Mike,

> Lets not forget that (last time I checked) there are some 5500+ lines of
> code in the Quixote package. Its not a fair comparison to make between a
> general framework like Quixote and say a custom, point solution that solves
> only a single problem - such as a CGI script that makes a database
> connection, does something and terminates.


I did not mean to discredit anyone's hard work. Quite the contrary is true. I
am thankful for this excellent peace of software. But what is the use of
an appserver that takes (possibly due to misconfiguration somewhere for which
I have to take the blame) 2 to 3 seconds to process a single request?

I would be glad if I could find the problem and solve it in order to achieve
similar results to yours, at least to process 2 to 3 requests per second
rather than to have to wait for 2 to 3 seconds for one request to be
processed.

There must be something strange in my setup. LWN via the internet, although
being a bit sluggish, is a hell of a lot faster than my setup via loopbak
networking...

Eckhard

> ab -n 2000 -c 8
>                     requests/second         transfer rate
> index.html       2877                           4487.94
> demo.cgi           12                             18.24
> demo.lrp          835                           1248.12
>
>
> Impact on the box:
> index.html - Running the test against straight html is very quick as you'd
> expect. It runs so fast that its difficult to gauge the effect on the
> machine except that its clear there is lots more capacity sitting in
> reserve. Load might be around .40 - .50 / 60-75% idle.
>
> demo.cgi - Load average on the box peaked at over 8.50 and the CPU's were
> 0% idle (mostly user)
>
> demo.lrp - Load about .30 - .50, CPU idle 35%, very approximately
>