[quagga-users 14273] Re: OSPFv3 30 second data path switching

Lennart Sorensen lsorense at csclub.uwaterloo.ca
Mon Mar 28 03:09:44 BST 2016


On Sun, Mar 27, 2016 at 08:55:36PM -0400, Hristo Asenov wrote:
> Hello Quagga Community,
> 
> A couple of months ago I needed to perform a test with ospf6d, where a router is connected to a host through two redundant router links. I simulated a router failure at the best path and waited how long it takes for the OSPF6 protocol to detect the failure and switch over packet forwarding to the second path. The version of Zebra and OSPF6 we used is 0.99.17. We did not change the hello interval (which by default is 10 seconds) and left the queue hold-down timer in Zebra set to 500 msec. I calculated the time to switch over packet forwarding in Linux Data Plane to be about 30 seconds. Could anyone help me understand where the extra 20 seconds delay is coming from?
> 
> I understand that there is some delay in SPF recalculation and installing new route through Netlink into kernel, but don’t know if this latency is long enough to justify the extra 20 seconds. I also performed the same test in GNS3 on Cisco image, and the data plane switchover delay there was about 7 seconds.
> 
> Greatly appreciate any help on insight on this issue,

The dead interval is 3 times the hello interval.  After all a packet
could be lost and you don't want the network to make drastic changes
due to one dropped packet.

-- 
Len Sorensen



More information about the Quagga-users mailing list