Volano (makers of Volano Chat) is a company that is naturally interested
in servers that support lots of simultaneous threads. For several years
they've been publishing their benchmark results on the abilities of
various OS+JVM combinations to scale up to thread counts in the thousands.
There are some *very* significant (order of magnitude) differences between
combinations.
The report makes pretty interesting reading. Most recent edition was May
29, 2001:
http://www.volano.com/benchmarks.html
Craig McClanahan
On Tue, 14 Aug 2001, Pier P. Fumagalli wrote:
> Santosh Pasi at [EMAIL PROTECTED] wrote:
>
> > Hi Jeff,
> >
> > Thanks for your reply. And good link, really useful.
> >
> > Here are results of our troubleshoot :(
> > On PIII 600, 256 Mb RAM, IDE Harddisk, RedHat Linux -
> > total no. of max. connection 990-1000 and finally tomcat crashes. But
> > still apache, ldap and other applications are still running perfectly ok.
>
> You'll never handle 10.000 connections on this thing... Get a decent
> operating system and more ram :). If you're using native threads, there's a
> limit of processes the kernel can spawn (and since threads are processes on
> linux, you might be hitting that limit), and a limit on the number of file
> descriptors open system-wide (and that's the other one you might be
> hitting), or file descriptors open per-process (1024, so pretty close to
> what you're observing, I believe that's it).
>
> > On RS/6000, AIX 4.3, 1000Gb RAM, etc
> > Total no. of max. connection 1600-2000 and finally reponse time
> > increases specially from tomcat. But still apache, ldap and other
> > applications are still running perfectly ok.
>
> Dunno anything about AIX... But if response time increases, I believe it's
> because you're using green threads, and the JVM slice takes more time to
> figure out what green thread is active and what is not, rather than serving
> requests... Try native threads...
>
> Pier
>
>
