VPLS Experience

Wed Oct 12 07:19:33 UTC 2005

On 10/12/05, Mohacsi Janos <mohacsi at niif.hu> wrote:
> On Wed, 12 Oct 2005, danazster wrote:
> > I keep hearing that VPLS is a Good Thing.
>
> It would be if it would be supported and would be compatible.
> - There are different vendor implementations: LDP oriented or BGP
> oriented. And there is some movement to be RADIUS oriented also.

The RADIUS rfc expired long ago. It looks only like a BGP/LDP
discussion now, and I would really like to hear of any impact this has
on performance between the two, eg does LDP converge similarly, or
paths disappear faster?? Are there any practical differences between
the two people have seen?

> - Some vendors are supporting only in certain HW and line-card
> combinations.
>
> So if you dive into VPLS you might expect lots of debug and "blood and
> sweat" and fights with your vendor....

So what else is new? Sounds like life as we know it. I can list
GE/SDH/Sonet combos that don't work on certain line cards and chassis,
OS revs. Has been the case for years.

Basically I don't want to get into a vendor debate so won't comment on
which vendor/s this applies to.

I really want to hear about what people are actually doing.

> > Indeed for certain design models it seems to offer some real
> > advantages. There doesn't seem to be anywhere near the level of
> > stability concern that we saw when MPLS first came out - there was a
> > really strong negative about MPLS in 1999 at nanog, behind the scenes
> > anyway. [1]
> >
> > Has anyone got any good deployment or lab test experience they can share?
> >
> > Obviously comment on debug tools, operating models etc would be cool.
> >
> > Basically I am hearing a bit of FUD about jumbo frames, dDOS and
> > multicast, but they look like design problems to me.
>
> They started to work on multicast support: VPLS currently does not support
> efficient multicasting

True, the requirements specs are in draft form now, but multicast
works better from what I can see in VPLS than it does in MPLS.

The theoretical issue is that PEs potentially relay multicast to ports
not part of the VPN is this true? Surely you can configure that out
with filters on the edge?

danazster