Analysing RPC and Testing the Performance of Solutions

Abstract

In distributed computing, network sockets provide mechanism for a process to establish a remote connection to another process and send messages back and forth. This interface makes possible a proper mechanism that allows a program running as a process on computer A to call a procedure or a function on remote computer B and pass parameters to it. In the case of synchronous Remote Procedure Call (RPC), processes on computer A need to wait for the finishing of execution of procedures on computer B. When the called procedure finishes, produces its result and passes it to the process on computer A that can continue execution. The question is what happens between the time of the remote procedure call and arrival of the returned values and how much the caller must wait for result. To answer that question, this paper describes the structure of third generation RPCs and analyses them putting the focus on performance and the way of marshalling parameters. To facilitate the choice between them this paper represents the results of performance tests carried out by the authors