On Apr 21, 2021, at 5:52 PM, David Farmer <[log in to unmask]> wrote:
>
> On Wed, Apr 21, 2021 at 5:36 PM Richard Laager <[log in to unmask] > wrote:
> On 4/21/21 3:04 PM, Frank Bulk wrote:
> > And to follow up on my previous question, is Arista falling bit short in
> > our situation, by not supporting a receive interval of 10 msec?
>
> I've had a couple vendors suggest not to make it that short. Brocade,
> for example, suggested 150 ms as a minimum. Arista was more vague, but
> from your error message, apparently their implementation doesn't even
> try to do less than 50 ms.
>
> Maybe think about this from another perspective, 10 ms is 100 times a second, 50 ms is 20 times a second, and 150 ms just over 6 times a second. I think 10 ms is probably being a little impatient.
Not the mention, the added CPU load on both ends dealing with said BFD packets 100x/sec.
We’ve generally seen 50-250ms used in practice. 10ms does seem super aggressive. We use 250ms x 3 here for backbone links and peers/transit that support BFD, and 750ms x 3 facing internal gear.
—
Andrew Hoyos
[log in to unmask]
