Re: Truly later thinking...Honey, grab me another beer

Joe Pluta skrev den 12-08-2008 17:01:

Thorbjørn Ravn Andersen wrote:

Depends on the language.

No, it doesn't.

Yes it does. Depends on the libraries called.

(And lets not even get into situations where the speed is not cpu-bound..)

Perl e.g. has very powerful text operators, is interpreted, but is still one of the best ways to massage large text files (Perl regular expressions are NICE).

But correctly written C code that does the same thing will run much faster.

First of all it takes MUCH longer to write correctly in C and secondly you are wrong as Perl is calling correctly written C code internally :)

(and I don't think I'll bite on why compiled Java is slow on the i ;-) )

Not sure what you mean here. Are you saying that Java on the i is slower than Java on a Windows machine? That's got nothing to do with the issue. On the same platform, though, I disagree: Java code compiled by the JIT compiler runs faster than the same Java code interpreted - on any machine, including the i.

I had a case doing heavy graphics where the JITC was very aggressive in doing optimization but catching in and overtaking in a minute or two.

Please also note that the java byte code IS compiled, but that further compilation is necessary to achieve necessary speed.

Now, if you mean using the CRTJVAPGM compared to the JIT compiler, that's a different question. CRTJVAPGM (which AFAIK is no longer recommended) does static optimization which can't compete with the dynamic optimizations of a good JIT compiler. If you want to learn more, start with Wikipedia:

Thank you. I have previously done some extensive benchmarking to identify which worked best for us, and I found that the JIT compiler (this is for Classic) was the fastest. Even faster than opt40.

I don't pretend to be an expert. Optimization is a class of programming way above my expertise. If you want to talk about people who really know how a computer works, the optimization folks are the very best in the business.

HotSpot is really tough programming. It is so nice that we all benefit from the lessons learned that runtime profiling and optimization gives the best performance, and that we get access to it just by using a language running on the JVM.