QOS PROBLEM

Hello

I have problems with a Cisco 3640 and QOS Service Policy

Infra :

LAN -> ROUTER -> SWITCH -> INTERNET

I create ACL and put them in a policy map with a priority of 1Mb/s

One client is in a class , second client is'nt in any class-map

When i simulate a congestion - rate limit of internet on the SW
The class of the 1st client (with prioriy of 1Mb) isn't reached
and the default class isn't falling to give bandwidth for the 1st client..

What can i do to give garanted bp to clients...

Cisco has IOS 12.3.
I enable IP CEF but no results after that.

First client :

class xxx
priority 1000


IF
!
interface FastEthernet3/0
ip address xxx
service-policy output yyy
load-interval 60
duplex auto
speed 100
!
ip http server
ip classless
ip cef
ip cef accounting non-recursive



Thanks to help me

Nico

 

Anyone can help me ??

 

It would be easier if you send the whole config..

 

what's config do you have need
i have include class map config & iface config ..

 

Ok, please repost it, and explain exactly what do you want to do.

 

I want to prioritize certain types of traffic
and have garanteed bandwidth . I have a 10Mb link
and if link is congestioned i want clients have
his garanteed bandwidth. so here is config.

policy-map servers
class client
priority 1000 (1mbits garanteed)

IF
interface FastEthernet3/0
ip address xxx
service-policy output servers
duplex auto
speed 100
!
ip http server
ip classless
ip cef
ip cef accounting non-recursive

acl are created and when i show policy map
there is traffic on class default and traffic on client
class but when the link is congested , client doesn't
reach his priority by dropping bandwidth on default class.

Nico

 

seems you are missing:

class-map match-all client
match ... (something)

You said that you have 10 Mb link but you have speed set to 100 Mb. I
aslo thing that you need to specify bandwith with 'bandwidth' command
(it makes sense with 'percent' command in policy-map) so you could put
it anyway.

One more thing... This will just reserve bandwidth in OUTGOING from
interface, not incoming. How did you measure that it does not work?

Can you provide us a 'sh policy-map int fa3/0' ?

 

One more thing... This will just reserve bandwidth in OUTGOING from
interface, not incoming. How did you measure that it does not work?

I limit the link to 2Mb and class default has 1.5
and client has 500k (not 1mb garanted).

Class-map: client (match-any)
30478215 packets, 30661485625 bytes
1 minute offered rate 1438000 bps, drop rate 0 bps
Match: access-group 110
30478211 packets, 30661481553 bytes
1 minute rate 1438000 bps
Queueing
Output Queue: Conversation 265
Bandwidth 1000 (kbps) Max Threshold 64 (packets)
(pkts matched/bytes matched) 2292/2185982
(depth/total drops/no-buffer drops) 0/0/0

Class-map: class-default (match-any)
286422400 packets, 77094054526 bytes
1 minute offered rate 2199000 bps, drop rate 0 bps
Match: any

 

This looks fine. Let's try something else. Try to specify bandwidth in
your client class but without LLQ, just plain bandwidth. Also try to
create class-default and add remaining bandwidth with 'remaining'
keyword. Also, it would be good to add WRED to class-default.

Which IOS do you use?

 

This looks fine. Let's try something else. Try to specify bandwidth in
your client class but without LLQ, just plain bandwidth.

Also try to
create class-default and add remaining bandwidth with 'remaining'
keyword.

Also, it would be good to add WRED to class-default.

Which IOS do you use

 

I enable wred. set bandw of IF to 2500 and do some stats + class default to
bw remaining.


Class-map: client (match-any)
35740553 packets, 35805033296 bytes
1 minute offered rate 1961000 bps, drop rate 0 bps
Match: access-group 110
35740550 packets, 35805029460 bytes
1 minute rate 1961000 bps
Queueing
Strict Priority
Output Queue: Conversation 264
Bandwidth 1000 (kbps) Burst 25000 (Bytes)
(pkts matched/bytes matched) 11/13706
(total drops/bytes drops) 0/0


class-map: class-default (match-any)
307755610 packets, 83114905807 bytes
1 minute offered rate 2180000 bps, drop rate 0 bps
Match: any
Queueing
Output Queue: Conversation 265
Bandwidth remaining 90 (%)
(pkts matched/bytes matched) 513/107104
(depth/total drops/no-buffer drops) 0/0/0
exponential weight: 9
mean queue depth: 0

IF
MTU 1500 bytes, BW 2500 Kbit, DLY 100 usec,
reliability 255/255, txload 247/255, rxload 117/255
Available Bandwidth 619 kilobits/sec
1 minute input rate 1151000 bits/sec, 472 packets/sec
1 minute output rate 4150000 bits/sec, 596 packets/sec

Strange no ? bandwidth is set to 2500 and 1m output is 4Mb !!

 

No, it's not strange at all. 'bandwidth' command will not affect real
bandwidth. It's just used to tell some routing protocols and some QoS
mechanisms to find real or wanted bandwidth.

Your output looks just fine. I hope it will work as it should.

 

Yes but why the client exceed his 1Mb, if i i told
the iface to have 2.5Mb of BW , why client exceeed
his priority ..

 

Because you have set the minimum bandwidth in the case of congestion. If
you don't want for client to get more than 1Mb (congestion or not) you
should use traffic shaping.

If there is no congestion, the client will use all remaining bandwidth,
not just specified amount.

So, if you want to limit clients traffic.. word for you is traffic
shaping and policing.

 

no i don't want use shapping. but how can i simulate
the congestion .?? if i limit the internet port to 2mb
the routeur 'll have congestion ??

 

There's a way you can force congestion - by using class-based WFQ and
shaping.

The configuration is as follows:

policy-map MY_POLICY
class servers
bandwidth 1500 (reserve 1.5Mbps for servers)
class telnet
priority 10 (telnet traffic in rate of up to 10Kbps - put in
low-latency queue)
class class-default

policy-map MASTER_LIMIT
class class-default - take action on all traffic
shape average 2000000 - limit BW to 2Mbps (out)
service-policy MY_POLICY - manage congestion (2Mbps and above) by
this policy

Interface Ethernet0/0
service-policy output MASTER_LIMIT


in this situation, congestion will occur in 2Mbps rate on the
interface, rather than "waiting" for 10Mbps rate.

if only "servers" will talk, they can reach up to 2Mbps. if any other
traffic will take place, it will take 500Kbps max (2 - 1.5), and
"servers" will talk in at least\up to 1.5Mbps rate.

even if line is fully congested, telnet sessions will still work fine
under 10Kbps rate (which is far beyond what one session really needs)
- beacuse they are promised to get minimal latency in this rate and
below.

if you wouldn't put the 'priority' command, telnets will suffer really
high latency and sessions might fall.

hope I could help.
Ohad.

 

I'll try this ....

But it's sad that's not possible to specify rate
on a SW after the router

Router -> SW -> NET
I can rate limit the port on SW..
It's 'll be more easy to rate limit
on each port on SW and when SW drop packets..
the router can manage the congestion ..
(for multi homed traffic)..

If i have one link of 10m and a second link
of 100M, what's the value i enter in the router

 

well, it's not necessary to do anything on the switch level...
assuming most of the traffic is TCP, the congenstion the router makes
affects end-to-end - it delays\drops packets and the origin
server\workstation window becomes smaller and the session "stables" on
2Mbps end-to-end. but it's beeing managed more efficiantly by the
router - packets are put into queues insted of just being droped.

Ohad.

 

and in this cause, you think routeur gives priority ??
and give 1mb to priority class & 1mb to all other classes ?

 

During Period of simulate congestion ..

Sh queue

Input queue: 0/75/15781/0 (size/max/drops/flushes); Total output drops:
Queueing strategy: weighted fair
Output queue: 0/1000/64/0 (size/max total/threshold/drops)
Conversations 0/2/256 (active/max active/max total)
Reserved Conversations 1/1 (allocated/max allocated)
Available Bandwidth 8444 kilobits/sec

Normal ?
I make a download on the client server (priority) and on a default server
the speed of these are similar so.. It works or not