User reports that when they use NIO, the implementation of the
gc locker with CMS is causing periods when threads doing get()s and put()s
on NIO buffers are getting blocked for several seconds at a time.
They have provided a test case that purportedly reproduces the
issues that they are having.
They claim that rewriting their application to avoid the use of NIO
is avoiding the blockage of threads involved in IO.
An email trail of the incident and its description and follow-up can
be found starting at the following URL :-