OSPF Fast convergence tuning

ashish-nagpal · ‎01-07-2010

Hi,

I need your suggestions over following details.

Brief Overview of network:
- Our network is pure level-3 level ISP backbone network. It is comprissed of mostly Cisco GSRs, 6500 & 7600 series devices.
- IGP used in network is OSPF only and all the links used in backbone network are 10Gige & 1 Gige Ethernet only.
- All edge and core routers in our network are receiving full internet routes through BGP but not redistributed in OSPF.
- There are multiple leaf areas conneced to backbone area and each area is comprised of almost 10-25 devices.
- There is huge database of OSPF and the timers used in OSPF are default.

Problem Description:
- Few days back we faced link flap multiple times on one of our backbone router, which caused reconvergence of entire OSPF database. This convergernce took a long while because of huge OSPF database.
- In order to achieve faster and stable convergence we are looking to fine tune OSPF timers using OSPF SPF/LSA throttling with default parameters.

Need suggestion, if it is recommended to implement OSPF SPF and LSA throttling in core backbone network of level-3 size ISP and what are the suggested values, which can be used when entire backbone is based on 1/10Gige ethernet links.

Regards

Ashish Nagpal

amit.bhagat · ‎01-07-2010

Hi Ashish,

Changing OSPF Hello and Dead intervals using ip ospf hello-interval and ip ospf dead-interval commands respectively, from every OSPF-enabled interface is one option. The range is 1 - 65535. But considering that your network is very huge, sounds like a lot of work to me (which is okay if you are paid on hourly-basis ) since the values should match for each OSPF neighbor to form adjacency.

To me, the OSPF areas should be designed such that these triggered updates are not send across the whole network. If possible, try to design the network such that some areas are made STUB. Also, try to summarise at OSPF ABR router.

Other ideas are most welcome.

HTH.

Regards,

Amit.

Giuseppe Larosa · ‎01-08-2010

Hello Amit,

what you say about using OSPF stub areas and area filtering is agreeable.

I had done some search on OSPF throttling on the duplicated post under WAN forum

https://supportforums.cisco.com/thread/1003218?tstart=30

the feature is interesting because makes OSPF self adaptive to rate of changes instead of having fixed hold time between SPF calculations.

As I wrote there, the self adapting behaviour can be a benefit and I think that values used are not critical the first two values can be low, with third timer high to protect from flapping.

If the database is big, it may be time to apply inter-area filtering, you may need to export only /32 from non zero areas to area 0.

This is a measure for scalability. These /32 cannot be summarized in an MPLS environment, if they represent PE routers but internal links in an area can be filtered.

Hope to help

Giuseppe

jcozzupoli · ‎01-08-2010

Hi Everyone,

Yes I agree with Giuseppe regarding the spf-flooding command. I implemented something similar in a MPLS VPN network I helped build. We were using IS-S as the IGP inside the MPLS domain, and we tweaked SPF to minimum (spf-interval 1 1 1) to achieve sub-second convergence.

HTH,

Joe.