Re: Java core dump size

From: Bob Bailin (72027.3605_at_compuserve.com)
Date: 10/27/04 

Date: Wed, 27 Oct 2004 20:12:03 GMT

"Jean-Pierre Radley" <jpr@jpr.com> wrote in message
news:20041027172454.GA21278@jpradley.jpr.com...
> Bob Bailin typed (on Wed, Oct 27, 2004 at 05:04:33PM +0000):
> | I'm trying to get the Java version of BackupEdge's edgemenu
> | program to work from an X-Windows session.
>
> I think Tom Podnar once wrote me that it just doesn't (yet) work on OSR 5.

Yeah, they're still trying. Not exactly a high priority project because
you can always get in thru a browser. But I am concerned that an
innocuous java applet can quickly bring a system to its knees by
dumping core and filling the root filesystem on a standard configuration.

>
> [ ... ]
>
> | Does Openserver 5.0.7 have a way to limit the size of core dumps?
> |
> | Why would the core file be larger than physical memory (512MB)?
> | Why would the virtual memory be at least 771MB for a Java app?
>
> AFAIK, core dumps are limited only by the ulimit.
>
> But you do have a lot of control over whether core dumps exist at all,
> what they are called, and even if they should all get written in a
> specified directory, e,g, /var/adm/coredumps. Read all about it in
> /etc/conf/pack.d/kernel/space.c, starting at line 344.

Wow, lots of extremely useful information, especially the ability
to uniquely name and locate your core files. No more periodic
find / -name core to clean up errant programs! (Although I was
surprised that certain gnu/java system library directories are unwisely
named 'core', presumably because they comprise the core files
of the product.)

I also see no reason not to send core files to a non-root fs, since
I can't restrict ulimit on this system.

Bob

Relevant Pages

  • workaround to start matlab on x86-64 OpenSuSE 10.2 linux
    ... Matlab R2006b would not start on our new Intel Core 2 Duo ... I found a workaround to replace the java version shipped ... cpu MHz: 1596.000 ... Interestingly, Matlab would work on the 4GB RAM machines, but not on ...
    (comp.soft-sys.matlab)
  • Re: jdk 1.3.1 in FreeBSD
    ... >And when I use gdb to trace the core file, ... ># gdb java java.core ... Segmentation fault. ...
    (freebsd-questions)
  • Re: Chucks plan
    ... The biggest issue is how fast you can suck information form external memory and move from core to core, but still we are looking Java performance on the realm of equivalent to 33mhz +, preferably 133mhz+. ... As far as speed goes, yes there would also be improvements, especially if compared interpreted, as Chuck selected combinations of x86 machine code instructions to match each coloforth instruction. ... ran it on an iTV chip and got horrid performance. ...
    (comp.lang.forth)
  • Re: JavaMail
    ... I take all your points, and it's just occured to me that Microsoft has the advantage of hindsight on some of these issues and can copy the nest bits from Java, but it also has crammed in a lot of functionality into ... I suppose one could ask that many of the Apache Commons projects could be incorported into the core framework, but yes, you do have to draw the line somewhere. ... But if I would have to make the decision I would go for a much smaller core, and a much better organized set of mandatory standard extensions, where I would require that a Java licensee has to implement each and every of them. ...
    (comp.lang.java.programmer)
  • Re: Stupidest game of the month. 11 move Mate.
    ... Just to make it tougher for Sanny's 3000 rated chess program... ... Core 2 Duo, the java applet is using both CPUs. ... single core chess program running as expected. ...
    (rec.games.chess.computer)