ISC DNS performance test

[Thomas Klausner]

ISC did a DNS performance test:
http://new.isc.org/proj/dnsperf/ISC-TN-2008-1.html

The result was:
Linux Gentoo 2.6.20.7 93,000
Linux Fedora Core 2.6.20.7 87,000
FreeBSD-7-CURRENT 200708 84,000
FreeBSD-6-stable 200708 55,000
FreeBSD 6.2-RELEASE 51,000
Solaris-10 DevelExpr 5/07 50,000
NetBSD-4.0-Beta 200708 42,000
OpenBSD 4.1-snap-20070427 35,000
Windows 2003 Server 22,000
Windows XP Pro64 5.2.3790 SP2 20,000

Cheers,
Thomas

[Hubert Feyrer]

Post by Thomas Klausner
http://new.isc.org/proj/dnsperf/ISC-TN-2008-1.html

I think we've all seen it, and we'd also all be very interested to see
this repeated with very latest NetBSD-current -- too many things have
changed in NetBSD that have been left out here.


- Hubert

[Jim Wise]

Post by Hubert Feyrer

Post by Thomas Klausner
http://new.isc.org/proj/dnsperf/ISC-TN-2008-1.html

I think we've all seen it, and we'd also all be very interested to
see this repeated with very latest NetBSD-current -- too many things
have changed in NetBSD that have been left out here.

Unless I'm mistaken, all OS versions used were from their respective
release branches. Since I don't _think_ we're advising people to use
bleeding-edge -current for their production DNS servers, this seems
reasonable.

If NetBSD 5.x will be an improvement in this regard, it's reasonable
to ask for a rematch once 5.x is branched.

What are some of the changes since 4.0 which will make 5.x better in
this regard?

- --
Jim Wise
***@draga.com

[Jason Thorpe]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Post by Hubert Feyrer

Post by Thomas Klausner
http://new.isc.org/proj/dnsperf/ISC-TN-2008-1.html

I think we've all seen it, and we'd also all be very interested to
see this repeated with very latest NetBSD-current -- too many
things have changed in NetBSD that have been left out here.

Unless I'm mistaken, all OS versions used were from their respective
release branches.

You're mistaken:

FreeBSD-7-CURRENT 200708 84,000

-- thorpej

[Jim Wise]

Post by Thomas Klausner

Post by Jim Wise
Unless I'm mistaken, all OS versions used were from their
respective release branches.

FreeBSD-7-CURRENT 200708 84,000

Hmm. I guess I'm not very up with FreeBSD's naming scheme -- isn't
this the release branch for 7.x -- analagous to NetBSD 4-BETA, which
was also tested, but distinct from FreeBSD-CURRENT?

- --
Jim Wise
***@draga.com

[Alfred Perlstein]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Post by Thomas Klausner

Post by Jim Wise
Unless I'm mistaken, all OS versions used were from their
respective release branches.

FreeBSD-7-CURRENT 200708 84,000

Hmm. I guess I'm not very up with FreeBSD's naming scheme -- isn't
this the release branch for 7.x -- analagous to NetBSD 4-BETA, which
was also tested, but distinct from FreeBSD-CURRENT?

Typically something -CURRENT means "top of tree", but of course
you do need to check the date because eventually 7-CURRENT becomes
7-RELEASE and subsequently 7-STABLE.

[Andrew Doran]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Post by Thomas Klausner

Post by Jim Wise
Unless I'm mistaken, all OS versions used were from their
respective release branches.

FreeBSD-7-CURRENT 200708 84,000

Hmm. I guess I'm not very up with FreeBSD's naming scheme -- isn't
this the release branch for 7.x -- analagous to NetBSD 4-BETA, which
was also tested, but distinct from FreeBSD-CURRENT?

I expect NetBSD-current would do better on this test than 4.0, although
there is still a lot we could do to improve the numbers on this particular
benchmark.

As far as I am aware an answering bind server communicates with the outside
world using a single UDP socket, which works somewhat in NetBSD's favour
given that our IP stack is single threaded at the moment.

Cheers,
Andrew