[j-nsp] IP/MPLS fast convergence

[Phil Bedard]

LFA relies on the topology to be one in which a failure won't cause a traffic to head back the other direction.  So topologies such as rings are not good because an adjacent node may send traffic back to a node with an upstream failure creating a microloop.  The router will not create a protection path in that scenario just use regular convergence. LFA works well in triangle or hub and spoke topologies.  You have to be careful with setting metrics as well. 

There has been some work on newer algorithms and ways to notify adjacent nodes either inband or out of band of a failure to ensure 100% coverage.   Today I do not think any vendor has implemented those features. 

As for the design it really depends on the SLA.  If you do not need 50ms Traffic restoration LDP should work fine.  On a metro network IGP convergence is pretty fast these days probably less than 500ms. 

Phil

On Dec 21, 2011, at 3:23 PM, Amos Rosenboim <amos at oasis-techs.net> wrote:

> Hello Serge,
> 
> Thanks for sharing your insights, it's valuable.
> Can you explain why only some of the paths are protected ?
> 
> Regards,
> 
> Amos
> 
> Sent from my iPhone
> 
> On 21 Dec 2011, at 21:47, "Serge Vautour" <sergevautour at yahoo.ca<mailto:sergevautour at yahoo.ca>> wrote:
> 
> Hello,
> 
> We did started a greenfield deployment 2 years ago. We had no requirement for FRR and stayed clear of RSVP. We did implement LDP + OSPF LFA since it was just an extra knob and gave us something for free. We used Link Protection.
> 
> 
> The caveat with LFA is that it will not protect 100% of your paths. A given link failure may re-route some of your traffic via LFA while the rest has to wait for standard OSPF convergence. WANDL did some LFA coverage analysis for us and if I remember correctly based on our topology we have ~70% of paths covered.
> 
> 
> I ran some tests on our Prod network before it went live. LFA did in fact allow sub-50ms convergence. For paths that weren't covered by LFA in a worst case scenario, I got about 300ms. Not too bad. Junos seems really fast at converging even without LFA. We use MX960s and MX80s.
> 
> I hope this helps.
> 
> Serge
> 
> 
> 
> ________________________________
> From: Amos Rosenboim <amos at oasis-tech.net<mailto:amos at oasis-tech.net>>
> To: juniper-nsp at puck.nether.net<mailto:juniper-nsp at puck.nether.net>
> Sent: Wednesday, December 21, 2011 2:34:22 PM
> Subject: [j-nsp] IP/MPLS fast convergence
> 
> Hello All,
> 
> I'm planning a greenfield IP/MPLS network for a mobile operator.
> The requirements are to support MPLS services (mainly L3 VPNs but also some VPLS), enforce  strict but fairly simple CoS model, and support fast convergence.
> No requirement for CSPF based TE.
> 
> Traditionally I'de set single hop RSVP LSPs (from access/edge) to core just for the sake of FRR, and tunnel LDP inside these LSPs.
> This way I would get FRR without the burden of full mesh RSVP LSPs.
> 
> However in the last two years I read more and more about LFA, IP/LDP FRR and similar technologies.
> 
> I'm considering to drop RSVP in favor of LFA and LDP, but was wondering if anyone is actually using this in the field, if so what is the impression.
> 
> Regards
> 
> Amos
> 
> 
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net<mailto:juniper-nsp at puck.nether.net>
> https://puck.nether.net/mailman/listinfo/juniper-nsp
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net<mailto:juniper-nsp at puck.nether.net>
> https://puck.nether.net/mailman/listinfo/juniper-nsp
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/juniper-nsp