[c-nsp] OSPF fast convergence

[Phil Mayers]

Walter Keen wrote:
> When redesigning an OSPF service provider network, (default values, with
> many gig-e links).  Aside from fixing link cost issues (100mbit is
> treated the same as gig-e at the moment) should I look at sub-second
> timers in OSPF 'ip ospf dead-timers minimal .....' Or BFD.  It looks
> like either would require an IOS upgrade, but I have seen lots of
> discussion about bugs in BFD.  This is only for core interfaces (all
> cisco 7600 series).  We'll be adding MPLS and iBGP on top of this after
> it's completed.

Common advice seems to be to make actual link-loss detection fast, in 
preference to using BFD. That said, I know some people use BFD.

Assuming you're using LAN cards, you may want to see if you can make 
router links as routed rather than SVI interfaces. Though routed 
interfaces are implemented internally as VLANs, presentations I saw from 
Cisco claim that this:

int G7/1
   ip address ...

...will detect link-loss (much) faster than this:

int Gi7/1
   switchport mode access
   switchport access vlan 300
int Vlan300
   ip address ...


Also, the OSPF process/SPF timers (as opposed to hello timers) are 
relevant for fast convergence (rather than link-loss). I did some 
research recently and concluded that, with a mostly-empty OSPF table 
i.e. bulk of routes in BGP, the following settings were both safe, and 
considerably "better" than the defaults:

router ospf 1
  ispf
  nsf
  timers throttle spf 10 100 5000
  timers throttle lsa all 10 100 5000
  timers lsa arrival 80

...again based on reading presentations from Cisco and others advice.


HTH