[Exim] Re: Caching DNS and Exim (Vadim Vygonets)

I just subscribe to the digest format and I noticed the little fray
about DNS caching and Exim. I thought I would contribute my two
cents...

We use exim to deliver over 6 million emails each day.

We have tried running BIND on dedicated machines, but the that was
problem (I'll get to that). We run a local cahcing-only version of bind
on each machine. We found that the hit we take from loosing the shared
cache of a single server is not worth the chance of that nameserver
acting up. We have had a lot of problems with BIND not responding
correctly.

AOL.COM no MX records? ;) Not what you want to see when you have 800000
messages to be delivered there...

Someone brought of context switching and thrashing due to that. We have
about 400 exim queue runners at any given time. If you look at the
nature of exim (and any MTA with similar design) causes A LOT of
blocking I/O which generates context switches anyway. So, who cares.

All of our emails are unique... So we do not have multiple recipients on
any message. So, this is a solid 6,000,000 messages at about 10GB
outgoing. Outgoing is the only time DNS caching matters, we shouldn't
consider incoming messages.

We are using about 40% of the capacity of our systems, so we should be
able to send about 18 million on our systems ;) We run caching
nameservers on the same boxes as exim.

6 in total soon to be scaled to 12 (for redundancy).

> Honestly, with 16 GB mail traffic a day, I doubt the extra lookups cost
> me much bandwidth. Further, it takes a much longer time to detect that
> a remote name server that you use as forwarder is down. That happens
> only rarely, but it can kill a busy system.

How much ingoing and outgoing.. The 16GB could be deceiving.. You should
look at quantity of messages not counting the size at all.

> So the answer probably is: It depends if a local DNS cache is useful.
> There is more to think about than just resource usage.

Really useful (we have tried it both ways). It works best for us to
have separate caches on each machine...

--
Theo Schlossnagle
33131B65/2047/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7