jdk-1.4.2p7 crashes with no indication of why

[Nicolas Gieczewski]

> > A couple of days ago I moved a very thread-intensive Java application
> > from a server running Linux to a server running FreeBSD 5.3-RELEASE and
> > jdk-1.4.2p7. The server has an Athlon XP 2600+ processor and 512 MB
> > of memory.
> > 
> > The application services around 700 concurrent clients during peak hours
> > and creates at least one thread for each connection, with many connections
> > requiring 2+ threads. The total number of threads during peak hours is
> > usually between 1600 and 2200.
> > 
> > The problem: During peak hours, the JVM dies several times an hour with
> > no indication of why. Nothing is written to stdout or stderr, there's no log
> > file, no core dump, and nothing in /var/log/messages.
> > 
>
> Its very difficult to have any clues since there isn't much information
> to be had :(.
> 
> Try updating to 5.3-STABLE as another poster has suggested.  If the
> problem persists, try using java_g instaead of java and maybe run that
> under gdb and see if you can get a trace.

I upgraded to 5.4-PRERELEASE yesterday and the problem persists.

I've noticed something that may be of interest, though: Until
yesterday, I had been running the JVM with the -Xmx256m parameter.
If I didn't, the JVM would run out of memory and threads would die with
java.lang.OutOfMemoryError exceptions. This happened on Linux as well,
and using -Xmx256m got rid of those errors. However, on FreeBSD,
-Xmx256m appears to be the cause of my random, silent crashes: If I
change it to, say, -Xmx128m, the JVM no longer crashes. But it does
run out of memory because 128 MB is not enough, so that's not an
acceptable solution. To summarize:

-Xmx256m: Fine on Linux, random silent crashes on FreeBSD.
-Xmx128m: java.lang.OutOfMemoryError exceptions on both OSs.

Does this help at all? By now I'm pretty much sure that the crashes I'm
experiencing are memory-management-related; otherwise they shouldn't
go away when changing the value of -Xmx.

Thanks,

Nick