8.3 vs 8.4

[Jan Gyselinck]

On Sat, Dec 06, 2003 at 06:47:54AM +0000, Paul Vixie wrote:
> Jan Gyselinck <bind-users at lists.b0rken.net> writes:
> 
> > It's a fact that a dual CPU Sun with BIND9 performs equally well
> > (worse in some case, better in other) than a BIND8.  Only, it
> > requires the double amount of memory (if not more) and it uses
> > the two CPU's while BIND8 can only use one CPU.  What an
> > improvement.
> 
> we're working on that, thanks in part to a contract with the unix vendors.

Nice to hear that.

*snip* *snip* *snip*
> > No, that's not a show stopper.  But if there's no advantage in going
> > to BIND9 (other than support for the bleeding edge DNS additions),
> > the 'minor' pains are just an additional reason why not to migrate.
> 
> how about, people who have been in the source code for both systems, have
> absolutely refused to run BIND8 after BIND9 got stable (9.1 or so.)

I don't contest that.  But when one is tied to using certain hardware
and making sure things run, that's the least of my problems.
 
> > 1 GB RAM is not enough for BIND9 for a modest ISP.
> 
> so, we recently upgraded a sun in our lab from 256MB to 1.2GB, for $200.

Well, it's not an issue to get 2 GB in a server, the issue is that the
hardware can't handle running BIND (if 8) with more than 1 GB.
Oh, yeah, it's no bleeding edge hardware, I know.

> > A 'bit'?  Twice as fast is not what I call 'a bit'.  Makes you wonder,
> > BIND8, a product of coding and coding and fixing etc etc etc,
> > which makes some horrible source code, which is still faster than
> > BIND9, a product redesigned from the ground.
> 
> like i said, we're working on the performance delta.  it turns out that
> BIND9 has some unfortunate overgeneralities that were due to protocol
> elements like bitstring labels, which are no longer present, and DNSSEC.
> ripping these out and/or making them optional is taking some time, but
> i think in the first half of 2004 you should be pleasantly surprised.

Well, here's hoping ...
 
> > Obviously for a lot of us those extra features don't outweight the
> > performance loss from BIND9.
> 
> it's a rare system indeed that's running in the headroom delta from BIND8
> to BIND9.  in other words if you're running a BIND8 system at CPU loads
> which frequently exceed 40% utilization then you've got other problems.

Ofcourse, but we're talking mainly about resolvers here.  Those other
problems are broken stub resolvers or client programs (requesting 1000
of times the same thing, when it can't be resolved) and authorative
DNS servers that don't answer (the famous non-cached ServFail).
 
> AOL is now selling a pee cee for USD$299 that would run BIND9 just fine.

Oh, but they don't sell Suns?  A pitty.
(don't convince me, I don't make the decisions :/)
 
> we run BIND9 on f-root, which is one of the world's busiest name servers.

But it's an authorative.  It never has to go out to fetch results.
We do not have any issue with any of our authoratives.  Unless one has to
run a big zone on it.  Then the difference between BIND8 and 9 becomes
very clear again.  Do make some tests with the .com zone, you'll see.
 
> if you frame the question as "why is it so slow?" then my answer is, we're
> working on it.  if on the other hand you frame it as "i havn't got enough
> hardware to run something that slow" then my answer is "then you havn't
> got enough hardware to run anything else, either, please go spend $299 on
> a new AOL pee cee."

If it were me, the place would be filled by PC's ...  If only it were me
that could make that decision :/


Jan Gyselinck