[gclist] Name that hypothesis
Henry G. Baker
Wed, 4 Dec 1996 17:30:11 -0800 (PST)
> Perhaps something like "the probability an object will die is inversely
> proportional to its age". This still leaves "age" undefined, but I
> think its easier to pin a meaning on it than "young" and "old" as
> It's then up to the implementor to define/measure the "death probability
> curve" for a given application and see if it has properties an allocator
> can take advantage of (as Henry pointed out, its only when this curve
> has discontinuities that a generational GC helps).
I didn't mean to imply that the curve had to have discontinuities, but
that it had to have a different shape/slope in some places than the
curve for the 'null hypothesis'. A clever algorithm should then be able
to tune itself to whatever differences exist between the two curves, just
like a clever compression algorithm can take advantage of whatever
redundancies that exist in the input data.