Re: [PWE3] [mpls] WG Last Call for draft-ietf-pwe3-endpoint-fast-protection-01 - RFC4447

[Loa Andersson <[e-mail]>]

Yakov,


On 2014-08-08 21:27, Yakov Rekhter wrote:
> Loa,
>
>> Yakov,
>>
>> On 2014-08-08 16:02, Yakov Rekhter wrote:
>>> Loa,
>>>
>>>> Authors, Mingui, Stewart
>>>>
>>>> On 2014-08-08 04:46, Mingui Zhang wrote:
>>>>> Hi Stewart,
>>>>>
>>>>> I think authors would say the S-PE stitching method involves
>>>>> the control plane processing during the repair procedure. They
>>>>> emphasized their method uses data plane & local repair, which
>>>>> can be faster.
>>>>> Here, I want to raise one issue:
>>>>
>>>> It seems that we take it for granted that the method proposed on this
>>>> draft is faster than e.g. the e2e protection a la mpls-tp. Why is that?
>>>
>>> To answer your question let me quote from "Network Recovery" by JP
>>> Vasseur, Mario Pickavet, Piet Demeester (Section 5.8.1, page 336):
>>>
>>>     With global protection, rerouting is performed by the head-end
>>>     LSR, which means that this requires for the head-end LSR to receive
>>>     the failure indication to reroute the affected traffic onto their
>>>     respective backup paths (whose paths have been precomputed and
>>>     signaled). So in terms of recovery time, the delta between global
>>>     and local protection is the failure indication signal propagation
>>>     time to the head-end LSR.
>>>
>>> Yakov.
>>>
>>
>> Well, I won't argue that this is what is said in the book, but it is
>> not what I asked about.
>>
>> mpls-tp runs e.g. 1:1 or 1:n, i.e. 1 working lsp (or pw) is protected
>> by one or more pre-established  lsp's or pw's. detection time can be
>> as low as 10ms and the switch over is far shorter.
>>
>> While I understand that the draft is much more resource conservative,
>> what is the foundation to claim that is faster?
>
> Local protection does not involve propagating failure indication
> beyond the point of failure.
>
> Do you think that the detection time at the head-end can be less
> than the time it takes to propagate failure indication from the
> point of failure to the head-end ?
>
> Furhermore, even if we assume that for some topologies the propagation
> time could be 10ms, could it be 10ms irrespective of the distance
> and the number of hops spanned by a PW ?
>
> Yakov.
>

MPLS-TP style protection does not involve propagating (i.e. does not
involve sending a a failure indication from the point of detection to
the head end). The failure at the head end is e.g. detected if you
loose three consecutive keep alive message traveling in the data plane.

If you send one keep-alive message every 3.3 msec, you can reach a
time to repair this is
- detection time (approximately 10 ms)
- transfer time (typically usec's)
- switchover time (good guess is approximately 10 ms)

MPLS-TP protection was designed to met the 50 msec requirement, it is my
guess that is about the same as what this draft achieves.

Note that I'm not say that the protection mechanism described is "bad"
or "slow", to the contrary I think it is both useful and fast, only that
we are talking of time to repair for the to mechanisms.

/Loa

-- 


Loa Andersson                        email: [e-mail]
Senior MPLS Expert                          [e-mail]
Huawei Technologies (consultant)     phone: [phone]