[bind10-dev] more about cache (Re: cache effectiveness)

JINMEI Tatuya / 神明達哉 jinmei at isc.org
Wed Mar 6 23:12:32 UTC 2013


At Wed, 6 Mar 2013 08:42:40 +0100,
Michal 'vorner' Vaner <michal.vaner at nic.cz> wrote:

> > In any case, I doubt the use of shared memory works as a silver
> > bullet.  We'll then just suffer from interprocess synchronization
> > problems instead of inter-thread ones.
> 
> There are no silver bullets, of course. The use of shared memory has the
> advantage we need to synchronize only the shared part. We don't have to worry
> about all the support libraries around, like libcc, libconfig, liblog. Are we
> sure if they are thread safe, or at least reentrant?

Okay, I see the point, but in that sense I think we already opened the
can of worms as we introduced threads in b10-auth; we cannot free from
that discussion anyway.  If this (the DNS record cache for the
resolver) is the only possible place we would use threads, it may be
more appealing to consider eliminating them in the cache too,
addressing or ideally eliminating performance drawbacks, but since
that's not the case, other tradeoffs should be more seriously
considered.

That said, I'm not necessarily opposed to considering the idea of
a dedicated cache process.  This is an opportunity to do something
completely new, so we can give any idea a try as long as our research
phase allows.

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.


More information about the bind10-dev mailing list