Using code shipping to optimize Remote Procedure Call

Remote Procedure Call (RPC) has become the primary method for communicating between client and server processes in distributed computing systems. While RPC is a valuable abstraction, the performance of an application structured using RPC can suffer because each remote call requires a high-latency message exchange. This paper presents a new technique called code shipping that automatically reduces the number of remote calls by folding multiple calls into a single RPC. Code shipping relies on compile-time changes to client programs performed by a source-to-source translator to generate code that is shipped to a special remote procedure located on the server. This remote procedure executes the shipped code locally, thereby transforming what had been multiple remote procedure calls into local procedure calls with a corresponding performance improvement. This approach also makes it easier to adapt to different client access patterns without loss of performance or the use of manual call folding techniques that require redesign of the server interface.