[concurrency-interest] Software Transactional Memory

Joseph Seigh jseigh_cp00 at xemaps.com
Sun Mar 25 18:42:30 EDT 2007


I took down that STM package I had out on sourceforge.  It had a few
implementation errors which I didn't notice until I changed the testcase
implementation from a LIFO queue to a FIFO queue.  It was fun debugging
that one.

I'm not going to put an up fixed package out there because I'm not real
happy with the performance and the scheduler artifacts on a single
processor system make it impossible to demonstrate scalability for
the time being.

Question about ConcurrentLinkedQueue.  It uses AtomicReferenceFieldUpdater
rather than AtomicReference.  Why is that?  The former is a lot slower 
than the
latter according to my measurments and allocating extra objects didn't 
seem to
make that much difference so having an extra object for AtomicReference 
shouldn't
be a problem.

--
Joe Seigh


More information about the Concurrency-interest mailing list