Lambda the Ultimate

Lately I've been thinking of garbage collection in terms of being a lazy algorithm. That is, I see a lot potential similarities between lazy evaluation and garbage collection. One of the main benefits of lazy evaluation is the modularity it provides by allowing finer-grained intermingling of programs (the separation of generation from selection). Garbage collection follows this pattern closely by allowing the allocation of memory anywhere, without having to worry about deallocation. Also, the freeing of unused memory is done dynamically on demand, instead of being determined at compile time. Again, similar to the execution path of a lazy program. And lazy evaluation makes it easier to reason about correctness, but harder to reason about efficiency. Much the same could also be said about GC.

I think the unification of the two concepts could inform both areas. I could see a form of strictness analysis performed on our 'lazy' garbage collector which would produce a region inference algorithm or a compile time garbage collection algorithm. And maybe garbage collection research could help in developing better evaluation strategies. Although I have a harder time seeing it, might a generalization of the idea behind generational GC applied to an evaluation strategy lead to something similar to Optimistic or Eager evaluation? (That is, spend more time on executing younger thunks, etc.)

Are there any papers out there which discuss the lazy aspects of garbage collection?