Wired Intelligent Edge (Campus Switching and Routing)

Reply
Highlighted
Contributor I

Convergence time for BFD, OSPF and Directly connected

I am configuring OSPF routing for one custoemr. I am trying to minimize convergance time as much as possible and I am compairing convergance time between different options and whether BFD adds any value on top of OSPF.

 

1- For directly connected routers, is there eany added value for BFD? once the interface is down, its IP address will be removed rom routing table immediately (VLAN will be down as this is the only interface in this VLAN).

 

2- For in-directly connected routers (L2 devices inbetween), I couldnt find any advantage by configuring BFD. As per documentation (for 5400 and 8400), the minimum time to decalre the router down is 3 seconds while I can acheive less than this with OSPF timers.

 

This is from BFD documentaion in 8400

 

  • If the minimum time interval is set to 1000 milliseconds, then bdf detect-multiplier should be set to at least 3.
  • If bdf detect-multiplier is set to 1, then the minimum transmit interval should be set to at least 3000 milliseconds.

 

and this is from OSPF documentation where dead timer can be set to as low as 1 second

 

ip ospf all dead-interval 1-65535
Assigns the specified dead interval to all networks configured on the VLAN. (Operates the same as the ip
ospf dead-interval option, above.)
Default: 40 seconds; range 1–65535 seconds 

Highlighted
Contributor II

Re: Convergence time for BFD, OSPF and Directly connected

Hi,

 

I had the same question and searched the BFD timers on the latest release (10.04). This is what I found:

 

 

  • Exceeding a maximum of 20 BFD sessions with interval values of 300ms. Spurious sessions flaps will occur when the limit of sessions is exceeded.

  • Minimum intervals of 300ms are only compatible with the async_vxlan mode (BFD sessions across VxLAN) and is not user configurable.

  • Setting minimum transmit time interval between 500 ms and 1000 ms, and bfd detect-multiplier less than 3 might result in spurious flaps.

 

To avoud those spurious flaps, the minimum would be 3 (detect-multiplier) x 500ms (interval) = 1.5seconds for a state change.

 

The pros about using BFD are:

  1. it's agnostic to routing protocol, you can use it with OSPF, BGP, static routing...
  2. it has very little overhead meaning less CPU usage
  3. the most important one (for me): simplicity. When you configure BFD you won't need to adjust timers on a routing protocol that might break something else along the way. If you don't adjust those values right on your whole infrastructure, OSPF neighbors won't bring up adjacencies or even worse, you will have unpredictible behaviours

 

All that said, I believe that those timers will improve on future releases as the code matures.

 

*I forgot to add:

1- If for some reason that OSPF VLAN is on another port, that VLAN won't go down when the port that connects to the neighbour router goes down (ie. OSPF timers kick in).

Also if you have severa routers speakin OSPF inside the same VLAN, this applies as well.

 

 

Best regards,

 

Aarón

 

Highlighted
MVP Guru

Re: Convergence time for BFD, OSPF and Directly connected

Hello,

For directly connected nodes, there is no interest in using BFD.

For indirectly connected nodes, you can set min Tx interval to 100ms, which leads to 300ms switchover with detection multiplier being 3.

Highlighted
Contributor II

Re: Convergence time for BFD, OSPF and Directly connected

Hey Vincent,

 

for directly connected nodes using copper it won't matter but if you are using fiber optics it might.

 

On copper if the link fails it will bring ports down on both routers.

With FO if a single strand fails, router A will bring the port down but router B will still have light keeping the port UP.

Highlighted
Contributor II

Re: Convergence time for BFD, OSPF and Directly connected

Actually UDLD might be better for my last post.

 

The interval range goes from 200 to 90000ms * retries range is 3-10. The fastest you can go with UDLD is 600ms so this might be a better solution for fiber optic links.

Highlighted
MVP Guru

Re: Convergence time for BFD, OSPF and Directly connected

With 10G-SFP+ connectivity, UDLD is almost useless. Fiber strand cut detection is native in 10G standard. UDLD might still be useful to detect patching errors (very limited niche-case).

Highlighted
Contributor II

Re: Convergence time for BFD, OSPF and Directly connected

Thanks I didn't know that. About the timers, on your first answer you said you can use 100ms as minimum transmit interval.

 

Reading on the 10.04 release that's the case for 832x switches, for 8400, 6400 and 6300 the minimum is 500ms. Is there someway we can do subsecond convergence with the 6300 gear?

 

I'm on a project trying to see if this switches fit to our customer needs.

Highlighted

Re: Convergence time for BFD, OSPF and Directly connected

Also set BGP route update delay to 0, helps in many cases :) Also you can set connection retry time lower so BGP forms faster after the link comes up

Search Airheads
cancel
Showing results for 
Search instead for 
Did you mean: