> This is precisely the problem with mark/sweep collectors, as I
> understand it. The effort is proportional to *all* the data, dead as
> well as alive.
If you know that some data is 'dead' before the GC runs, though, I imagine
that its space can be re-used, thus reducing the volume of the 'all the
data' that needs checking. For example, a thousand serial
allocate/deallocate cycles can all use the same memory space. Ought to
reduce swapping, too.
- Eric