[gclist] Thread safe GC

[Stefan Monnier]

Jim Larson <larson@kesey.jpl.nasa.gov> writes:
> sequentially.  Even though we had two dozen processors trying to
> do GC work, all but one of them were idle after an initial flurry
> of activity, and had to wait for a long, deep list to be traversed.

If I understand you right, you had a single collector but parallelized.
What you could try to do is to shift the parallelism a little higher by
splitting your memory into a few regions that are GCed in parallel (and you
have of course to maintain lists of inter-region references).

Of course this can only "get you rid of" the lenthy list if some of the pointer
in the list is one of the inter-region references.


        Stefan