Re: [Idr] draft-ymbk-sidrops-rov-no-rr

[Job Snijders <[e-mail]>]

On Mon, Nov 15, 2021 at 02:07:37PM -0500, Jeffrey Haas wrote:
> On Mon, Nov 15, 2021 at 11:00:43AM -0800, Randy Bush wrote:
> > > this little hack is merely intended to warn and to allow
> > > constrained routers which can not keep full adj-rib-in to do rov
> > > with less damage.
> > 
> > sorry.  to be clear; in this day of lotso ram and many core cpus, it
> > would seem hard to justify not keeping the full adj-rib-in.
> 
> I'm not deeply following operational practices for the use of rpki-rtr
> right now.  Is there current operational wisdom about how often you
> permit your caches to churn?
> 
> If so, perhaps a quick comment in the draft that you can minimize
> impact of these older routers by minimizing the cache churn?

An interesting suggestion.

Some RPKI RP cache implementations already default to 'once an hour'. An
internet-draft WIP [5] suggests something along the lines of "between
once an hour and every 10 minutes", however the currently published RFC
(8182) says something that suggests along the lines of (rephrased) "poll
as frequent as every minute". There are quite some RP implementations
deployed in the wild which adhere to that published RFC, and
unfortunately have not yet been updated to implementation versions which
folow the suggested RPKI fetching slow down.

This phenomenon probably needs to be tackled via both backporting some
kind of fix to 'older hardware' (either the suggested hack, or enabling
"soft-reconfig inbound always"' by DEFAULT on older versions), and via
the RPKI RP cache implementations (but that now seems covered in all
current versions).

Kind regards,

Job

[1]: https://datatracker.ietf.org/doc/html/draft-ietf-sidrops-rpki-rov-timing
[8182]: https://datatracker.ietf.org/doc/html/rfc8182