Dimitrios Vyzovitis <[EMAIL PROTECTED]> writes:
>
> I have a 2xPIII-500 box and it kicks butt (both ibm and blackdown jdk).
> Go for it!We have had a lot of problems with native threads on SMP machines, for
all JDK's tested (Sun 1.1.x, Sun 1.2.x, and IBM). This includes JVM crashes
and mysterious behavior which we can't attribute to bugs in our own code.We are running fairly complicated applications with many threads and
"interesting" interactions between them. A lot of people may never run into
these problems because most multithreaded applications are not so complex.
Fair enough, but I guess that it heavily depends on the particular
application, its design and stress, and particular library (glibc) or vm
bugs you might bump into.
My experience has been positive for some tens of interacting threads
- but I don't know what happens when you stress it to hunderds or really
complicated apps (like the ones you refer to). It is still an interesting
experiment for me (and I'll probably figure out for myself in the next
few months, in which case
your words of caveat are more than welcome ;-)
The only way we've been able to achieve stability is to use Sun JDK 1.1.7v3Personally, I still advise for an smp box, even if you end up using a green-threads based vm thanks to
with green threads, which clearly defeats the purpose of using SMP systems.
arcane bugs in native threads implementations. At the very end, you may fire up two (or more)
communicating vms and still get a performance increase - especially if your box is rich in memory and
the problem at hand is decomposable with somewhat limited communication between the
interacting parts.
And as an aside, you don't kill the only processor when your vm engages in computationally intensive
operations.
-- dimitris mailto:[EMAIL PROTECTED]
