Naming CSCI 6900/4900

Unreferenced Objects in Dist. Systems Objects no longer needed as nobody has a reference to them and hence will not use them Garbage collection is necessary as objects consume resources In non-distributed systems garbage collection is much simpler than distributed systems Cross machine references, transitive references, references as parameters complicate matters Objects referencing recursively but not by anyone else

The Problem of Unreferenced Objects An example of a graph representing objects containing references to each other.

Reference Counting The problem of maintaining a proper reference count in the presence of unreliable communication. Counting the number of references of an object Simple reference counting works well for uni-processor systems, but encounters problems for distributed systems

Copying References Across Processes a)Copying a reference to another process and incrementing the counter too late b)A solution. Race conditions can easily arise Need careful design to overcome the problems

Weighted Reference Counting a)The initial assignment of weights in weighted reference counting b)Weight assignment when creating a new reference. Designed to overcome race conditions of simple reference counting Each object has a fixed total weight When a remote reference is created, half the weight is assigned to the new reference Object can be safely removed when the weight becomes zero

Weighted Reference Counting - 2 When a copy is created of a remote reference, the new copy receives half the weight When a reference is deleted, its current weight is subtracted Object can be safely removed when the weight becomes zero

Indirection in Weighted Referencing Creating an indirection when the partial weight of a reference has reached 1.

Generation Reference Counting Each proxy stores two numbers –A counter indicating the number of times the proxy has been copied –A generation number indicating how the proxy was created Original skeleton maintains info. about copies for each generation When a proxy is removed the original skeleton is informed about the generation number and number of copies

Reference Listing All the above approaches assume reliable communication Idempotent Operations – can be repeated safely without affecting the end result –Results depend on whether an operation has been invoked at all How to make adding deleting references idempotent? –Maintain an explicit list of references rather than counter –Check for duplicates when performing operation

Identifying Unreachable Entities Reference counting and listing cannot deal with unreachable entities with recursive referencing Need techniques to check all entities for unreachability – Tracing based garbage collection Mark & Sweep Collectors –Mark phase: Entities are traced from root set and marked –Sweep phase: Examine all entities and remove unmarked –Three coloring approach –Each entity is initially white –Entity is gray if found reachable but references need to be explored –Mark it black if reachable and explored completely –All white nodes are unreachable and can be removed

Distributed Mark & Sweep Each process starts a local garbage collector –These garbage collectors run in parallel Proxies, skeletons and objects are colored Initially all proxies, skeletons and objects are white When an object residing in process P is reachable from a root in P, it is marked grey –All proxies in the object are marked grey When a proxy is marked grey, its skeleton is marked grey, which in turn marks the associated object grey Local garbage collectors collect white objects Un-scalable and requires reachability graph to remain unchanged – STOP & GO approach

Tracing in Groups Designed to deal with the scalability issue Processes are organized into groups for scalability purposes –Group is a collection of processes Combination of Mark & sweep and reference counting Basic idea – collect garbage within each group and then recursively consider larger groups Assumption – Remote references are implemented as proxy-skeleton pair

Steps in Group Tracing 1.Initial Marking – Only skeletons are marked 2.Intra-process propagation – Propagate marks from skeletons to proxies 3.Inter-process propagation – Proxies to skeletons 4.Stabilization – repeat two steps until stability is reached 5.Garbage reclamation

Marks in Group Tracing Skeletons – Hard or Soft –Hard implies skeleton is reachable from proxy outside the group, or object in rootset of a process –Soft implies reachable only from proxies within the group Proxy – Hard, Soft or None –Hard implies proxy is reachable from object in root set –Soft implies proxy is reachable from skeleton marked as soft –None – Proxy is not reachable from skeleton or proxy in root set

Tracing in Groups (1) Initial marking of skeletons.

Tracing in Groups (2) After local propagation in each process.

Tracing in Groups (3) Final marking.