Had another thought today: is garbage collection a factor?  Sun and IBM 
has spent a lot of time tuning the behavior of Java garbage collection, 
especially in server VMs.

I noticed after running a couple of ab -n 1024 -c 16 <url> tests on my 
PowerBook G4 (867Mhz) that free memory had gone to zero.  That makes me 
wander if some garbage collection might have been occurring during my 
last run (since it was much slower).

On Nov 20, 2003, at 5:46 PM, Michael Rueger wrote:

> Stephen Pair wrote:
>
>
>> excellent) on the Windows VM, I'd say there's something quirky in the 
>> Windows socket code.  On Unix, Comanche never has any failed requests
>
> Andreas and I had a lengthy discussion about that tonight, suspecting 
> my network code (once again).
> But, it might actually be a problem having to do with the semaphore 
> signal handling within the VM. Andreas will certainly have a more 
> detailed explanation (and a fix :-) for this soon :-)
>
> Michael
>
>
>