[j-nsp] Application of L2 VPN in Real World Scenario

[David Ball]

  We've been using L2VPN exclusively for PtP customers and it works
just fine.  Haven't tested the scalability much yet, as we have less
than 100 of them on our busiest box, and it's a T-series with ample
resources.  I dabbled in L2Circuits recently, and found that I liked
them as well, especially due to the vendor interop that existed (built
ckts to a couple Foundry XMRs and even MRV OS9000s), and also because
I personally found the configuration much simpler.  We may look at
moving to L2ckts in the future.
  Same goes for VPLS.  Have been using BGP-signalled up until now, but
recently tested the now-supported LDP-signalled flavour with another
vendor and interop was pretty good.
  As for the earlier question about why people use L2VPN vs. VPLS, I
think this really comes down to resource consumption concerns on
smaller platforms, mainly in the form of maintaining MAC tables.  We
avoid it if we can, but certainly do have several VPLS customers, each
with a couple thousand MACs.

David


2008/7/16 Mark Tinka <mtinka at globaltransit.net>:
> On Wednesday 16 July 2008 21:19:47 Abhi wrote:
>
>>     And how well do L2 VPN and L2 Ciruit scale up in the
>> SP environment and which are more preferred ones for
>> deploying these services.
>
> This is likely to start a heated debate, but I guess scaling
> of L2VPN's comes down to the signaling + discovery
> mechanisms in play (and perhaps, which "Kompella" you
> prefer as in the case for VPLS).
>
> Someone chime in in case I over-simplify the issue:
>
> Cisco primarily went for Martini L2VPN's based on targeted
> LDP sessions to signal l2circuits between PE routers.
> Juniper support this also.
>
> Juniper, on the other hand, felt that BGP should be used to,
> both, signal and autodiscover L2VPN's, including VPLS.
>
> Later, Cisco merged both schools of thought and, through the
> Rosen draft, support BGP Autodiscovery but maintain LDP
> signaling.
>
> Personally, I'm for the Juniper proposal. It makes sense to
> me, but like I said, this could get religious :-).
>
> There is another scaling technique (also supported by Cisco)
> called H-VPLS (Hierarchical VPLS), in which the concept of
> u-PE's (user-facing PE routers) and n-PE's (network-facing
> PE routers) are used to partition the VPLS domain into
> smaller, more manageable chunks that need to hold only a
> subset of LDP forwarding information, rather than that of
> the whole network.
>
> Choose your poison :-).
>
> Mark.
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/juniper-nsp
>