffic to fill the Be bucket, when using single rate.
Remember, dual token with one thing, dual rate is another. Every time you
specify Be, there is two buckets. Every time you specify cir and pir, there
is two buckets and two indendent rates for each token.
Regards,
This archive was generated by hypermail 2.1.4 : Sun Sep 04 2005 - 17:01:20 GMT-3