[j-nsp] MPLS fast convergence without link information

[Phil Bedard]

On 12/29/11 1:48 PM, "Mark Smith" <ggglabs0 at gmail.com> wrote:

>On Thu, Dec 29, 2011 at 6:45 PM, Phil Bedard <philxor at gmail.com> wrote:
>> BFD on the IGP session would be the best option.  We currently run ours
>>at 300x3 on Juniper boxes and lower on some other platforms with better
>>hardware processing of BFD packets.   If you are using aggregates
>>something like 802.3ah may work better as BFD control traffic has a
>>tendency to just take a single link on Junipers.  It is honor miss
>>though whether L2 circuit provider can tunnel that traffic.
>
>Thanks for comments.
>
>Correct me if I am wrong, but doesn't enabling BFD on the IGP cause
>IGP to reconverge in case of fault?
>
>Is there any (reasonable) way to utilize link-node-protection or FRR
>if link state information is not usable between MPLS routers? Or are
>these fast repair mechanisms only usable in case of dark fiber (or
>copper)?
>
>To put this in another way: I cannot expect sub-second recovery times
>unless the failure detection is triggered by interface link down?
>
>I had impression that one could use BFD to trigger facility backup
>(bypass LSP) in sub-second time (e.g. 3x 50ms). This appears to be
>incorrect.

BFD on the IGP will trigger a fast protection switch with either RSVP-TE
MPLS FRR or IP LFA (IP FRR) being used. The state of RSVP on a link on
Junipers is explicitly tied to the IGP state, so when the IGP goes down it
will trigger the facility bypass.  As for 3x50ms you can try it out in the
lab and see how it goes, maybe with newer offloading to the PFE the timers
can be run faster.  There are vendors out there who can run as low as 10x3
and get sub-50ms restoration times, but I'm not aware of Juniper being one
of them.  I've seen Junipers running at 100x3 and a facility bypass
restoration happens in about 310ms, so definitely sub-second.


>
>I would not expect the L2 service provider to be able to tunnel
>ethernet OAM (CCM etc) traffic.

Yep it is hit or miss.  The equipment we use for L2 circuits does support
it.  802.3ah operates at an entire port level and many who are using L2
circuits are hub and spoke with VLANs separating remote sites, so it's not
real useful, but for point to point it does work well.

>
>I wonder why haven't I found good information about this in the 'net.
>Am I blind or do all major players always have dark fibers (or
>similar) at their disposal? How about unidirectional fiber failures?
>At least in some cases (e.g. all autonegotiation stuff disabled)
>unidirectional fiber results in link up on the other (where rx sees
>light) end and link down on the other. How are these handled?

I think you'll find even with those who have dark fiber based networks,
many still use BFD because there are forwarding plane failures which still
result in link-state remaining up.

>
>
>I really appreciate comments and discussion.
>
>_______________________________________________
>juniper-nsp mailing list juniper-nsp at puck.nether.net
>https://puck.nether.net/mailman/listinfo/juniper-nsp