[Zope] Zope performance too low ?

[Michel Pelletier]

> -----Original Message-----
> From: Christian Andreassen [mailto:christian@riddlers.com]
> Sent: Thursday, June 17, 1999 4:33 AM
> To: 'Brian Lloyd'; zope@zope.org
> Subject: RE: [Zope] Zope performance too low ?
> 
> 
> > I'm not that familiar with ab, but you don't give a lot of
> > information on your setup here (processor, memory, load
> > (discounting zope), publishing mechanism (pcgi? ZServer? mod_pcgi?),
> > details of the DTML used), which would be helpful to put this kind
> > of information in context.
> 
> I forgot. The server is a Pentium 200 with 96MB RAM, running Redhat
> Linux 5.2, Apache 1.3.4 and Zope 1.10.2. There was no other 
> significant
> load on the server when I did the tests. The publishing mechanism is
> pcgi.
> 

Hmm.. is the first request from the benchmark software the request that
fires up Zope?  Remember, this is a 5 second+ delay between the very
first request and it's response while Zope loads.  This can explain why
you got less than 3 requests per second.

Also, a 50 request trial is pretty short.

After seeing your mail, I ran some benchmarks of my own.  I have
distilled out everything but the requests per second figure, like you
did.  Note, that my benchamarks are a completely different setup than
yours, which probably helps to explain why I got 20 times the
performance you did.

Machine:  PII 350Mhz, 256M RAM
OS:  Red Hat Linux 5.2
Runtime:  No X Windows, one emacs, two shells, various daemons

free report before benchmarks:

             total       used       free     shared    buffers
cached
Mem:        257048     138332     118716      18432      44312
72200
-/+ buffers/cache:      21820     235228
Swap:       258744        228     258516

Standalone benchmark:

Description:
 Ran Zope 2.0 alpha 2 as: "python1.5.2 z2.py -d 216.164.72.2"
 Ran ab as: "./ab -n 1000 http://aldous:9673/test"

 The test document contains:

 <!--#var standard_html_header-->
 <h2><!--#var title_or_id--></h2>
 <p>
 This is the <!--#var id--> Document.
 </p>
 <!--#var standard_html_footer-->

 Benchmark output:

 Concurrency Level:      1
 Time taken for tests:   27.523 seconds
 Requests per second:    36.33
 Transfer rate:          10.79 kb/s received


Same as above, but with two concurrent requests:

 Ran ab as: "./ab -n 1000 -c 2 http://aldous:9673/test"

 Bechmark output:

 Concurrency Level:      2
 Time taken for tests:   26.026 seconds
 Requests per second:    38.42
 Transfer rate:          11.41 kb/s received


Same as above, with 4 concurrent requests:

 Concurrency Level:      4
 Time taken for tests:   26.738 seconds
 Requests per second:    37.40
 Transfer rate:          11.13 kb/s received


One problem with *both* of our benchmarks is that they are running on
the same machine in both cases.  Also, what are we benchmarking against?
Another 'application server'?  How would ASP compare?  or PHP?  How can
you compare two wildly different technologies (PHP has no
infrastructure, object model, object orientation, standard managment
interface, standard security model, acquisition etc. the list goes on
for miles)?  I suspect Zope would blow the doors off of any old CGI
script that has to spawn on every hit.

And requests per second is our only benchmark parameter, what about
speed vs. object database size?  ZODB is so well engineered I suspect a
benchmark in this area would be promising.  and how about DTML rendering
speed?  Or ZServer vs. PCGI?

What this proves, is that benchmarks are very strange things.  I trust
politicians more than I trust benchmarks, including my *own* benchmarks.

The reality of the situation comes from real life experience.  Zope has
been proven to scale extremly well, and will only get better.

> BTW, how is www.zope.org configured? Do you see the same 
> drastic difference
> between static pages and Zope pages there?
> 

No, the performance is quite nice.

> 
> Christian Andreassen: christian@riddlers.com
> 
> 
> _______________________________________________
> Zope maillist  -  Zope@zope.org
> http://www.zope.org/mailman/listinfo/zope
> 
> (For developer-specific issues, use the companion list,
> zope-dev@zope.org - http://www.zope.org/mailman/listinfo/zope-dev )
>