Thanks for posting this Greg. The details of observable reordering of results that is discussed on this page is the biggest issue that River and other “old java code” has when moving to these newer CPU architectures. This stuff just didn’t use to happen with older cache behaviors and the changes to JIT optimizations as well. And, as the uses of Hashtable and Vector get removed, the implicit cache coherency effects of those synchronized accesses of old, just make all kinds of things not “happen” correctly.
It will be great if there is a path for Peter’s branch to a release so that we can not fight all of these things over and over again as intractable bugs. Gregg Wonderly On Feb 11, 2014, at 11:31 AM, Greg Trasuk <tras...@stratuscom.com> wrote: > Mainly for the archives… > > I came across an interesting article on the Java Memory Model. > > http://gvsmirnov.ru/blog/tech/2014/02/10/jmm-under-the-hood.html > > Cheers, > > Greg Trasuk >
