There is an aspect of Zope that drove our team into using Webware--if there
were two simultaneous long-running requests, Zope seemed to block or queue
additional web requests until one of the two completed. Our application has
a need to "shell out" to processes that may take 30-45 seconds to complete,
and we'd still like web requests to get processed while two or more users
have such processes running. Webware is handling this situation just fine.
Prior to adopting Webware, I ran the Apache Benchmarker against a stock
Webware install, with two simple Oracle queries in each page. I didn't have
any trouble, or even much of a slowdown, with 100 simultaneous clients doing
a total of 10000 requests, each hitting Oracle twice. (Note that this was
using DBPool, which significantly improved performance because of Oracle's
high cost of setting up a connection.)
We have had some occurrences of threads seeming to wedge (mentioned just
recently on the list), and we're trying to figure out why. But in general,
Webware is performing very nicely for us. (Caveat: We're not fielded yet,
but we've got about 10 developers hammering with automated unit
tests--against their own instances of Webware/WebKit.)
As others have noted, this is not proof, just anecdotal evidence.
David Hancock | dhancock@... | 410-266-4384
From: Geoff Gerrietts [mailto:geoff@...]
Sent: Wednesday, March 05, 2003 5:06 PM
To: Geoffrey Talvola
Subject: Re: [Webware-discuss] help evaluating WebWare
Quoting Geoffrey Talvola (gtalvola@...):
> What kind of hits per second are you getting on the Zope servers during a
> peak hour?
Hits per second is pretty low -- between 1 and 3 per server. The pages
are very dynamic, though, and response time varies from almost
instantaneous (1 second or less) up to 20, depending on the page and
the current load. Some of these delays are expected to get no faster
than 3-5 seconds turnaround, because of reliance on external data
sources -- but they start piling up when you're serving a couple
requests a second and the requests take a few seconds to complete.
> If your bottleneck is Zope's CPU usage, then I'm quite confident that
> Webware will make a huge improvement. It is a much more lightweight
> architecture, and your servlets are just python code, so they can pretty
> much run as fast as any Python code, whereas Zope has a lot of machinery
> layered on top of Python so it ends up being much slower, at least when
> you're using DTML. Note that PSP (included with Webware) and Cheetah (a
> separate package) are templating solutions for Webware that compile their
> templates down to Python code, so if you use them, you'll still get
I think I'm confident Webware would make a huge improvement, too. And
numbers will decidedly help. But who's got the biggest app (especially
most dynamic) who's using Webware? Is there any site I can point at
and say "these guys do something similar to what we're doing, and
they're having success!" I think there's a sense among some of our
folks that we got burned by trying to be "first" using Zope and ILU,
and they don't want to be first.
> I know I've seen some benchmarks on this list before, but I can't
> find them now. Is there any reasonable way to search the archives?
If you've got some good keywords, I'll download the whole archive and
search it with Mutt.
> I'm not sure that anyone has yet implemented a high volume
> application like this using Webware. There's certainly no reason
> why it shouldn't work fine.
Again, it's more about how can I sell it to others, on a short
timetable. I can build a sample app, but I can't build a sample app
that will prove the case.
Geoff Gerrietts "By doing just a little every day, I can gradually
let the task completely overwhelm me."
<geoff at gerrietts net> --Ashleigh Brilliant
This SF.net email is sponsored by: Etnus, makers of TotalView, The debugger
for complex code. Debugging C/C++ programs can leave you feeling lost and
disoriented. TotalView can help you find your way. Available on major UNIX
and Linux platforms. Try it free. http://www.etnus.com
Webware-discuss mailing list