QoS for MPLS VPN (noobie question)

Unanswered Question
Feb 9th, 2009
User Badges:

Hi All,


I need some help in understanding how to configure the point-to-cloud (hose) model for MPLS VPN's so we can offer some QoS guarantee for our MPLS VPN customers.


I've seen lots of explanation about what the hose model is but no configuration examples which is what I'm after.



Lab Topology:


CE1 -- PE1 -- P -- PE2 -- CE2


If we take the very simple topology above and we want to offer a 256K ICR and 512K ECR at both CE's, what is the necessary config to achieve this on the PE and P? Is it a simple police statement in both the output and input service-policy of the PE-to-CE interface to rate limit it accordingly or am I missing something here? Is there any marking done on the PE's? Can someone please draft me a config for my example?


Continuing on with the example above, does the hose model dictate that CE2 can only send 512K to CE1 and receive 256K from CE1? This leads me to my other confusion - what if CE1 and CE2 both have a 2M/2M SHDSL connection into the MPLS VPN cloud - aren't they expecting that they should be able to send/receive 2M/2M to/from their other sites? Wouldn't this policing mean that they've paid for a 2M/2M circuit but it's being rate limited to 512K/256K so wouldn't they be better off to have bought a 512K/256K ADSL connection???


Thank you in advance.


Cheers.


Andy

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
Giuseppe Larosa Tue, 02/10/2009 - 12:05
User Badges:
  • Super Silver, 17500 points or more
  • Hall of Fame,

    Founding Member

Hello Andy,

first of all a lab with three CE should be used.


if you have only two sites you can define clearly a point to point QoS model.


When it comes to provide QoS services when the VRF sites are a few to many a point to cloud model could be used:

instead of trying to define how much traffic can be sent/received between site1 and site2, how much can be sent/received between site1 and site3.

But this is not scalable and can lack the flexibility to use unused bandwidth when available.


and so on the idea is to define two rates:

how much can be sent to the cloud

how much can be received from the cloud.


To avoid huge losses is reasonable that a site can send less traffic then how much it can receive this allows for two, three sites to send traffic to a fourth one.

This also allows to accomodate customer topologies with a clear HQ site with much greater needs then other sites.


Configuration is actually simplified because it is just enough to police differently in the two directions without discrimination of actual destination site.


note: the access link speed and the servide speed a customer gets are two different parameters and it is common practice.

So the customer actually pays for the service.

The access link speed has to be better then the service to make it feasible.



Hope to help

Giuseppe



asaykao73 Thu, 02/12/2009 - 17:48
User Badges:

Hi Giuseppe,


Yeah I'm very confused about the access link speed and the service speed.


I'm trying to think like a customer and if I paid for a 2M/2M (site1) then I'd expect to send/receive this much to/from all my other VPN sites. I mean if site2 and site3 were both sending me 2M each then obviously I would get congestion at site1 because I've only got a 2M pipe.


But site1 was only able to receive 512K ICR and site2 and site3 were sending me 1M each, wouldn't I still hit congestion because eventhough I've got a 2M pipe, I'm being policed at 512K???


Thanks.


Andy

Mohamed Sobair Thu, 02/12/2009 - 22:58
User Badges:
  • Gold, 750 points or more


Hi Andy,


At some statment you are pointing to Policing, In another you are pointing to QoS marking! what is exactly your requirment? If you want to apply QoS marking , Service provider uses different techniques (Models) for Impleming QoS. Each model has its own characterstics, and they are:


1- Pipe Model.

2- Short Pipe model.

3- Short Model.


In Short pipr model, the forwarding of the traffic is based on the LSP diffserv information (Customer Marking), while In short model, the Traffic is forwarded based On the Tunnel LSP information. The Ip precedence bits are copied into the MPLS experemintal bits. If the marking done based on DSCP, then the 1st 3 bits are copied. The marking is preserved when the label is pushed or Swapped, However its not preseved when the label is Poped. So it really depends on the SP QoS Policy.


Looking at your last paragraph, the ADSL offers a symetrical Bandwidth. UPlink with rate and downlink with different speed. A gain what Service level agreement do you have with the Service provider?



HTH

Mohamed

asaykao73 Thu, 02/12/2009 - 23:18
User Badges:

Hi Mohamed,


Thank you for your reply.


I understand how packets are marked from IP to MPLS and MPLS to MPLS and MPLS to IP via the various models you've outlined.


My research into QoS for MPLS VPN has lead me to believe that there's two viable QoS solutions that SP can offer.


1/ point to cloud (hose model)

2/ point to point (pipe model)


For any-to-any MPLS VPN's, Ivan Pepelnjak wrote in his article below that the hose model is the most viable solution for SP.


"The hose model is the only viable model for any-to-any service, including full mesh MPLS VPN service and Virtual Private LAN (VPLS) service."


In Cisco's white paper "Cisco IOS MPLS Quality of Service", they also talk about the two models that SP can use to provide QoS for MPLS VPN's.


Reference:

MPLS QoS: Implementing the best model for guaranteed service

http://searchtelecom.techtarget.com/tip/0,289483,sid103_gci1338871,00.html


Cisco IOS MPLS Quality of Service

http://www.cisco.com/en/US/tech/tk436/tk428/technologies_white_paper09186a00800a4455.shtml


I'm interested in applying the point-to-cloud (hose) model in a lab environment. I believe this is done by simply policing the ICR and ECR on the PE-CE interface on the PE router but can't find any configuration examples or case study to help me.


1/ I was wondering if anyone has a config of the hose model because I haven't been able to find any on the internet and cisco web site.


Eg: Lab set up.


|--CE2

CE1--PE1--P--PE2--|

|--CE3


ALL sites are to have a 256K ICR and 512K ECR.



2/ I'm confused at the actual access line speed and the service speed to/from the VPN cloud. Access speed can be 2M/2M SHDSL as an example, yet the service or SLA provided by the SP might be 512K ICR and 256K ECR. I'm a bit lost as to why a customer would accept a lower service/SLA than their access speed.


Thanks.


Andy

asaykao73 Thu, 02/12/2009 - 23:19
User Badges:

Oops the lab diagram is all screwed up.


Should be CE1 connected to PE1 and CE2/CE3 connected to PE2.


Cheers.


Andy

Actions

This Discussion