[j-nsp] Cisco IGP fast convergence

[Colby Barth]

All-

JUNOS has a few config options and we have interop'd in the field for sometime with Cisco's IGP FC config's:

SPF IW value defaults:

  IOS-XR default: IW: 50ms, 2ndaryW: 200ms, MaxW: 5000ms
  JUNOS defaults: IW: 200ms, MaxW: 5000ms, RRs: 3
The Cisco alg. is an exponentially increasing timer, whereas ours is not.  As an example, if a router receives a new LSP at time 0, Cisco will run a full SPF 50ms later.  If a 2nd LSP is rx'd, the router will not run another SPF computation for a minimum of 200ms.  If a 3rd LSP is rx'd then the next SPF IW timer is 400 msec and so on and so forth until the maximum wait achieved.
In the JUNOS implementation, we will run consecutive full SPFs on 200 ms intervals 3 times and then we wait the holddown interval before we run another full SPF.  Forcing the implementations to approach one another requires config on both sides.  The end result is you really want to tune the IW values to match, reduce the IOS-XR secondary wait timer and reduce both the JUNOS and IOS-XR holddown timers to approach SPF alignment.
JUNSO config:
spf-options {
    delay 50;
    holdtime 3000;
    rapid-runs 3;
    }

IOS-XR SPF configuration:
  spf-interval initial-wait 50 secondary-wait 100 maximum-wait 3200 level 2

I don't agree that you want the SPF IW to = 0.  In the case of SRLG's in the network you may want to have a bit of delay in there to try and capture all the incoming LSPs before running an SPF.  This is a personal preference and network dependent but would be my recommendation.  Your also gonna want to change the IOS-XR lsp generation timers.

IOS-XR SPF configuration
  spf-interval initial-wait 50 secondary-wait 100 maximum-wait 3200 level 2

This will align JUNOS and IOS-XR in that regard.

There are also changes recommended for cnsp intervals, LSP generation intervals & lsp intervals that require changes (mostly in IOS-XR) that you should consider.  Unicast me if you are interested in a thorough explanation and JUNOS and IOS-XR config's.

thanks,

-Colby


On May 29, 2012, at 2:38 PM, Saku Ytti wrote:

On (2012-05-29 07:08 -0700), Doug Hanks wrote:

Or just use BFD ...

Not really solution for SPF timers though. Juniper is clearly lacking in
SPF configuration options. What you operationally want is 0 delay for SPF
calculation to start, to cover typical event of single link going down as
rapidly as possible. (Sure for some topologies LFA will cover that)
But if LSPs keep coming in, you want to exponentially back-off, to run SPF
less and less frequently.

Same thing as interface up/down dampening, here also JunOS has just static
values, no exponential back-off.

Both cause you to choose rather conservative value in JunOS, while in IOS
you can rock aggressive values.

--
 ++ytti
_______________________________________________
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