JDK version ?

Godmar Back gback at cs.utah.edu
Wed Mar 24 09:21:37 PST 1999


> 
> It is also the case that kaffe is a high-performance virtual machine.  I
> haven't personally run any benchmarks (don't have anything from Sun
> installed here) but I've seen benchmarks in which kaffe does very well.
> 

My take on this is that you should expect Kaffe with jit to perform about 
as good as Sun's 1.1.7 interpreter for *real* applications.  For small
benchmarks, you can see Kaffe outperform Sun and even Sun with TYA
at times.  If you have an application where kaffe substantially lags 
behind Sun's 1.1.7, please let us know.

Given the existence of Microsoft's JVM, calling Kaffe high-performance
may not be the right word.  But: here's a tremendous chance of making
a contribution.  Basically, the complete JIT in Kaffe is waiting to
be ripped out and be replaced with an optimizing one. [3]

> Kaffe's AWT is, IMO, a strong point too.  It doesn't use native widgets
> (which you may find distasteful), but it's very fast, and it actually
> performs according to spec, unlike most sun versions I've seen.
> 

You know, about the whole "performs according to spec".  I'm not really
sure what that means.  Java is a Sun invention.  They defined both the
"spec" and did several implementations.  Since they control the 
standardization process, they always have the option of either fixing the 
"spec" or fixing their implementation, or just to ignore discrepancies.  
All of which they have been doing and are still doing, and they aren't
even shy to admit it.

Transvirtual and others are attempting to move some aspects of the control 
over Java to an independent body [1], but similar attempts have failed in 
the past (see the case of NIST of real-time Java [2]).

It remains to be seen how the struggle between attempts to create an
open standard and the fear of people that Java may disperse into 
different standards without Sun's control will turn out.

Until then, I'm afraid we have little choice but to pick and prefer  
a given JDK reference implementation from Sun over whatever "spec"
they release.  Of course, we should also lobby for making Java truely open.

	- Godmar

[1]: http://www.transvirtual.com/pressrelease/dase-proposal.html
[2]: some background is here: http://www.newmonics.com/webroot/rtjwg/qa.html
[3]: http://www.kaffe.org/develop.html#cool


More information about the kaffe mailing list