Hello,
there are mainly two contributions to FRR convergence time as far as I know.
First you have to detect the failure, second the router switches over to the predefined backup LSP.
The second portion is only a local rewrite of the LFIB and should be practically instantaneous.
The first one - failure detection - is the major contribution. In SONET/SDH a failure condition in the optical network is propagated in certain bytes in the SONET/SDH frame header, thus it is rather quick to detect failures.
With ethernet that might be different. Assume two routers connected through a LAN switch. In this case a link-down event in one router will only be detected by means of keepalives. Those might be your routing hellos or in MPLS TE also RSVP messages. With standard timers that means several seconds.
In case you have back-to-back fibers then the convergence time will be considerably lower. You should be able to reach less than a second, but I have not measured this myself - this is "second hand" information.
Another option to look at: f.e. in IS-IS we can lower keepalive timers to less than a second and hold timer to one second. This should be designed carefully not to introduce unwanted instability into your routing.
In any case you cannot reach 50ms IP routing convergence time.
The question should really be, which convergence time is acceptable in your environment, i.e. which applications require less than a second outage and whether it will justify the efforts.
Hope this helps! Please rate all posts.
Regards, Martin
