Cental article:

David Plainfossé and Marc Shapiro, "A Survey of Distributed Garbage
  Collection Techniques".

Questions (please return answers to the questions before monday,
  October 21., so that they can be reviewed by another participant
  for the coming wednesday):

(Reference-based dgc)

1. Why 'entry items' and 'exit items'?  Why don't remote objects not
   just point directly to their target objects?

2. Consider the following weighted reference counting scheme: It works
   as the one presented in paper, only each entry node only has a
   single weight, which is initally set to 0.  Once a reference is
   requested, the entry node's weight is decremented by a certain
   number (determined by either owner of entry node, or by requestor,
   that is owner of exit node), and the corresponding exit node is
   incremented by the same number.  Copying of references is as in the
   paper.  An entry node is deallocated once its weight reaches 0
   (again).  Does this work?  (Is it correct to deallocate once an
   entry node's weight reaches 0 again?)  How does this scheme compare
   to the one presented in the paper?

3. Arne suggested use of infinite precision numbers for weights that
   allow exact division by 2 arbitrarily often.  What are the
   advantages and disadvantages of that scheme?  Would it work at all?

4. Once an exit node runs out of weight to divide, it can either
   request more weight from the corresponding entry node, or it can
   function as an 'indirect' entry for new exit nodes, dishing out
   weight of its own.  Is this better or worse than asking for more
   weight (and in what sense "better" or "worse")?

5. Reference weights are used to avoid race conditions.  Can't these
   just be avoided by using TCP instead of UDP?

(Tracing dgc)

6. Argue why it is correct to deallocate data whose time stamp is less
   than the least of redo-times in the network.

7. (Optional) Hughes writes in his article that the redo counter on a
   local processor can go both up and down.  How is this possible?
   Why doesn't it just go up?

8. What happens if a space fails in the tracing-with-timestamps collector?

9. Suggest a way to overcome this problem.

10. Why is it only necessary to send colour messages for black exit
    items in the tracing-within-groups collector?

11. When do the global marking step (or propagation of colour step)
    terminate for this collector?