DNS Performance Numbers
ohartman at zedat.fu-berlin.de
Wed Nov 22 09:19:58 PST 2006
Mark Bucciarelli wrote:
> On Wed, Nov 22, 2006 at 04:43:49PM +0100, O. Hartmann wrote:
>> Mark Bucciarelli wrote:
>>> On Wed, Nov 22, 2006 at 11:52:47AM -0200, Marcelo Gardini do Amaral wrote:
>>>> The results were discussed in the following threads:
>>> I see the speed differences are major, but don't have a good
>>> idea of what 15,000 DNS queries per second means. Is the
>>> following interpretation correct?
>> Means: as I see the 'numbers', they roughly show the
>> performance of an untuned TCP/IP stack and/or of the speed the
>> kernel can handle the 'subject of being tested for'.
>> My purposes for the specific OS may be not very common and not that
>> specific to DNS.
> I'm going to build a DNS server in the next couple weeks, and the
> difference between 6.1 and 4.11 in Marcelo's tests was dramatic.
> Are you saying Marcelo's test results do not apply to my
No, I complain about the dramatic performance drainage of FreeBSD and
would also say, that my main purpose for an UNIX driven box isn't the
service for network like routing, DNS and others. But I'm frightened by
the poor network performance when I have MPI in vi>
ew, for cheaper clusters GbE is still the common way of linking nodes
and this will be my 'bottleneck'. So performance of the above shown
tests may have the same dramatic impact to my purposes as to yours, b>
ut I do not know exactly. It is only a guess and it should in both cases
be subject of more investigations.
>> i did and do not care about 10% or 15% of > performance gain or
>> drainage if I could stay with the familiar OS
> I agree 100% with this idea, and that is why I am trying to get a
> better gut sense if this performance difference really matters
> for my application (web and email servers).
In some aspects I think this test we saw lead to some major problems and
I would guess if the performance drainage of the DNS/network service is
substantl, there will also be a overall drainage of performance, not
even in one particular application aspect. Look at BIND and NDS, do they
really suffer from the same systematic problem? If not, the performance
drainage comapring UP/SMP and 4.11/6.X seems to be representative.
Changing to Linux is an option, but the last one, because it will end a
ten years 'friendship' with Berkeley UNIXes.
Ok, don't want to be offending or a troll, stopping here ...
Serious: I miss performance tests, test suits on which some operating
system performance competitions could be made. In the past nearly every
new release of any operating system was benchmarked but nowadays this
seems to be subject of the specific persons who need to make decissions
about the OS choice.
More information about the freebsd-performance