Metro 3750 QoS

[jaurbano]

We are having an issue setting up traffic shaping on our Metro 3750 switches.  We have 5 different data flows (see diagram) on 5 different VLAN's.  When B is not flowing, we want A and E to be able to burst.  However, we haven't been able to set this up successfully.  Whenever we go over our max of 60 Mpbs, B drops below 43 Mpbs and A and E remain constant.  We have tried setting precedence for B but it doesn't seem to solve the issue.

Searching the web, we see the "shape peak" command.  However, it's not available on the switch we are using.  We are running Cisco IOS version 12.2(35)SE5.

[Lei Tian · Accepted solution]

Hi John,

I will try some configure like this

policy-map shape_60M

class class-default

shape 60M

service-policy child_policy

policy-map child_policy

class class_a

bandwidth 1M

police 4.6M

class class_b

bandwidth 43M

class class_c

bandwidth 10M

class class_d

bandwidth 1M

class class_e

bandwidth 5M

police 20M

HTH,

Lei Tian

[jaurbano]

I haven't tried the police with the bandwidth command.  I have tried the "shape average" command, but it seems to take over the bandwidth for the msision that is supposed to be given a guaranteed 43Mbps.  I will give it a try.  Thank you.....

[jaurbano]

At one time, I could have sworn I saw a "shape peak" and a "shape average" command on the Metro.  I don't see it anymore.   Perhaps it was on a different switch.  The policing, in conjunction with the bandwidth command should have the same result.

[Lei Tian]

Hi John,

Based on the command reference, shape peak is not supported on 3750 metro. I think that is a feature for software based device. You are correct, shaping and policing essentially is same. Can you try the config and see what's the result?

I think without congestion, that should be fine. class a will get max 4.6M; class e will get max 20M. During congestion, is the question whether the parent level shaper know which packet to drop.Because shaper is FIFO queue, when queue is full, packet will be tail dropped. If other classes are more aggressive, it could affect class b's rate.

I don't have device to try but I think WRED is supported on the parent shaper; if it does you can try apply WRED on the parent level shaper. Configure the WRED to drop class a and e before drop class b.

HTH,

Lei Tian

[jaurbano]

The use of bandwidth, in conjunction with policing did not work.   We could never reach our peak, and the throughput was very choppy.  Also tried the bandwidth, along with the "shape average" command.   Still no joy.

[Lei Tian]

Were you testing it when total taffic > 60M? How did you test it? could never reach the peak means class B cannot get 43M?

Regards,

Lei Tian

[jaurbano]

No... we were not flowing all data flows at the same time.   At most, I was flowing the 5-20Mbps data flow, and the 43Mbps.

I was expecting the data flow with the configured with the policing to drop to it's minimum when the 43Mbps data flow was flowing.

[Lei Tian]

Hi,

If you only sent out traffic in that 2 classes, then 43M will be guaranteed, and the other class will use the left over 17M. Bandwidth command reserves the minimum bandwidth during congestion, it is not a upper limits. Policing, on the other hand, can limit the max bandwidth at any time.

If you can test with all traffic, with congestion, each class should get a percentage close to what is configured.

Regards,

Lei Tian