[c-nsp] ISIS Fast Convergence (ASR920?)

[adamv0025 at netconsultings.com]

> Oliver Boehmer (oboehmer)
> Sent: Thursday, March 01, 2018 10:18 AM
> To: Jason Lixfeld; Cisco-nsp
> Subject: Re: [c-nsp] ISIS Fast Convergence (ASR920?)
> 
> 
>     Hey,
> 
>     There seem to be some conflicting suggestions for ISIS fast convergence
> timers, and I can’t seem to understand why that would be.  The former
> example is ISIS in a LFA FRR environment, the latter is from a general best
> practise guide. I can’t imagine LFA FRR or not would matter to the best
> practise would it?
> 
>     ASR920 + LFA FRR[1]:
> 
>     spf-interval 5 50 200
>     prc-interval 5 50 200
>     lsp-gen-interval 5 50 200
> 
> 
>     IE: XE 16.5 (Everest) (which can also run on an ASR920)[2]:
> 
>     spf-interval 5 1 50
>     prc-interval 5 1 50
>     lsp-gen-interval 5 1 50
> 
> 
> 
> SPF timers is generally a design decision, so the values above are just
> reflecting different design approaches: Choosing an initial wait of 1ms (the
> latter settings, i.e. spf-interval 5 1 50) tunes the network for optimal reaction
> for link failures, so routers will immediately start to re-route, with the risk of
> a subsequent SPF required if the failure was a node failure and additional LSP
> updates from other nodes are required to judge this.
> Hence a little more conservative setting uses 50msec initial wait, allowing
> more LSP updates to come in before the new SPF is calculated..
> 
> With LFA-FRR, core failures can be handled differently, so a little less
> aggressive initial-wait makes a lot of sense in this case..
> 
> Either way: even without LFA-FRR, difference between 1 and 50 msec is
> marginal and not noticeable in practice, so why bother much __
> 
Agree with Oliver.

With any FRR mechanism in place there's absolutely no point in tuning any SPF timers (lowering the timers will just put unnecessary strain on RE CPU).

Just would like to point out that in case of iBGP FRR it might be desirable to tune LSP generation and propagation (depending on your SLAs).
This is because the "PIC Core" can do its magic only once the ingress router is informed about the failure of the primary egress PE.
(and just recalled our old discussions with Oli on inter-as or hierarchical MPLS and how it imposes a finite limit on convergence times and how SR egress protection solves this -back then SR was just on paper).
Oli,
Just would like to ask whether there's equivalent of the Junos mpls egress protection (for RSVP).  

adam

netconsultings.com
::carrier-class solutions for the telecommunications industry::