Performance: BIND 8 vs BIND 9

[phn at icke-reklam.ipsec.nu]

Simon Waters <Simon at wretched.demon.co.uk> wrote:

> "Sasso, John IT" wrote:
>> 
>> I've been hearing second-hand that the performance of BIND 9 (w.r.t. #
>> requests/sec) is, well, terrible compared to BIND 8.  I wish I could give
>> specific revisions, but such was not given to me.  I have looked on the 'net
>> as well as Nominum's site for any papers or other info for BIND 8 vs BIND 9
>> benchmarks, but have not been able to find any.  I am told that BIND 8 will
>> do about 8000 reqs/sec, whereas BIND 9 fairs worse.
>> 
>> Is there any substantiation to these claims?

> Yes

> BIND 9 has lower maximum throughput. More recent versions of
> BIND 9 close the gap. BIND 9 also scales better with CPU as it
> is threaded (with BIND 8 you could run multiple instances on one
> machine, but that is wasteful of other resources), but you need
> two CPUs (more or less) to match the performance of 8.

> The actual requests per second depends on hardware you have. 

> In many real life situations DNS performance is limited by
> network bandwidth not server or OS resources, many big sites run
> DNS happily on a few low end SUN or Intel PC's running BIND
> without any significant load. 

> What sort of CPU and network utilisation do you see on your
> current nameservers? How many requests per second do you get at
> peak?

>>  We were considering on moving
>> to BIND 9.2 in the near future, but not if its going to degrade performance.

> Somethings are more important than performance.

>> Any web pages and/or papers you can refer me to regarding performance issues
>> and comparisons would be greatly appreciated!

> Rick Jones at HP's Labs (Cupertino) has written a number of
> reports. Including direct head to heads of BIND 8 to BIND 9 on
> identical hardware, and running MS DNS on Win32, versus BIND on
> GNU/Linux and HP-UX. (For the record MS DNS on Win32, beat BIND
> on Linux, which beat BIND of HP-UX, all on the same IA64 IIRC,
> which I think backs up my statement that there are more
> important things than performance nicely). 


Simon,
Do you have an URL to these tests ?




> He usually posts here when new ones are out, so check the
> archive.


-- 
Peter Håkanson         
        IPSec  Sverige      ( At Gothenburg Riverside )
           Sorry about my e-mail address, but i'm trying to keep spam out,
	   remove "icke-reklam" if you feel for mailing me. Thanx.