The MPLS WG Archive

["Phani Koganti" <pkoganti@newbridge.com>]

I have been doing some thinking on this area for a while, but am not sure
if i am in the right direction. Please correct me where ever neccessary.

For Qos routing across autonomous systems isn't it required that there be a mechanism to figure out whether we have the required Qos end to end.

Doesn't Bandwidth broker concept address the inter-domain Qos routing ?

Instead of having a centralized entity per AS as in bandwidth broker architecture, if all the routers running BGP can generate bandwidth and QoS related attributes for the routes generated from the AS (along with the labels). As a result based on the policies the routers can have multiple routes to a destination mapping to different Qos/bandwidths.

So seems like we need a well defined mechanism to export the traffic engineering attributes of the IGP protocols to the BGP protocol and also, the intermediate ASs when forwarding the routes received need to figure out the QoS/bandwidth available to the next-hop and forward the cumulative QoS/Bandwidth constrained route to other ASs.

Phani

Phanidhar Koganti
Newbridge Networks Inc.

Peter Ashwood-Smith wrote:

 

Currently the MPLS working group has not addressed this subject in any depth although there was a draft presented in the routing area group today in Adelaide that started to talk about solutions.

As far as I know there are basically two ways to attack this problem, or at least two general flavors.

piecewise source routing. In this approach you compute a gateway based on the destination address and then do normal flat source routing to that gateway. Go through the gateway and repeat. In other words you are doing piecewise traffic engineering but letting the exterior gateway protocols figure out which gateway you should be going to next. Obviously there are problems with this approach since it puts everything though the same gateway. I actually helped built a piecewise solution like this a while ago. It has some good advantages in that local (at least local to an AS) repair is natural to do as are local modification etc.

hierarchical source routing. This is what ATM does, you compute a hop list hierarchically against a hierarchical topology database etc. etc. I'm not sure this approach is appropriate for the Internet due to the massive size of the problem but I'm really not an ATM hierarchical routing guy.

I would hope that the MPLS working group or the Traffic Engineering working group can start looking at these issues, possibly with some simulations of the different approaches against a set of real AS topologies. Some of the PNNI architects and BGP architects are active members  of the MPLS working group so you can be sure that it will cause some heated debate ... should be extremely interesting and challenging.

Cheers,

Peter
 

-----Original Message-----
From:   Andrzej Czerczak [SMTP:Andrzej_Czerczak/HeadQ@netia.pl]
Sent:   Tuesday, March 28, 2000 6:04 AM
To:     mpls@UU.NET
Subject:        MPLS routing accross AS boundaries

Hi,
 Where can I get information on the subject of MPLS QoS routing accross
Autonomous System boundaries?

Thanks
Andrzej