[kea-dev] Good news and bad news

Francis Dupont fdupont at isc.org
Mon Dec 3 16:20:42 UTC 2018


Thomas Markwalder writes:
> On 11/30/18 7:12 AM, Francis Dupont wrote:
> > It is a known defect
> And where is it written down?

=> nowhere I am afraid...

> > and the fact it was not reproduce in the threaded
> > receiver is not an accident.
> 
> Nor did I suggest that it was, well done for doing it that way, Francis.

=> I tried to write code which is more fair (easy at the original is
not fair at all...). I remember to have taken a short time to see how
to organize the loop the best way.

> >   Note in the real world it should not happen
> > or with other words do not expect an overloaded server to perform well.
> 
> It can happen in the real world, and I suspect the more interfaces one
> has the worse it becomes.

=> linear list is already an issue when you have a lot of interfaces.
I saw some kernel code to be heavily modified for this reason so it is
something which can happen for Kea. Note the select vs poll/kqueue
tradeoff is of the same kind: select does not scale.

> and while the starvation was not as pronounced, it is still present:
not fair => someone is starving...

I don't know if we received complains about this issue?

Regards

Francis Dupont <fdupont at isc.org>


More information about the kea-dev mailing list