I have kept quiet on this mailing list but can't help putting in my two
cents (not that its worth more than that).
As I have read the e-mails, I have been reminded of the evolution of
integrated circuit design. Back in the early 70's there was a push to design
asynchronous circuits. They held the appeal of being faster than their
synchronous cousin. Several good designs by top engineers and leaders in the
field were developed. However, it was soon discovered that although these
circuits worked in theory, variations in the materials used for the
circuits, and/or the environments in which they ran made their behavior
unpredictable. As a result, synchronization points were added to the
circuits and the circuits were clocked to advance the signals. As history
has shown us, the imposition of synchronization did not hinder development
but in contrast, fostered it.
What I'm hearing on this mailing list is the same thing. We have programmers
who want to write asynchronous programs. These are some of the top minds in
the field. They have designed some quite clever and unique algorithms.
However, variations in the algorithms and/or in the systems they are run on
have made their behavior unpredictable. Since Java is designed to run on
multiple under-lying systems, just like the asynchronous circuits were
expected to work in multiple environments, the solutions appears to be what
Bill proposed. Being a proponent of we should learn from history and not
repeat it, I would have to back the proposal of only guaranteeing behavior
if and only if synchronization is used. To me these appear to be the same
problem only under a different wrapper.
As I said to begin with, this is just my 2 cents, I'm only a graduate
student so feel free to ignore it. :)
Marie T. Conte
IMPACT Research Group
Center for Reliable and High-performance Computing
Electrical and Computer Engineering Department
University of Illinois
1308 W. Main St.
Urbana, IL 61801
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:22 EDT