From: | Rudi Chiarito |
Date: | 24 Sep 2000 at 21:02:00 |
Subject: | Re: AMIOPEN: XML and Java, a uncharted territory? |
On Fri, Sep 22, 2000 at 10:46:54PM +0100, Jim Peters wrote:
> If, as you suggest, Tao's JVM does a really good job of real-time
> garbage collection, then I am more than willing to throw out
> reference-counting. I'm cautious, still, until I've tried it for
> myself, and got an understanding of the algorithm (if possible).
I've only tried it with one program, which, though, was expressely
designed to stress the collector. That's a very good start, although I
suppose more tests would be necessary - I don't know of any. At the same
time I'm sure Tao has done a lot of work in this area, as it's one of the
questions they must have been asked more often from potential clients.
The documentation (jintro.pdf and usingjava.pdf) says the collector
uses a mark and sweep approach and, more importantly, can be interrupted
by system events. It also goes into some details about techniques used to
reduce the impact of GC on system performance.
> To me garbage collection is a compromise - it's never going to be as
> efficient as C-code with all the alloc()s and free()s in the right
But in large systems you eventually end with a lot of code that deals
with nothing but allocations and deallocations. And with truely reusable
software one has always to check who is suppose to free or allocate
stuff.
Of course, still as of today a lot of programmers misplace or forget
free()s, so it must mean that maybe manual management is not the ideal
solution. I prefer a slightly slower program that manages memory
correctly to a faster program that once in a while frees the wrong
region of memory...
> can tolerate in your application. If Tao have found a really good
> solution that I really feel I can count on, I will be very happy
> indeed. Good enough for low-latency sound generation, do you think ?
I'm not the most appropriate person to answer that, I fear. I haven't
tested the collector much. I'm not going to tell you things I am not
completely convinced of.
> I'll be very interested to see this when it is released.
Hopefully there will be an introduction available soon.