Higher-order Distributed Computation over Autonomous Persistent Stores

The traditional approach for building distributed applications is by calling a procedure in another store using an RPC mechanism. However, an RPC requires a round-trip network delay for every call and makes each store dependent on the availability of other stores. A solution to this problem is to migrate the remote objects needed to the client store, and in particular the remote procedures themselves. After building an initial higher-order RPC we observed that in a persistent programming language the transitive closure of references for many objects includes large parts of the store, creating significant problems in efficiency and semantics. In this paper we present a refined programming model based on two new mechanisms: migration by substitution to avoid migrating the entire transitive closures; and persistent spaces that help with preserving object sharing both within and between stores.