[Seaside] HTTP Performance

[Avi Bryant]

On Nov 19, 2003, at 5:21 PM, Nevin Pratt wrote:

> Sven Van Caekenberghe wrote:
>>
>> That doesn't change the fact that you need a certain level of 
>> performance to be able to serve public sites. And IMHO the test was 
>> reasonably fair for all. The goal of the dynamic vs. static page was 
>> to get an idea of the overhead for dynamic pages.
>
> In my experience, a lowly 1.3 Ghz Pentium running Squeak, Comanche, 
> and Seaside can still overdrive a relatively fast DSL internet pipe.
>
> The pipe to the internet is usually the bottleneck.

I've been doing some benchmarking.  One thing that helped a lot for 
static pages was terminating the UI process (Project uiProcess 
terminate).  Note that this will (obviously) make your image 
unresponsive to the mouse and keyboard, but CPU usage goes down and web 
performance goes up.  I was able to get over 300 requests per second 
with a simple static page and Comanche on a 1Ghz G4 Powerbook.  I'd 
recommend doing this before deploying an image (although maybe there's 
a more reversible way of getting the same effect?).

This was using the unix VM (3.6g2).  For some odd reason, doing this on 
the Carbon VM (3.6.1beta3) had the opposite effect - performance 
dropped to 2 requests per second (!).  I'll have to look into this 
more.

I also noted that my current development branch of Seaside, where I've 
been doing some memory optimizations, also seems to have gained some 
speed - it's about 50% faster in some cases than the current release.  
I'm sure I can improve this further.