[armedbear-devel] CLOS thread safety
dkirkman at ucsd.edu
Sat Apr 24 20:01:43 UTC 2010
On Sat, Apr 24, 2010 at 11:36 AM, Eric Marsden <eric.marsden at free.fr> wrote:
>>>>>> "dk" == David Kirkman <dkirkman at ucsd.edu> writes:
> dk> On my machine, a half dozen or so threads will reliably die with either:
> dk> Missing slot MOP::NAME
> On my machine (4-core AMD64 running Linux), I'm not able to
> reproduce the bug with the server JVM, but it triggers very easily
> when running in interpreter mode (java -Xint).
> OpenJDK / Java 1.6.0_18
I was using the client jvm (2-core macbook), java 1.5.0_22. I can also
trigger it the the server jvm, but possibly with lower frequency.
But in trying out the server jvm, I realized that that code fragment is less
reliable than I thought -- sometimes it generates dozens of errors, and
other times I can run it with no trouble ...
Note that 1000 threads takes a lot of memory -- if I start it with the default
jvm parameters I just thrash. I don't get errors, but it does not actually end
up computing anything.
On my machine, if it does not generate errors, it starts generating output
after about 10 sec, and is done withing a minute. If it generates errors,
everything slows down a lot -- it dos not ever seem to finish (just keeps
my cpus pegged at 100% until I get tired of waiting for it).
More information about the armedbear-devel