I find this contribution interesting.
We have had similar reflexions. Attached is a paper (to be published)
summarizing some of our work related to this topic.
- We have the same starting hypothesis: may be the common approach with
explicit end to end full mesh of MPLS LSPs is not the only solution. However
it permits optimal mapping of traffic onto the network resources, it has
some drawbacks from a scalability and complexity point of view.
- The question is whether we can find an alternative solution relying on IP
native shortest path routing (OSPF, IS-IS) possibly enhanced by some MPLS
LSPs. We can identify two kinds of solution:
1) try to realize the optimal routing pattern: this is the approach
described in the draft, or the one called "generalized ECMP" in our paper
which is the same approach although we do not propose the realization of
load splitting with MPLS LSPs;
2) try to find a good (may be not optimal compared to the optimal multipath
routing) solution relying on IP native shortest path routing enhanced with a
limited number of LSPs: we describe such an approach in our paper.
The difficulties we see in selecting such a routing strategy are the
following:
1) try to find a strategy that is good on operational networks because
sometimes you will often be able to find "worst case" topologies on which
such strategies perform very badly;
2) such solutions are kind of static because link weights cannot be changed
too often. However if the solution uses a (limited) number of LSP, changes
of the LSP may enable the routing to adaptat to traffic fluctuations (as
suggested in the draft). Idem in case of topology changes (link failure).
These topics however are not easy to study;
3) finally there are many options in integrationg IP native routing with
MPLS LSPs (advertise the LSPs in the IGP or not, how to modify the SPF
algorithm to take into account LSPs, how to select the tunnels metrics)
Do people agree on such an approach? Comments welcome.
Nicolas MICHEL
Yufei Wang wrote:
> FYI:
>
> We just submitted a draft on a new approach for
> traffic engineering. It solves the unequal traffic
> splitting problem of the IGP link metric tuning
> method. It's also capable of limited dynamic load
> balancing. The URL is
> http://www.ietf.org/internet-drafts/draft-wang-te-hybrid-approach-00.txt
>
> Yufei
>
> __________________________________________________
> Do You Yahoo!?
> Get personalized email addresses from Yahoo! Mail - only $35
> a year! http://personal.mail.yahoo.com/
-- ____________________________________________________________________France Telecom R&D
Nicolas MICHEL FTR&D/DAC/OAT/ALB e-mail: nicolas.michel@francetelecom.com tel: +33 01 45 29 41 77 fax: +33 01 45 29 60 69
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:03 EDT