webserver performance (was Re: Why so few Python jobs? (and licenses))

[Alex Martelli]

Paul Rubin wrote:

> "Alex Martelli" <aleax at aleax.it> writes:
>> > These programs are all way too slow for a high-traffic site on the
>> > hardware we used back then.  Even with today's hardware, it would be
>> 
>> Have you actually _measured_ the performance and scalability of a web
>> server architecture based on asyncore/asynchat, versus more traditional
>> architectures based on multiple processes and/or multiple threads?  I
>> haven't done it systematically at all, but the few informal attempts
>> I did do have consistently amazed me.  Event-driven programming seems
>> to have extremely interesting characteristics in this sense.
> 
> Take a look at the benchmarks for thttpd and other servers, at
> www.acme.com, if you're interested in that question.  I haven't
> personally done measurements but Jef's look right to me.  We played
> with the Java web server and it was dog slow.

Thanks for the tip!  The exact URL is:
        http://www.acme.com/software/thttpd/benchmarks.html
and they do show that even-based ('select') servers leave all others in the 
dust (700+ to 1000+ requests per second on the test configuration, versus 
250 for Apache and 45 for architectures based on Java threads).  The test 
config is a 300 MHz UltraSparc, but I don't have a good feeling for what 
that implies in terms of overall machine performance.  They do make the 
point that any of these servers can saturate a T1 line when running on a 
Pentium-100, though:-) -- so unless you're serving on a fast LAN-based 
intranet using rather oldish hardware, server performance shouldn't be a 
crucial issue, should it?


Alex