We've concluded that the HotSpot server mode is unusable on our RedHat 6.2
SMP system with Sun 1.3.1rc1. Performance tests that run in under 15
seconds in client mode are still running after half an hour with
-server. The test also throws occasional CreateExceptions that don't
happen with other JVMs.
It's hard to compare platforms, but under W98 and JDK 1.3.0 server mode
tests complete, but are noticeably slower than the default client JVM.
It'd be interesting to try this with the 2.4 kernel, but our single RH 7.1
machine is both a single CPU, and dedicated to running a java source
website, www.gjt.org. BTW I haven't run any throughput tests, but 7.1
feels *fast*.
Kirk Yarina
At 01:47 PM 4/20/01 -0700, you wrote:
>I *HAVE* seen problems with the HotSpot server under SMP with the Linux 6.2
>kernel.
>
>Under heavily loaded conditions, the JVM would 'pause' for somewhere in the
>range of 30 to 60 seconds. This would occur more frequently when the load
>was higher. After struggling with many potentially solutions it was
>eventually solved by running in interpreted more (-Xint).
>
>-----Original Message-----
>From: elephantwalker [mailto:[EMAIL PROTECTED]]
>Sent: Friday, April 20, 2001 9:05 AM
>To: Orion-Interest
>Subject: RE: smp w/jdk
>
>
>Erik,
>
>This is very ugly. I have two more support calls with Sun on my support
>contract. I will call them to see what's happening with this bug. How they
>could release a product that has such a serious bug is beyond me. There
>seems to be a workaround, though:
>
>in Bash, try setting the environmental variable as follows:
>_JAVA_SR_SIGNUM=16
>export _JAVA_SR_SIGNUM
>
>Regards,
>
>the elephantwalker
>
>-----Original Message-----
>From: [EMAIL PROTECTED]
>[mailto:[EMAIL PROTECTED]]On Behalf Of Erik Bogghed
>Sent: Friday, April 20, 2001 8:24 AM
>To: Orion-Interest
>Subject: RE: smp w/jdk
>
>
>
>My experience is that almost all 1.3.* variants (Sun, IBM, Blackdown)
>breaks under heavy load on Linux (Redhat 7). It's a recognized bug and
>is said to be fixed (even thought I found out that that was not the case)
>in 1.3.1 RC1.
>
>See http://developer.java.sun.com/developer/bugParade/bugs/4355769.html
>for some info about this.
>
>The only JVM I tested that did not break under heavy load was JRockit.
>To simulate the load I used several instances of Apache JMeter.
>
>I ended up using Blackdowns 1.2.2.
>
>Regards,
>
>Erik Bogghed
>
> > -----Original Message-----
> > From: [EMAIL PROTECTED]
> > [mailto:[EMAIL PROTECTED]]On Behalf Of elephantwalker
> > Sent: den 20 april 2001 06:07
> > To: Orion-Interest
> > Subject: smp w/jdk
> >
> >
> > I just noticed that the readme for Sun's 1.3.0.02 recommends against smp
> > kernels on Linux.
> >
> > Is that an issue? I haven't noticed a problem, had anybody else noticed a
> > problem with Linux and Sun's jvm with the smp kernel?
> >
> > Regards,
> >
> > The elephantwalker
> >
