dnsperf and BIND memory consumption

[Vinny Abello]

> -----Original Message-----
> From: bind-users-bounce at isc.org [mailto:bind-users-bounce at isc.org] On
> Behalf Of Vinny Abello
> Sent: Thursday, August 07, 2008 12:58 AM
> To: JINMEI Tatuya / 神明達哉
> Cc: bind-users at isc.org
> Subject: RE: dnsperf and BIND memory consumption
>
> > -----Original Message-----
> > From: JINMEI Tatuya / 神明達哉 [mailto:Jinmei_Tatuya at isc.org]
> > Sent: Thursday, August 07, 2008 12:38 AM
> > To: Vinny Abello
> > Cc: bind-users at isc.org
> > Subject: Re: dnsperf and BIND memory consumption
> >
> > At Thu, 7 Aug 2008 00:26:04 -0400,
> > Vinny Abello <vinny at tellurian.com> wrote:
> >
> > > Huh... maybe I was right in the first place. I left dnsperf running
> > > and named ran out of memory. In my syslog I had a lot of these
> > > swap_pager_getswapspace failed messages followed by named finally
> > > dying (again, FreeBSD 7.0 STABLE AMD64, 4GB of RAM and the only
> > > software running is really BIND).
> >
> > Quick questions: did you enable threads?  If so, does that change if
> > you disable threads?  We've heard a similar report on a beta version
> > of 9.5.0 for FreeBSD, which reportedly only happened with enabling
> > threads (and only happened with 9.5, not 9.4).  I've tried to
> > reproduce it with a mostly equivalent setting of OS/hardware, but
> > never succeeded in seeing it by myself.
>
> OK. I've recompiled BIND 9.5.0-P2 (from ports) without threads enabled.
> I no longer see the memory leak at all. I'm running dnsperf and I see a
> constant of 18MB which is much more reasonable for what I am doing. For
> me it's easy to reproduce. Some more information that may help
> reproduce it:
>
> FreeBSD 7.0 STABLE AMD64 (cvsup'ed within the past week)
> BIND 9.5.0-P2 installed via ports with threads enabled
> Server is a Dell PowerEdge 2850 with 2 CPU's, Hyperthreading disabled,
> 4GB of RAM and a 36GB RAID1 array on a Perc4 controller (LSI MegaRAID
> chipset)
> Dnsperf run from a different server on the same network segment over
> Gig-E
>
> Interestingly, without threads I am seeing pretty much the same
> performance as with threads, but am only using one CPU and now have
> extra horsepower to spare. I know the maintainer of the BIND95 port on
> FreeBSD enabled threads by default, but I'm wondering if this should be
> reconsidered... that's more of an issue for a FreeBSD list, but if
> something is figured out, I can point the port maintainer to this
> thread (unless he's reading it already).
>
>
> I saw Danny state that Windows just had a fix made in regards to a
> memory leak specific to Windows socket code, but I'm wondering if what
> I was seeing was related to threads on Windows as well. I guess it's
> hard to tell without the fixed socket code to test with. I would
> recommend playing with dnsperf though to see if ISC can reproduce this
> with threads enabled.


One thing I am still confused about is why the performance of BIND is slower (when measured with dnsperf) when the test data is for authoritative data. I'm still seeing (with threads disabled now) non authoritative answers clocking in around 14k queries per second and authoritative answers at about 4k queries per second. Again, I tried localhost and I just tried the PTR record for 1.1.168.192.in-addr.arpa via queryperf. Dig confirms both are authoritative answers from my 9.5.0-P2 server (and I know they are based on the configuration), yet when doing queries for them I'm getting about 1/3 the performance vs. random lookups for non authoritative data. If anything I would think that data would be faster.

-Vinny