The MPLS WG Archive

[Florian-Daniel Otel <otel@ce.chalmers.se>]

I couldn't rezist to beat up the dead dog with just one more posting...;>

> Florian,

> The plain fact is that it just doesn't matter.  The shortest speed of
> light delays in any WAN application is on the order of a few
> milliseconds.  Queueing delay can be hundreds of milliseconds if the
> outbound interface is congested.  Modern routers forward IP in a few
> 10s of microseconds (on fast interfaces).  If the path is the same and
> the same queue is used, then the difference in forwarding speed
> between an IP route lookup and an MPLS label lookup isn't even
> measurable and in some (most?) routers the difference is exactly zero.


That was my point exactly when i said that this type of measurement,
even done with production router software, is irrelevant. As some
people already pointed out, IP routing software can be (and is) done
as fast as MPLS switching, e.g. hardw implementation, pipelined ASICs,
hashing, advanced searching techniques, routing table compression, etc,
etc. It's a whole world here, i don't even dare to open this
Pandorra's box here..

Back to the original posting, I really cannot imagine any non-trivial
tests aiming to prove that  'MPLS is faster than IP' (yes, the
statement doesn't make much sense). One such example might
(arguably) be the ability to do source routing at lower costs than
IP-in-IP tunneling/ IP source routing.

Again, MPLS strongest points aren't performance but the network
engineering capabilities it offers,

> Curtis

Cheers,

Florian.