RE: CBWFQ Congestion Allocations

Matt,
Consider this a crash course.
When you use the "Priority" command under a class you have moved into what
is known as LLQ (Low latency queue). Without you specifying this command
you are basically configuring "CBWFQ".
I will be dissecting your examples as much as I can for this explanation.
Remember "WHEN THERE IS NO CONGESTION THERE IS NO QUEUING".

class TEST1
priority percent 15

(This is a way of policing and guaranteeing bandwidth within a class. Take
it as saying this classs will receive the first and only 15 percent of
available bandwidth if congestion happens. This means it will be cut off
bandwidth usage as soon as it gets the 15%)

class TEST2
bandwidth percent 30

(This means this class will get a minimum of 30% of available bandwidth.
The key word here is "Minimum". This basically means this class will get
more if it is available)

class TEST3
bandwidth percent 30

(This means this class will get a minimum of 30% of available bandwidth.
The key word here is "Minimum". This basically means this class will get
more if it is available)

class class-default
fair-queue
random-detect

(The "class default" is what matches any other traffic not defined by you.
And it is there by default. When you say "fair-queue" this is going to
listen to the lower talkers over the higher talkers. Say ftp and SQL are
used for explaining this. Assume SQL traffic is coming to the router more
than FTP. Because you have fair queue configured instead of the default
"RED" (tail drop) under here, for every time the FTP decides to send
anything the router will listen and attend to it because it initially was
granting more bandwidth to SQL because SQL was the higher talker (I hope you
understand what am getting at.) Maybe someone else can throw more light on
this if need be.

In addition to you using "Fair-queue", you have also told the router to use
"fair-queue" on the basis of "IP-precedence" (layer two classification).
What this means is "I the router will listen to lower talkers but still use
0-7 classification to do this. "I would personally use dscp here instead of
the default IP-Precedence". You ask why, I have more possibilities for
marking 0-63, dscp is also a layer 3 marking so I wouldn't worry about
losing my markings when the traffic leaves my hop (router).

Now CISCO is known for telling us about best practices. But they still give
us the gun to shot ourselves in the foot. That is the way I see
"max-reserved-bandwidth". On an interface by default 25% is reserved. This
is used for lots of other stuffs the router does i.e routing protocols,
communication between devices etc.
So you see you technically cannot use more than 75% of an interface. But
guess what if you decide to shoot yourself in the foot you can go under the
interface and use the command "max-reserved-bandwidth" to change the default
25% to whatever you want and end up starving other critical traffics.

Hope my ranting helped a little.

-----Original Message-----
From: nobody_at_groupstudy.com [mailto:nobody_at_groupstudy.com] On Behalf Of Matt
Bentley
Sent: Tuesday, September 21, 2010 10:33 AM
To: Cisco certification
Subject: CBWFQ Congestion Allocations

Hi GS:

Question about the exact behaviors of CBWFQ.

Say you have this policy-map:

class TEST1
priority percent 15
class TEST2
bandwidth percent 30
class TEST3
bandwidth percent 30
class class-default
fair-queue
random-detect

So I understand the following:

TEST1 no congestion -> EF treatment for up to 15%, Can send up to line rate
with normal treatment (assuming none of the other classes are sending)

TEST1 congestion -> LLQ treatment for up to 15%, policed for greater than
15%

TEST2 no congestion -> Can Send Up To Line Rate (assuming none of the other
classes are sending)

TEST2 congestion -> AF treatment for up to 30%, excess ????

TEST3 no congestion -> Can Send Up To Line Rate (assuming none of the other
classes are sending)

TEST3 congestion -> AF treatment for up to 30%, excess ????

For TEST2/TEST3 congestion scenario, I see this statement from Cisco

http://www.cisco.com/en/US/tech/tk543/tk757/technologies_tech_note09186a0080
103eae.shtml

If class-default does not need it, the unused bandwidth is available for use
by class TEST1 and class TEST2 (they use different names for their classes).
If both classes need the bandwidth, they share it in proportion to the
configured rates. In this configuration, the sharing ratio is 30:30 or 1:1

Instead of trying to word my questions, if someone could elaborate the
allocations in the following scenarios, it would be immensely helpful.

1) Scenario#1
TEST1 is sending traffic at exactly at 100% of line rate
TEST2 is sending traffic at exactly at 100% of line rate
TEST3 is sending traffic at exactly at 100% of line rate class-default is
sending traffic at exactly 100% of line rate

How is the bandwidth going to be allocated between them?

2) Scenario#2
TEST1 is sending traffic at exactly at 15% of line rate
TEST2 is sending traffic at exactly at 42.5% of line rate
TEST3 is sending traffic at exactly at 42.5% of line rate class-default is
sending traffic at exactly 50% of line rate

How is the bandwidth going to be allocated between them?

Also, I don't think the max-reserved-bandwidth assists in determining how
much bandwidth the default class gets when congestion occurs, but not sure?

Thanks very much,

Blogs and organic groups at http://www.ccie.net
Received on Tue Sep 21 2010 - 15:38:59 ART