rm at fabula.de
Fri Apr 22 03:50:18 CDT 2005
On Thu, Apr 21, 2005 at 09:59:28PM +0200, Marc Battyani wrote:
> <jeff at inf.ed.ac.uk> wrote:
> > Quoting Marc Battyani <marc.battyani at fractalconcept.com>:
> > > "Brian Mastenbrook" <bmastenb at cs.indiana.edu> wrote:
> > > > I find configuring apache to be a royal pain, and not
> > > > worth the added complexity.
> > >
> > > That you dislike Apache is your right but that is not enough to make
> > > based web servers a worse solution.
> > That some people don't mind involving Apache isn't enough to make
> > Apache-based web servers a better solution either. :)
> The major interest of apache is this:
> About 70% of the servers are using Apache. (that's 43174442 ones)
> Having a solution that integrates well into Apache is called pragmatism.
> That Apache+mod_lisp is the most performing solution to make a Lisp web
> application is just a bonus.
This is a worldwide study - in many areas Apache is even more dominant:
from a recent (HTML/XHTML-validity) study for german (the language) web-
sites: 86.5% Apache, 10.6 % IIS ....
> Anyway the last time I made some benchmarks, most of the other web servers
> like aserve were fast enough, being able to serve more several hundred
> pages/second (that's more than 360K pages/hour) which is quite enough. It
> shows that HTTP handling is largely fast enough.
> IIRC the mod_lisp tests with Apache 1.3 were at 440 pages/s (1.584M pages/h)
Most pure Lisp servers are probably more than fast enough for what they
are used for (small/medium sized websites with lots of dynamic content).
With my own private testings Araneida wasn't even close in magnitude to
a 'C-only' Apache (and _did_ have some stability issues that could result
in a DOS attack :-(
> > I'm disappointed that so many Lisp tools seem to require
> > Apache + mod_something.
> > Is the reason that people think the basic HTTP and socket levels
> > are uninteresting, or is there some horrible complexity there that
> > they've trying to avoid having to deal with?
> The complexity is very low. (I myself once wrote a small 1.0 server in just
> a few lines) That's why they are so many of them...
Hah! Yes, a more-or-less HTTP/1.0 compliant server is rather easy to
write (iff you stay away from encoding issues that crop up in
the HTTP header already). Most sites i use/maintain really require HTTP/1.1
and that's rather more demanding ...
None of the 'simple' Lisp servers i've seen so far did handle all of the
nitty-gritty details of HTTP (client-server content negogiation etc.).
> > With some of these systems, I'm left wondering what exactly they do,
> > if they've turned so much over to Apache.
> > It's not that hard to write an HTTP server from scratch. That's
> > what I did the last time I had to make a choice of this sort, and
> > and I'm not sure there's anything in more recent systems that would
> > lead me to make a different decision.
> > What I'd like to see is some Lisp-alone system that handled the
> > HTTP level very well, including cases where connections were kept
> > open (I didn't do all of that in my server), and provided an easy-
> > to-start-using structure on top of that. If it came with a CLOS-
> > compatible OO database, that would be ideal.
> > The Uncommon Web structure looks like it might be good, for example,
> > but I haven't examined it in detail. Unfortunately, it seems to want
> > something else to do the HTTP part.
> Sure, I told you it's boring... ;-)
Probably not - but a lot of work (and a lot of that will require careful
"standards reading" <shudder> - and then learning the hard way that some
of the standards are handled _very_ special by some important clients.
That special treatment of popular broken clients might be worth using
> When I need a small stand-alone server for desktop applications I use the
> paserve HTTP layer with my framework.
Yes, that seems to be a very good solution.
Cheers Ralf Mattes
> Lispweb mailing list
> Lispweb at red-bean.com
More information about the lispweb