On Tue, Sep 03, 2002 at 10:03:15PM +0200, Marco Trevisan wrote:
> I found out that the best performing JVM on a single-processor Linux 
> machine is Blackdown-1.3.1 with green threads and the OpenJIT compiler.
> Other JVMs I tried are: IBM v1.3.0 - 1.3.1, Sun v1.3.1 - 1.4.x

If there is any gain in speed it's most likely because of the context
switching overhead of native threading (crossing kernel trap boundaries)
verses green threading which runs within userspace (fast context switching).

> It was a surprise for me to find out such a result, infact some 
> benchmark classes I've done in order to measure pure processing power 
> showed the IBM JDK as the best performer: nevertheless, when it comes to 
> running my webapp under Tomcat, Blackdown 1.3.1 with green threads and 
> OpenJIT makes it run _noticeably_ faster.

OpenJIT is good stuff. :)

> I looked at the NGPT home page, surfed the net and found some 
> interesting benchmarks : 
> http://www.opengroup.org/rtforum/jan2002/slides/linux/abt.pdf .

There purpose is to build a highly scalable and complete pthreads
system. LinuxThreads is missing a couple of things that are important
to the Posix specification, namely proper signal handling. The HotSpot
compiler code accounts for this brokeness with another signal handling
layer within the internal thread management code... That's got to be
addressed in the code before the threading system can run.

> At http://www.ibm.com/developerworks/oss/pthreads/, they say: "This 
> release is fully suitable as a replacement for LinuxThreads by either a 
> single user or group or an entire distribution."
> Does it mean that if I patch the kernel and install it on my system, my 
> JVM will use it?  I guess it's not so easy :-)

I believe what I've said to be accurate, but...

If it emulates all the bad signal handling in Linux, then it might work.
You can try it and report back to us. :)

bill


----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to