cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
2895
Views
5
Helpful
3
Replies

OSPF Best Practice for Keep Alive timers

George-Sl
Level 1
Level 1

Hello,

 

If I use Nexus 6ks and 5ks with 2k FEX, and SFP-10G x2 Port-channels, how should I find out how much bandwidth I am gonna lose from my links for hello intervals? and how low I can go for hello intervals for the fastest convergence?

 

thanks

 

2 Accepted Solutions

Accepted Solutions

Filip Knezevic
Level 1
Level 1

Well, timing is more up to you. I don;t think there is a specific formula except that the ration for timers is 4:1. You can reduce the timers to allow faster convergence, but in case the link is unstable it will flap much more with lower timers. 

So if you have a stable link, you can go with, for example, 4:1 instead of 40:10 for broadcast networks.

But if you want faster convergence, vest practice is to use BFD.

View solution in original post

Joseph W. Doherty
Hall of Fame
Hall of Fame

First, just curious, what's the chance of the port-channel links failing without hardware signalling port down?

With my prior employer, we found using hello interval of one second didn't seem to cause any issues, and although still "slow", its was 10x better than the default.

On devices that supported BFD, we used 250ms x4, i.e. 4x better than the above.

Normally we expected hardware to take the port down, so the hello timers were "just in case".

On devices that losing a link wouldn't trigger a port down, and which didn't support BFD (e.g. tunnels), I used, if supported, fast OSPF hellos, 500 ms x4.

On any of the above, especially with gig or faster LAN links, high frequency hellos weren't a bandwidth issue.

View solution in original post

3 Replies 3

Filip Knezevic
Level 1
Level 1

Well, timing is more up to you. I don;t think there is a specific formula except that the ration for timers is 4:1. You can reduce the timers to allow faster convergence, but in case the link is unstable it will flap much more with lower timers. 

So if you have a stable link, you can go with, for example, 4:1 instead of 40:10 for broadcast networks.

But if you want faster convergence, vest practice is to use BFD.

Joseph W. Doherty
Hall of Fame
Hall of Fame

First, just curious, what's the chance of the port-channel links failing without hardware signalling port down?

With my prior employer, we found using hello interval of one second didn't seem to cause any issues, and although still "slow", its was 10x better than the default.

On devices that supported BFD, we used 250ms x4, i.e. 4x better than the above.

Normally we expected hardware to take the port down, so the hello timers were "just in case".

On devices that losing a link wouldn't trigger a port down, and which didn't support BFD (e.g. tunnels), I used, if supported, fast OSPF hellos, 500 ms x4.

On any of the above, especially with gig or faster LAN links, high frequency hellos weren't a bandwidth issue.

Review Cisco Networking for a $25 gift card