Re: [babel] reliable route transfers

[Dave Taht <[e-mail]>]

On Tue, Mar 28, 2017 at 10:25 AM, Juliusz Chroboczek <[e-mail]> wrote:
>> Alternatively I was fiddling with using the "ack" message to try and
>> better rate limit route updates.
>
> Right.  The current reference implementation doesn't pace route updates --
> it sends a whole periodic route dump as a burst of as many packets as
> needed, rather than staggering updates across the update interval.  So
> even increasing the update interval will not alleviate the amount of data
> being sent as a burst.

Yes. basic tcp friendliness would be good here.... and burstyness to
some extent is also desirable for 802.11 aggregation.

> I'll see if I can hack together a solution on the plane.  It's not
> entirely obvious how to do it cleanly -- right now, the update deadline is
> a per-interface datum, it would need to become a per-(route,interface)
> pair one.

The point of confusion for me in the spec was that each route is
associated with an interval (currently) closely associated with the
hello timer.

There will always be a point (RFC970) at any bandwidth, where you have
more routes than you can get out in any given hello interval, and
either discarding them (intelligently), or pushing forward
(increasing) their potential update interval to something that can be
expressed within that bandwidth seem like answers, while still
adhering closely to delivering hello packets within their interval.

> Toke, how does your implementation schedule periodic updates?
>
> -- Juliusz



-- 
Dave Täht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org