My experience with NT 4.0 was that it had a limit of 64 threads (some of which
were needed internally), which meant that the most I could use was approx. 58.
NT 4.0 is [was?] a time-slice OS rather than a full multi-process OS which had
specific affects on thread processing. Perhaps MS fixed this with NT 5.0 (i.e.
Windows 2000/XP) and later, perhaps not.
MY preferred solution was SunOS (i.e Solaris Intel) which has better threading
but no great incremental cost (as opposed to using one of your 220's which is
then not being used appropriately, plus possible substantial incremental cost).
Still, my recommendation would be to throw the best hardware you can spare
at the problem (i.e. if you have a Sparc or 220 free, use them instead).
Be aware that you will probably need to be bump up the socket limit on *nix,
which can be done as follows:
ulimit -n unlimited
If you are running as 'root' this will actually be 'unlimited' otherwise it will
be
the 'hard' limit as set at system configuration (see the 'ulimit' manpage for
more
info). The Apache people suggested including this command in the users'
startup script (i.e. .bashc, .cshrc, or whatever).
Let us know how this works out.
Regards
<quote>
I seemed to have hit a bottleneck with repect to the number of simultaneous
threads that can be generated using Jmeter1.7 (march 18 nightly build ) on my
clients Load testing workstation (win2K, 128MbRAM 800Mhz P3). Can't get past 50
threads without the machine locking up on me and/or my sampling times go from
milliseconds to minutes. I have tried a number of tricks learned both from past
experience and from reading news groups, mail archives, the jdk 1.3.1 docs etc.
So far the only noticeable improvement has come from bumping the heap size up
from the java command line. That is in fact how I got the number of threads up
from 40 to 50. If any of you java magicians out there have OS and/or java config
tricks up your sleeve that might help me get this number up to 100-500 threads
please let me know. In the meantime my next course of action is to try and
setup the jmeter server on the solaris 220s where the web servers live and just
using the testing station as a client. Al!
though this is a different test entirely, I would expect to see a dramatic
improvement in the number of simutaneous users Thanks. -David-
</quote>
--
To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
