Re: [PWE3] WG Last Call for draft-ietf-pwe3-endpoint-fast-protection-01

[Stewart Bryant <[e-mail]>]

On 07/08/2014 16:01, Yimin Shen wrote:
> Hi Stewart,
>
> Thanks very much again for the review and detailed comments.
>
> We will consider your suggestions on the LDP TLV structures defined in this draft.
>
> Regarding the tradeoff between the need for egress PE node protection and the increased state and complexity in PSN, this is a common considration for all IP/MPLS node protection, not PWE3 specific.
SB> Agree
> We do see the requirement from service providers. I think this should be their decision on a per network basis. When they decide to do it, we have a solution available for them.
SB> Sure but what is the view of the WG. Do we want to recommend this 
approach as the best approach to the industry, or should we recommend a 
simpler approach that allows better layer separation. Perhaps some 
operators could comment on what their requirements are so we can better 
a better feel for how to make the complexity+scale trade off against 
protection coverage.

SB> Remember a standards track RFC is a serious recommendation of good 
practice.

>
> That said, this draft doesn't necessarily multiply the number of PSN tunnels. It does require one tunnel per <primary PE, protector>.
SB> Surely it requires on per protection group?

> However, if each primary PE is protected by only 1 protector, the number of tunnel will be the number of primary PEs, which is basically the same as the case where this mechanism is not used.
SB> Not necessarily. This is a consequence of your method. Without 
context labels you deliver the traffic on the ordinary IP path to the 
alternative T-PE
> This 1:1 relationship is achievable with the co-located model, and is especially achievable with the centralized model. Of course, in a more general case, if the avarage PE-to-protector ratio is 1:N, the number of tunnels will be multiplied by N.
SB> With the centralized model - is this a PSN function, or is the 
centralized protector a type of S-PE?

> But operators can always design their networks thoughtfully and manage this N to be lowever than the threshold that may lead to scalability problems. (Here, I'm not counting bypass tunnels in, because bypass tunnels exist in the existing IP/MPLS fast-reroute as well.)
SB>Yes, but with some sort PW stitching model this simply does not apply.

- Stewart