bind/os benchmarking tool

[Marc.Thach at radianz.com]

Jacob,
I don't think this dig approach that John suggests would get anything
meaningful.  Matt Simerson (see archives) did some benchmarking earlier
this year and documented his methodology, however I think what he did is
overkill for your interest.
Ask yourself this:  "What sort of performance do I expect/require?  What
sort of headroom (extra capacity) do I feel comfortable with?"
If the answer gives you less than a few hundred queries per second then
Matt's results imply that with a modern Intel platform you'll be OK.
You cannot validate the opinions on the performance of an IP stack by
measuring BIND performance.  I'd opt for routing performance myself, or raw
connection speed to a null socket or such like, in the latter case you will
need multiple connections, esp if you want TCP.
As for routing performance, I performed some benchmarks using FreeBSD and
Linux on the same platform a couple of years ago.  The PC was the same and
the NICs used were the same, but the drivers for the NIC were not the same
source code between the two OS.  FreeBSD was about 15 percent faster than
Linux routing between two interfaces.  I did not look at the Linux driver
code, but I did have reason to examine and alter the FreeBSD driver.  It
failed to take into consideration certain modes of behaviour of the NIC and
thus might have be prone to lockups at high bandwidth or over-large
Ethernet frames.  Perhaps the Linux driver was written better and was thus
slower, who knows?  Perhaps the BSD networking code is better.  For that
percentage variation, who cares?  Note also that I cannot remember what the
PC was but it was certainly no more than a PII-266 (it may even have been a
P90), and it peaked out around 70000 pps, so clearly given Matts figures of
(IIRC) 1000-2000 queries/s for a much more powerful PC, stack performance
way exceeds BIND performance.
Choice of OS for your DNS server should IMO therefore be based on factors
such as reliability, security, manageability, and support, rather than on
any small performance differences.
rgds
Marc TXK
________________________________________________________________________
The views expressed are personal and do not necessarily reflect those of
the organisation providing the mail address from which this message was
sent



                                                                                                                    
                    "john mcCaig"                                                                                   
                    <jmccaig at cobalt        To:     comp-protocols-dns-bind at moderators.isc.org                       
                    group.com>             cc:                                                                      
                    Sent by:               Subject:     Re: bind/os benchmarking tool                               
                    bind-users-boun                                                                                 
                    ce at isc.org                                                                                      
                                                                                                                    
                                                                                                                    
                    30/08/2001                                                                                      
                    20:53                                                                                           
                                                                                                                    
                                                                                                                    




You could write a simple shell script using "dig."

Dig gives you the total query time and you can ask for all sorts of records
including zone tranfers with it.

Good luck!

john

<jacob at freecommunity.com> wrote in message
news:20000920010747.J5432 at freecommunity.com...
>
> I would like to run a comparitive test between bind on freebsd and
> bind on linux.  I've searched for dns benchmarking tools and havent
> found any.
>
> I realize there are many environmental factors that affect bind's
> performance and that any benchmarking tool I could write would not
> adequately describe 95% of the real nameservers in operation.  What I
> am looking to establish is whether or not I can validate the opinion
> that BSD-derived kernels have a "better" ip stack and that they
> handle high load (a very general term I realize) better than other
> kernels.
>
> I have to decide between BSD and linux for this setup and I would like
> to know if there is a glaring difference between the two.  At this
> point I don't know.
>
> First thought would be a perl script to generate a large number of
> dummy zonefiles with a large number of records in them.  Then a
> threaded perl script to make a bunch of random requests at specified
> rates, calculating average response time at various load levels.
>
> You could turn up the volume of requests over time and see how things
> change.  If all other things are equal (same hardware, no other
> software runnning) it seems to me this would reveal if there was a big
> difference between the two.
>
> Does this sound reasonable?  Are there already tools available that
> I just haven't found that can do this?
>
> Thanks!
>
> jacob martinson
>
>
>
>