Hi!

>>Hi all,
>>
>>I have been looking through some of the capabilities of
> Comanche and I am
>>very impressed.  Seems like you can do anything in it that could
> do with
>>apache, but a much nicer/accessible interface.
>>
>>So my question is, what is the performance and scaling like.  I
> know the
>>common belief is that apache is unbeatable, but using a higher
> level
>>language it is possible to accomplish more:
>
> If I remember correctly from an old mail Comanche was around
> 1/3 of Apache for the number of transactions it can handle.

I did some benchmarking about 2 years ago but never got around to
publishing in a readable form (perhaps I posted about it) but from my
memory I recall:

1. I found some code here and there that I rewrote for speed. Never got
around releasing those changes. It was not fantastic improvements but
definitely worthy.

2. After the tweaks the bottleneck appeared to be header parsing. I bet
that code can also be improved.

3. Dan Shafer has been doing investigations on the lower levels, and if I
am not mistaken the Croqueteers have very recently discussed Socket issues
in the Unix VM (stumbled upon that) causing problems. I haven't myself
looked at this level and IIRC I used "keep alive" when testing which
probably hides a bit of the Socket issues.

4. Performance varies substantially with size of payload. The larger the
payload is the header parsing turns into a smaller issue.

5. Oh yeah, I only tested KomHttpServer itself with "in memory" payloads.
I never bothered with actual file serving which is a can of worms in
itself.

Since most of us have so much "other stuff" to spend our time on I haven't
had the time to revisit KomHttpServer tuning.

I have also felt that some parts of KomHttpServer is needlessly complex
(especially the whole module setup etc) - but I don't think it affects
performance in a negative way.

regards, Göran