MPLS and QoS 'traffic shaping' question

[desmith]

Hello!

I am wondering what the best way is to implement some sort of traffic shaping for a customer using an MPLS 'cloud' for their WAN connections.  They have several remote sites connected with full T1s, one with 2 x T1, and the central site is connected with 4 x T1s. 

Obviously, my concern is not to send traffic from the 4 x T1 (or the 2 x T1) site to a T1 site at a higher bandwidth than that site can handle.  In the past, I'd have set up FRTS per DLCI to match the bandwidth of the remote side.  I'm wondering what the current best practice recommendation is for this situation (i.e., a "full mesh" with BGP routes from each site directly to any other site). 

By the way, I have configured CBWFQ/LLQ outbound on all routers, and so the most critical traffic (VoIP) should be OK, but I'm concerned about shaping so that other traffic types don't get sent at higher bandwidths than the remote sides can handle.

Thanks for any suggestions/directions that you may have!

Deb

[vmiller]

First thing, make contact with the MPLS provider and find out what they support for this customers service level.

What ever you do will have to work in their framework.

[desmith]

Yes, thanks vmiller, I do understand the need to coordinate with the MPLS provider.  However, if the central site sends traffic at 4 times the rate that the far end can handle, we're very likely to have problems (no matter what the SP is doing for the customer).  So, I'm trying to figure out what the current best practice is for that central site's customer premise router.

Thanks again for your suggestion, and we certainly will do so!

[vmiller]

You will need to rate limit after you classify, and make it work with how the carrier handles QOS on the MPLS backbone.

You can start here to get cozy with the terminolgy. there are some nice design guides on the cisco site.

http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note09186a00800a3a25.shtml

[desmith]

Thanks, vmiller.  I was already completely familiar with the material at the link you sent; however, at the very bottom of that doc is a link to CBWFQ *inside* of GTS, and the ability to do *that* is the what I've been missing.  I have CBWFQ set up already, and shaping or policing on top of that (using GTS) differently to the different remote sites based on their connected bandwidth looks like exactly what I need.

Thanks again for your time!  It's very much appreciated!

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

You can shape on each sites egress to avoid over running the far side's ingress, however this is only practical if you have few sites and also assuming you can treat your topology as hub-and-spoke.  (I.e. a logical topology much like you might have for frame-relay or ATM.)

If you want to take advantage of L3 VPN on MPLS, then you need QoS support from your MPLS vendor for some options of egress processing from their MPLS cloud to each of your sites.  You still want QoS to manage egress congestion to the MPLS cloud but additionally you would mark traffic so that it will be properly treated upon MPLS egress.

The two common disadvantages of this approach are often the MPLS vendor might only offer rudimentary egress QoS policy options and they might charge more for activation (especially for real-time guarantees).

[desmith]

Thanks for your time and suggestions, Joseph.

I've followed the config examples for "CBWFQ inside GTS" at this link:

http://www.cisco.com/en/US/docs/ios/12_2/qos/configuration/guide/qcfcbshp.html#wp1002823

However, the config is not working. 

I'll paste in here the config example from that link, and then use it to explain the problem I'm having:

cust1-classes Configuration

Router(config)# policy-map cust1-classes

Router(config-pmap)# class gold

Router(config-pmap-c)# bandwidth percent 50

Router(config-pmap)# class silver

Router(config-pmap-c)# bandwidth percent 20

Router(config-pmap)# class bronze

Router(config-pmap-c)# bandwidth percent 15

cust2-classes Configuration

Router(config)# policy-map cust2-classes

Router(config-pmap)# class gold

Router(config-pmap-c)# bandwidth percent 30

Router(config-pmap)# class silver

Router(config-pmap-c)# bandwidth percent 15

Router(config-pmap)# class bronze

Router(config-pmap-c)# bandwidth percent 10

Customer Policy and QoS Features Configuration

Router(config)# policy-map cust-policy

Router(config-pmap)# class cust1

Router(config-pmap-c)# shape average 384000

Router(config-pmap-c)# service-policy cust1-classes

Router(config-pmap)# class cust2

Router(config-pmap-c)# shape peak 512000

Router(config-pmap-c)# service-policy cust2-classes

Router(config-pmap-c)# interface Serial 3/2

Router(config-if)# service out cust-policy

So, in my config, when I entered the equivalent commands to these:

     policy-map cust-policy

       class cust1              

I got an error message that the class-map cust1 wasn't configured yet.....hmmmm...this example made it look like that wasn't necessary.  However, I issued the (equivalent for my config) command to:

     class-map cust1

Then, the policy-map accepted the command, and I proceeded to apply the shape and service-policy commands.

So, I thought all was well.  However, *no* traffic is matching my embedded CBWFQ service-policies.  No packets, no bytes.  (All packets are matching the class-default class-map.)

Any suggestions?  Does anyone have a complete known good working config that I could see?

(I should probably mention that I have defined a bunch of class-maps that are used by the (equivalent of) the embedded "service-policy cust1-classes", "service-policy cust2-classes", etc.)

Thanks again for any time/suggestions/help!

Deb

Message was edited by: Deborah Smith

[Joseph W. Doherty · Accepted solution]

Disclaimer

The  Author of this posting offers the information contained within this  posting without consideration and with the reader's understanding that  there's no implied or expressed suitability or fitness for any purpose.  Information provided is for informational purposes only and should not  be construed as rendering professional advice of any kind. Usage of this  posting's information is solely at reader's own risk.

Liability Disclaimer

In  no event shall Author be liable for any damages whatsoever (including,  without limitation, damages for loss of use, data or profit) arising out  of the use or inability to use the posting's information even if Author  has been advised of the possibility of such damage.

Posting

For your T1 sites, try something like:

class-map match-any VoIP

match . . .

policy-map CBWFQ

class VoIP

priority percent 30

class class-default

fair-queue

For your multiple bundled T1 sites, try something like:

class-map match-any Site1

match . . .

class-map match-any Site2

match . . .

class-map match-any SiteN

match . . .

policy-map Shape_T1

class Site1

shape average 1300000

service-policy CBWFQ

class Site2

shape average 1300000

service-policy CBWFQ

class SiteN

shape average 1300000

service-policy CBWFQ

[desmith]

Hello, Joseph.

Thank you for your suggestion.  I have set up something similar to what you suggested. 

I *thought* that all was well.  However, for some reason the target shape rate is twice the configured peak rate:

sho policy-map int mu1

Multilink1

  Service-policy output: QoS-Shape-T1-Out

    Class-map:   (match-any)

      5266361 packets, 1472713784 bytes

      30 second offered rate 58000 bps, drop rate 0 bps

      Match: access-group name -Subnets

        5266361 packets, 1472713784 bytes

        30 second rate 58000 bps

      Queueing

      queue limit 64 packets

      (queue depth/total drops/no-buffer drops) 0/3070/0

      (pkts output/bytes output) 5263291/1470261787

      bandwidth 760 kbps

      shape (peak) cir 1400000, bc 5600, be 5600

     target shape rate 2800000

FYI:  The interface Mu1 consists of 4 x T1.  It is exactly twice over-subscribed by either single or dual T1s at the remote sites.  So, you can see in the config below that I'm setting the guarenteed minimum bandwidth for each single T1 site to 760k (a bit less than half of a T1), and trying to shape to a peak of 1400kbps (a bit less than a full T1).  I've been poring through docs on CCO, but so far I'm baffled as to why the target shape rate is being set by IOS to 2800k. 

Thanks very, very much for your help already!  If you have any idea on why the target shape rate is twice the CIR, I would greatly appreciate hearing it!

Best Regards,

Deb

P.S.

By the way, I tried to use 768k for the bandwidth, but it seems that with IOS 15.x at least 1% of the available b/w of the physical interface must not be reserved (i.e., left for the automatically created class "class-default").  So, I used 760.

-----------------------------------------------------------------------------------------

Here is an excerpt of the relevant parts of the configuration:

class-map match-any Network-Control

  match ip dscp 48

  match ip dscp 56

class-map match-any VOIP-Bearer

  match ip dscp 46

class-map match-any VOIP-Signaling

  match ip dscp 24

!

policy-map QoS-LLQ-Out

  class Network-Control

    bandwidth percent 5

  class VOIP-Bearer

    priority percent 18

  class VOIP-Signaling

    bandwidth percent 5

!

class-map match-any Site1

  match access-group name Site1-Subnets

class-map match-any Site2

  match access-group name Site2-Subnets

!

policy-map QoS-Shape-T1-Out

                      class Site1

     bandwidth 760

    shape peak 1400000

    service-policy QoS-LLQ-Out

class Cashmere

    bandwidth 760

   shape peak 1400000

    service-policy QoS-LLQ-Out

!

[Joseph W. Doherty · Accepted solution]

Disclaimer

The   Author of this posting offers the information contained within this   posting without consideration and with the reader's understanding that   there's no implied or expressed suitability or fitness for any purpose.   Information provided is for informational purposes only and should not   be construed as rendering professional advice of any kind. Usage of  this  posting's information is solely at reader's own risk.

Liability Disclaimer

In   no event shall Author be liable for any damages whatsoever (including,   without limitation, damages for loss of use, data or profit) arising  out  of the use or inability to use the posting's information even if  Author  has been advised of the possibility of such damage.

Posting

Your 2x is probably because you're using shape peak, not shape average (as I suggested).

I also suggest shaping at about 85% of nominal bandwidth, to allow for L2 overhead not accounted for by a L3 shaper; becomes important for VoIP.

Lastly, with FQ in clas-default, you often don't need any other classes other than LLQ. I.e. you probably don't need the explicit classes for Network-Control or VoIP-Signaling.

[desmith]

Thank you, once again, Joseph.  "Shape average" (yes, as you suggested) did take care of the 2x target shape rate problem (*that* was unexpected: shaping  to a "peak" sounds like a not-to-exceed, vs shaping to an average). 

Once again, I offer my thanks and appreciation for all of your time and suggestions!

Deb