The MPLS WG Archive

[Claus Gruber <claus.gruber@tum.de>]

Hello group,

>	  * Are you saying that that a strict MPLS LSP
>	  path can be duplicated with OSPF? 
>  
>
In OSPF and its extension ECMP (Equal Cost Multi Path) multiple paths 
towards one destination IP address are possible. The traffic is 
distributed equally. I.e. if there are three shortest paths (equal 
length) traffic is distributed 33%/33%/33% on outgoing interfaces. The 
traffic is distributed using a flow-based hash function ('per 
destination') to keep the ordering of packets of one flow (TCP problem) 
or in a round robin manner ('per packet').

Using the configurable OSPF link weights many MPLS paths can be 
duplicated with OSPF. However, MPLS has the advantage to distinguish 
packets (treat differently)  towards the same destination IP address 
that are entering the network at different edge routers. Additionally an 
unequal traffic distribution on multiple (not shortest) paths is possible.
(Note: A different routing for packets belonging to differnt type of 
service classes was included in OSPF v1 but is no longer included in 
OSPF v2 RFC)

With this, MPLS has an advantage considering traffic engineering. 
However, if a network operator optimizes the OSPF link metrics the 
difference is not that big for working traffic (without a failure). I 
did some optimizations both for ECMP and MPLS indicating this.

However, I think the main advantage of MPLS is the possibility for 
resilience. OSPF as well as IS-IS or RIB, ... are distributed protocols. 
In case of a network element failure it takes some time until failure 
notifications (LSAs) are broadcast and all routers have updated their 
forwarding table (with transient effects: loops).
When tuning timers it is however possible to reduce the convergence time 
of OSPF or IS-IS to some hundred milliseconds even for larger networks. 
I did simulations and I think SPRINT reported sub-second convergence 
times with patched Cisco routers already.

With MPLS one is able to prepare for a failure and establish backup 
paths in advance of possible failure patterns (e.g. Fast Reroute 
configuration). This is - at least to my knowledge - not possible with 
OSPF since it is a distributed protocol and there is only one valid 
forwarding table in a router. MPLS thus, is able to react locally and 
prepared (backup path already defined - "protection') upon failures 
which is of course faster (some tens of ms) than a distributed 
restoration approach.

The main advantage is the distribution of the backup traffic. After 
failure detection and rerouting new paths are calculated according to 
the old link metrics in OSPF. These metrics however were adapted for 
working traffic and might not be well fitted for the new (reduced) 
topology. MPLS backup paths can be chosen independently of the working 
paths (without a failure). Optimization results showed that MPLS has 
advantages considering required capacity.

And: A (good) optimization of link metrics and MPLS paths takes its 
time. In MPLS this can be performed offline for probable failure patterns.
If one might change an MPLS path this path is affected only. When 
changing OSPF link metrics routes towards different destinations are 
changed and cannot be changed only for one destination or for one 
ingress-egress relation. This makes optimization and TE more complex.

As a summary:

- MPLS has potentially shorter convergence times (allows local protection).
- Not much difference in TE for working traffic results (at least as far 
as I calculated and know)
- MPLS allows an independent traffic engineering for resilience issues 
(including failure patterns)
- Adaptive traffic engineering is possible without potentially changing 
a lot of routes towards multiple destinations.
- Forwarding is 'wire speed' both for IP and MPLS today.
- Virtual Private Networks (VPNs) are easy to deploy with MPLS.
- MPLS paths traversing multiple ASs are in development.
- MPLS Multicast is in development.

As a drawback: MPLS is an add-on technology and if not configured static 
a kind of IGP to detect topology is required. Paths have to be refreshed 
and additional switching tables are required in each router.

Sprint has some very nice white papers (to be found on their homepage): 
Summary: MPLS is good and has advantages - currently we can do what we 
need and want with tuned OSPF.

It depends on the requirements... but MPLS has many advantages.

Regards,
Claus

-- 

______________________________________________________________________

Claus Gruber
Institute of Communication Networks    Phone: +49 89 289 23508
Munich University of Technology        Fax:   +49 89 289 23523
Building 9, Room 1932                  mailto:claus.gruber@tum.de
Arcisstr. 21, D-80290 Muenchen         http://www.lkn.ei.tum.de/~claus
______________________________________________________________________




_______________________________________________
mpls mailing list
mpls@lists.ietf.org
https://www1.ietf.org/mailman/listinfo/mpls