BIND 8.3.0 is now available

Nate Campi nate at wired.com
Sat Jan 12 05:44:41 UTC 2002


On Fri, Jan 11, 2002 at 12:47:11PM -0800, Nate Campi wrote:
> 
> On Sat, Jan 12, 2002 at 12:00:28AM +1100, Mark_Andrews at isc.org wrote:
> > 
> > 			BIND 8.3.0 Release
> > 
> > BIND 8.3.0 is the first release of 8.3 and contains new features not found
> > in 8.2.5.
> > 
> > Highlights vs. BIND 8.2.5:
> > 
> > 	EDNS0 support.
> 
> So has anyone done any load testing to see if the EDNS0 checks make
> 8.3.0 as slow as BIND 9 for recursive queries outside it's authority
> (outside records)?
> 
> If not, I'll be doing this later today. I have production resolvers that
> *cannot* perform as poorly as BIND 9 does. I really hope 8.3.0 isn't as
> bad.

Here's the load testing I promised. It appears that the EDNS0 checks in
8.3.0 don't adversely affect it, though it's still not as fast a
resolver as dnscache:

  BIND 8.3.0 on localhost (dual PIII solaris x86, 1 gig RAM):
  
   queryperf -q 200
  
  Statistics:
  
    Parse input file:     once
    Ended due to:         reaching end of file
  
    Queries sent:         49827 queries
    Queries completed:    48038 queries
    Queries lost:         1789 queries
  
    Percentage completed:  96.41%
    Percentage lost:        3.59%
  
    Started at:           Fri Jan 11 21:20:48 2002
    Finished at:          Fri Jan 11 21:23:07 2002
    Ran for:              139.289560 seconds
  
    Queries per second:   344.878683 qps
  
  
  BIND 8.2.5 on remote host (dual PIII linux x86, 1 gig RAM):
  
   queryperf -q 200
  
  Statistics:
  
    Parse input file:     once
    Ended due to:         reaching end of file
  
    Queries sent:         49827 queries
    Queries completed:    46765 queries
    Queries lost:         3062 queries
  
    Percentage completed:  93.85%
    Percentage lost:        6.15%
  
    Started at:           Fri Jan 11 21:16:44 2002
    Finished at:          Fri Jan 11 21:19:13 2002
    Ran for:              149.779228 seconds
  
    Queries per second:   312.226205 qps
  
   
  BIND 8.2.5 on remote host (single proc sparc netra t1 105, 256 RAM,
  solaris):
  
   queryperf -q 200
  
  Statistics:
  
    Parse input file:     once
    Ended due to:         reaching end of file
  
    Queries sent:         49827 queries
    Queries completed:    48749 queries
    Queries lost:         1078 queries
  
    Percentage completed:  97.84%
    Percentage lost:        2.16%
  
    Started at:           Fri Jan 11 21:14:31 2002
    Finished at:          Fri Jan 11 21:15:59 2002
    Ran for:              87.994650 seconds
  
    Queries per second:   553.999590 qps
  
  dnscache on localhost (same dual PIII solaris x86, 1 gig RAM):
   
   queryperf -q 200
  
  Statistics:
  
    Parse input file:     once
    Ended due to:         reaching end of file
  
    Queries sent:         49827 queries
    Queries completed:    49329 queries
    Queries lost:         498 queries
  
    Percentage completed:  99.00%
    Percentage lost:        1.00%
  
    Started at:           Fri Jan 11 21:24:39 2002
    Finished at:          Fri Jan 11 21:25:50 2002
    Ran for:              71.166479 seconds
  
    Queries per second:   693.149369 qps

Hosts marked as "remote" are in the same datacenter, separated only by
a few switches. The records used were hostnames pulled from (resolved)
web logs, about one thousand of them repeated 50 times to make a file
with 50,000 records to look up. This tests the initial resolution and 
subsequent service from the cache.

Each nameserver was stopped and started before testing to clear the
cache (without a doubt). I had originally installed dnscache on this
host for testing but left it in place when I couldn't find anything else
that performed as well. It is now our primary production resolver for a
busy datacenter.
-- 
Nate Campi | Terra Lycos DNS | WiReD UNIX Operations

Mount attack on Osama Bin Laden 
 # mount /attack osama:/bin/laden 



More information about the bind-users mailing list