On Mon, Nov 06, 2000 at 07:17:23AM -0800, Matt Brown wrote:
[snip]
> > In Apache, pre-forking is useful because it is one big fat whale.
> >
> > If you take a look at WN, for example (http://www.wnserver.org/), that
> > doesn't pre-fork, you'll see that it shows similar or better
> > performance.
>
> I know Dan has commented on this kind of thing before, and I'm sure
> he'd actually tested it. The gist of his comments as I remember was
> that fork() and exec() have gotten an unjustified bad reputation among
> UNIX programmers which they DO NOT DESERVE. Forking a small process
> is quite cheap, and exec is even cheaper.
Right now, fork() and exec() (funny how exec() actually doesn't exist :)
are the only calls that I trust to be portable over a wide range of
UNIXes.
> I think expending effort on an ugly hack of an optimization like this
> is poorly considered when there are far more fundamental causes of
> poor performance than this in qmail.
Very correct. Useful things would possibly be redesigning the scheduler
so that high concurrency could be reached much faster while injection is
still happening.
Even djb states (in THOUGHTS?) that there is room for improvement.
Greetz, Peter
--
dataloss networks
'/ignore-ance is bliss' - me
'Het leven is een stuiterbal, maar de mijne plakt aan t plafond!' - me
