Snapshot-GC (was Re: Linux and GC)

[Marc Wachowitz]

Years ago I've read several papers for such gc algorithms (though I
don't remember any authors or titles); they called it "snapshot gc".

Note that to avoid excessive allocation or synchronization overhead,
you'd better think very carefully about the interface between those
processes, concerning relative speeds (the collector's work should be
nicely distributed over the time until another collection is due, which
by itself does already depend on the allocation/deallocation rates/sizes
and memory fragmentation), giving back a possibly very large free-list,
etc. Doing this well is hardly a trivial task, and unless you have great
algorithms and heuristics for these aspects, I wouldn't expect snapshot
gc performance to be competitive with more conventional strategies.

-- Marc Wachowitz <mw@ipx2.rz.uni-mannheim.de>