The MPLS-OPS Archive

Cell Relay Retreat>MPLS-OPS Archive>month:2001-Jun> msg00321



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

Re: (HELP) MPLS QoS end-to-end !!!

  • From: Robert Raszuk <raszuk@cisco.com>
  • Date: Mon, 25 Jun 2001 13:46:25 -0700
  • CC: MPLS Operations <mpls-ops@mplsrc.com>, Alexander Marhold <alexander@marhold.at>
  • Organization: Signature: http://www.employees.org/~raszuk/sig/
  • Resent-Date: Mon, 25 Jun 2001 18:26:40 -0400
  • To: Sivaramakrishna Iyer Krishnan <krishnan@research.bell-labs.com>

Hi Siva,

> My question is how does R1/R2/R3 and R4 differentiate between MPLS
> traffic (this means that the traffic carries a MPLS label) and BE
> traffic.
> Is their a work around ?.

Well it does not really matter for the scheduler if the packet is a mpls
packet or not. It also does not matter if this came via preestablished
TE-LSP or not. Each packet during it's life in the router is described
internally. One of this description is it's qos taken from IP Prec field
or MPLS EXP field etc ...

So the treatment for all the packest will be identical if they carry the
same diffserv code point characteristics. 

Reg the R4 itself if you set on R1 different value of qos into MPLS
header rather then it is in IP packet you may use explicit null label
which will carry the modified qos value to the very end.

> Also, R3 puts implicit null into the packet before sending it to R4 and

There is no such a thing as putting "implicit null label into the
packet" ;). Reception of implicit null binding simply mean do PHP.

> I can tell it to put explicit null in which case R4 will do label
> popping.

Yes.

> How does, R4 send traffic to SB2 and provide QoS ???...

Again based on the EXP field or IP prec of arriving data.

> BIG QUESTION : How to provide end-to-end QoS ????...

SMALL ANSWER: The same way as you would do it with IP and no MPLS at
all. An extenstion the this answer could be that diffserv aware TE may
provide you with some help here and support easier provisioning.

> I think I should configure CAR at R1, R2, R3 and R4 on the
> MPLS path so that any traffic from the source IP should be
> dropped.

Well this is precisely the issue with not 100% of TE in your netowrk. If
you don't have 100% of your external traffic taking the TE-LSPs via
transiting via your netowrk you are out of control and the only help TE
could do for you is the hot spot bypass.

> Do I need to do WFQ ?.

Some sort of intelligent scheduling will be needed. WFQ or rather CBWFQ
could be an option.

R.

> Sivaramakrishna Iyer Krishnan wrote:
> 
> Hi Robert,
> 
> Need some help.
> 
> Topology :
> 
> SB1--- R1 ---R2 ---R3 ---R4 ---SB2
>                                   |
>                                   --- SB3
> 
> SB - smart bits
> 
> Created a MPLS tunnel from R1(loopback) to R4(loopback).
> Any traffic that I send from SB1 to SB2 is going through the
> tunnel (no policy routing for now). The traffic gets labelled at
> R1 (ingress router) and has no label when it reaches R4 owing to PHP.
> 
> I want to reserve 40 Mbps bandwidth for MPLS traffic and
> let best effort have 60 Mbps of traffic. R3 pops the label before
> it reaches R4 as R4 is its last hop.
> 
> My question is how does R1/R2/R3 and R4 differentiate between MPLS
> traffic (this means that the traffic carries a MPLS label) and BE
> traffic.
> Is their a work around ?.
> 
> Also, R3 puts implicit null into the packet before sending it to R4 and
> I can tell it to put explicit null in which case R4 will do label
> popping.
> How does, R4 send traffic to SB2 and provide QoS ???...
> 
> BIG QUESTION : How to provide end-to-end QoS ????...
> 
> I think I should configure CAR at R1, R2, R3 and R4 on the
> MPLS path so that any traffic from the source IP should be
> dropped.
> 
> Do I need to do WFQ ?.
> 
> Will class based and policy based queueing help ?.
> 
> Thanks in advance.
> 
> Thanks,
> Krishnan.
> 
> -------
> The MPLS-OPS Mailing List
> Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
> Archive: http://www.mplsrc.com/mpls-ops_archive.shtml

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml