Avi Cherry wrote:
> > > > We're considering buying a dual processor machine to run our java
> > > > programs but if we can't get something running with a jdk on an smp
> > > > kernel there probably won't be much point.
>
> I've actually been running the Blackdown JDK 1.2.2RC4, with the JIT
> on a 2.2.14 SMP 2-CPU box for several months now without having any
> of these sorts of problems. I suggest that it's probably a issue
> with your configuration, rather than a bug in the JVM(s).
>
> You mentioned you were running 2.2.12? Try updating to 2.2.14. Also
> try updating to the latest glibc, since that has the pthreads code in
> it.
That was me that mentioned jitc problems on a 2.2.12 configuration (using
Blackdown JDK 1.2.2 RC4). Based on the following I don't think it's a kernel
/ library configuration problem:
I'm actually running a stock Redhat 6.1 kernel on a dual PIII 600 server
(2.2.12-20smp).
I did weeks of serious testing earlier in the year and had intermittent
problems with the sunwjit jitc.
At another list member's suggestion I tried running tests on the TYA jitc
over the weekend (with good results).
My test app does heavy i/o (JMS) and heavy XML parsing and generation, so
it's a good mix of CPU and i/o. There are around 20 threads and a few
heavily contended locks (so it has 'Bang'-like characteristics, but with a
server-app-like profile).
I ran for 42 hours continously at heavy loads using the TYA jit without any
problems (the test was manually terminated).
I then ran the identical configuration with sunwjit and got @ 5 hours at the
same load before the server JVM terminated abruptly. On a retry it
terminated within seconds. This is consistent with behavior from many
previous test runs.
I switched the jitc back to TYA and it's still running without a hitch. (On
my tests TYA is marginally faster than sunwjit as well).
In the past I've tested Borland's javacomp, but was unable to get it run on
a native threads SMP configuration for more than a few seconds (though,
being fair to Borland, that is not a supported / designed for
configuration).
I'm delighted with what I've seen of TYA so far. I've got a serious internal
testing cycle coming up later next month and will add TYA to the tested
configurations.
I have not done any JDK 1.3 / Hotspot testing, but am nervous about the beta
readme statement:
"Sun does support or recommend running the Java 2 SDK on SMP kernels."
Should that be "does support _and_ recommend" or "does _not_ support or
recommend"?
Has anyone got results from 1.3 testing on SMP yet?
--
David Marshall mailto:[EMAIL PROTECTED]
VM Systems, Inc. http://www.vmguys.com
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
