[c-nsp] SR (was BGP Signalled VPLS)

Tassos Chatzithomaoglou achatz at forthnetgroup.gr
Fri Aug 9 16:10:50 EDT 2013


Just wondering, is it possible to somehow create a segment path from an SDN controller, using more labels than the ones supported in an intermediate hop?
Will the SDN controller be able to take into account such per-node limitations?

Although not an expert on this area, it amazes/worries me that although IETF docs are in very early stage, vendor EFT code already exists.
It looks like most things are already predetermined/locked-in and no surprises should be expected.

--
Tassos

Saku Ytti wrote on 9/8/2013 21:55:
> On (2013-08-09 18:24 +0300), Nitzan Tzelniker wrote:
>
>> regarding SR one thing that I thing is a problem with it is the number of
>> labels you can push on the smaller hardware platform (ASR901/Juniper ACX)
>> if you want to do a strict path you will have to push a lot of labels and
>> in these platforms you are limited.
> I wonder what the actual HW limitations are, ASR901 I think is CSCO
> internal, ACX is BCM.
>
> Is the documented mpls label stack depth limitation how many levels of
> abstraction you have (lookup VPN label in VRF table, then lookup IGP label
> in LDP/IGP, two lookups, two levels of abstraction, essentially two
> lookups). 
> If this is the limit, then in explicit route case this limit might be
> exceeded by returning more than 1 label as rewrite information of single
> lookup, in explicit route, you'd know whole stack after one lookup, it
> would just return more data.
>
> IPv4 tunnel lookup returning 20B rewrite information might not be any
> different to lookup returning 5*4B labels.
> Or lookup result being IPv6 header might be analogous to returning 10
> labels.
>
> I however don't have any data, this is purely conjecture.



More information about the cisco-nsp mailing list