GC statistics (was Re: big project ported)

Godmar Back gback@cs.utah.edu
Wed Oct 27 10:58:00 GMT 1999


 Matt,

> 
> I strongly recommend that people interested in improving GC performance in
> GCJ look at the papers on the ExactVM from Sun Labs, two of which are at:
> 	http://www.sun.com/research/techrep/1998/index.html
> Called "The GC Interface in the EVM" and "GC Points in a Threaded 
> Environment". Perhaps the papers you really want to read are referenced 
> from these.
> 	
> It sounds like what many of you are talking about is generational and
> exact garbage collection, something which the ExactVM (now called "EVM")
> does very well. It should be possible to implement a very efficient 
> GC system in GCJ -- but it's probably best to draw on the work that's been
> done before in this area!
> 

I didn't discuss the issues revolving around generational collection;
this is a whole other bag of issues.  

My understanding is that Sun's route is to require C code to cooperate
by using JNI.  In fact, this is mentioned in the EVM paper, see page 26.
Cygnus's approach is a different one with CNI, for various reasons as
discussed on this list.

That said---there's clearly plenty to learn with regard to the other
aspects of allocation and other aspects of collection, such as.

I definitely agree with you that most of the issues aren't exactly
new ones, and we're not discussing the earth-shattering issues here---
this is not a research list, but a list about engineering issues of
gcj.  However, not all people on the list may be familiar with the
issues, and often discussion and exchange of information helps 
people to learn.  Having said that, let me contribute a pointer to
a voluminous reference on gc issues [1]

	- Godmar

[1] the Garbage Collection Bibliography
http://www.cs.ukc.ac.uk/people/staff/rej/gcbib/gcbib.html
I especially recommend this as an intro
@inproceedings{wilson:lncs92,
        AUTHOR = {Paul R. Wilson},
        TITLE = {Uniprocessor Garbage Collection Techniques},
        BOOKTITLE = {Proceedings of the International Workshop on
                Memory Management},
        YEAR = {1992},
        PUBLISHER = {Springer-Verlag},
        ADDRESS = {St. Malo, France},
        MONTH = sep,
        notes = {Lecture Notes in Computer Science 637}
}



More information about the Java mailing list