Hello everybody,
it seems the magic of Collecting Garbage is still
something misterious for number of people. Although
there are number of issues about it (f.e.
www.javaworld.com/javaworld/jw-08-1996/jw-08-gc.html
).
The point is : I made a kind of Garbage Collector
for and along with database (Java BTW). And I
followed to "Compacting collector" model. Yes I do
have some overhead when controlling my objects and
freespace into the huge byte array (this is exactly
my database is - only on disk) - but I have no
delays, because I found the smallest portion of
information to be replaced ( I can control the size )
, so I can estimate the time for every tack of GC.
OK, for sure it's not perfect yet but I did it
without single loop - for looking up died objects and
free space.
I beleive the absolute GC should be based on the
issues : controlling the atom optimization size and
suporting all handlers to void loops.(GC should not
depend on the size of heap).
I'm ready to share this knowlege with the Java
community and especially with Sun (if Sun wants to)
to the general benefit.
I just wonder : I made my stuff for 2-3 years
(eventually) so it's not a big deal, why Sun haven't
implemented it yet. What it is lacking off : money or
ideas ? (With full respect to Sun - I like Java very
much and beleive in it's future).
------------------------------------------------------
BTW : my single-threaded JavaDataBase showed the
following efficiency in comparison with MS SQL Server
6.5 on NT4 (P2, 300, 128MB) - it shows the efficiency
of given GC approach , OK not directly :
write (records/second) read (records/second)
SQL 250 10000 (cash I believe)
JDB 50 (average ,GC on) 500 (pure disk ops)
I know how to improve it (thru pool of threads and +++)
but I don't know how to implement rollback - because
of the nature of collected garbage. And it doesn't
have SQL parser yet along with some usefull stuff.
So it's not a advertising. The product is not ready
yet.
Best regards.
Alex Semenov. Wired Solutions Inc.
[EMAIL PROTECTED]
---Doug Twilleager <[EMAIL PROTECTED]> wrote:
>
> There are two parts to this. First is how the JVM
does GC. This can cut
> down on the number and length of GC pauses. This
is mostly out of our
> control. We give our input to the JVM team often.
The current JVM's
> aren't very good at this, but some of the upcoming
JVM's are getting
> better.
Guys this is not too complicated to be not solved for
Java 2.
>
> The second part is how well Java 3D prevents GC
from happening. We have
> worked very hard to minimize anything that would
cause GC to happen.
I think this is not a good approach.
> It's an ongoing thing. The upcoming 1.1.1 release
has a number of changes
> that reduce our GC causing activities even more.
>
> So, yes we know about the issues, and it is very
important to us. And, we
> will keep working it until it gets better.
Take a hand, it's free...
>
> Doug Twilleager
> Java 3D Team
>
>
_________________________________________________________
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com
=====================================================================
To subscribe/unsubscribe, send mail to [EMAIL PROTECTED]
Java 3D Home Page: http://java.sun.com/products/java-media/3D/
