Deloy writes:
> Has anyone seen this article?
>
> http://www.news.com/News/Item/0%2C4%2C16257%2C00.html?nd
>
> What is Sun up to?  Do you think this is going to affect the ISO's
> decision on the Java standard?

In case you haven't read any of the articles, Sun got caught with its
hand in the cookie jar last week when the people who do the CaffeineMark
benchmark discovered that Sun had been touting the performance of Java
on Solaris as significantly faster than the fastest Intel systems by
using a version of their JIT compiler that cheated on the benchmark by
recognizing the benchmark code and either emitting hand optimized code
or outright lying (it's not clear which) about the results.

The compiler actually contained a 200 byte sequence which exactly matched
the object code of the benchmark program.

Sun had already removed this beta compiler from their web site, but they
were still advertising the results of the benchmark.  The people who
produce the benchmark put out a press release explaining what they had
discovered after Sun apparently didn't respond to their requests for a
retraction of the benchmark results.

Sun initially denied that they had done anything wrong, and even if they
had then "everyone else does it too and people expect it".

Yesterday someone higher up in Sun admitted what had happened (the article
referenced by the URL above) but claimed that it was an "accident" that
they released this "highly optimized" beta version of the compiler.  Duh.

I suspect we may yet see an even more complete apology from even higher up
as the realization sinks in that it *appears* as though Sun has committed
a serious fraud and their reactions so far have looked like denials and
excuses.  I suspect that the people responsible may not be working for
the company a week from now.

I don't think this will have any dramatic effect on Sun's ISO submission
request, but it certainly gives Microsoft or any other anti-Sun group a
strong counter argument to the "Microsoft is evil, Sun is good" argument.
Sun has been trying to give the impression that they hold the moral high
ground at least relative to Microsoft, and their handling of this situation
(not so much that it occurred in the first place) is making them look
pretty bad.

The fact that the whole thing happened at all is more humorous than
annoying.  I suspect someone in the compiler lab got "cute" and then
showed the result to someone in marketing who couldn't resist writing
a press release about it.  I doubt that the activity was in any way
prompted or condoned by people higher up (although how much pressure
was being placed on the compiler lab to "beat Intel/Microsoft" would
be interesting to know).

Unfortunately I think Sun's early denials had an element of truth when
they said that "everybody does it".  While actually embedding the object
code of a benchmark into the compiler to match against is unusual, way
to much of our money is spent designing and tuning products like CPU
chips so that they perform well on benchmarks rather than real world
applications.  I'll bet that if you visited each compiler lab in the
world that is working on a Java compiler/VM implementation, you
would fine a white board with the names of the benchmark suite components
on it and current times for their implementation versus the competition
and also find that that's where most of the money/brains/time is going.

G.