[Seaside] Benchmarking

[Avi Bryant]

On Apr 12, 2005 12:04 PM, Avi Bryant <avi.bryant at gmail.com> wrote:
> On Apr 12, 2005 9:47 AM, Dmitry Dorofeev <dima-sender-3c337a at yasp.com> wrote:
> 
> > I run some tests about 6 months ago with Seaside Sushi Shop.
> > Seems like it is 1 or 2 page per second rate on a 1Ghz FreeBSD box.
> 
> Hm, that's a lot slower than the benchmarks I've done.  Maybe I need
> to do some more.

Benchmarking is a strange business; I'm never sure how much to trust
these numbers.  For example, the "ab" ApacheBench tool reports that it
can hit the /seaside/counter page on my machine in 16ms on average. 
JMeter, doing the same test, reports 200ms.  Where's that difference
coming from?

With that in mind: I just put together a little test script for the
Sushi Store with JMeter.  With one concurrent user, on my somewhat
poky (by server standards) 1Ghz Powerbook, I was getting an average of
300ms per request.  This scaled linearly, ie, with 10 users I got ~3s,
with 100 users I got ~30s.  However, the whole time I was doing this,
Squeak was never taking more than 30% of my CPU.  JMeter took another
30% or so, and the rest of the CPU was idle.

At any rate, these tools may not be very useful for absolute numbers,
but presumably they're still saying something useful about relative
numbers, so I'll see what further profiling I can do.

Avi