[gclist] GC and OODBMSes: any experience?

Giuliano Carlini giuliano@carlini.com
Tue, 03 Dec 1996 14:23:08 -0800

Lassi A. Tuura wrote:
> Note that I am currently not interested in garbage collecting the
> database

You should keep an open mind. There are some algorithms in
the works that might make it feasible. They haven't been
given any real world testing yet, but that should occur in
the near future.

I sure wouldn't want to trust the integrity of a multi-
petabyte datadase to expilict deallocation if GC was an option.

> (GCing 10 petabytes would take a bit too long :-)

There are algorithms that can run in a separate process
and never stop the mutators at all. I don't remember,
but I think the running time of the GC is linear w.r.t
database size. Using a generational approach should
allow you to recover most unreferenced space quickly.

To repeat myself so I don't raise false hopes: these
algorithms need to be tested in practice to check that
they work in the real world.