Highly reliable synchronous and asynchronous remote procedure calls

The remote procedure call mechanism currently available in the distributed computing environment (DCE RPC) is believed to obey the same programming paradigm as the local procedure call programming paradigm. In this paper we observe that error handling differs significantly between the local procedure call paradigm and the DCE RPC paradigm, with the DCE RPC paradigm being the more complicated. After discussing DCE RPC error handling, we proceed to describe a new form of remote procedure call, the highly reliable remote procedure call (HR RPC). The goal of the HR RPC is to make remote procedure calls highly reliable, that is as reliable as local procedure calls, and to make the remote procedure call programming paradigm truly the same as that of the local procedure call programming paradigm. The authors have implemented prototype UNIX(R) HR RPCs using a minimal queued transaction processing system (MQTPS) that operates between a client application program and the DCE RPC mechanism, and have extended it to provide asynchronous remote procedure calls, as well as priority and concurrency control. Hence, the prototype is implemented as middleware that runs in UNIX user space and is disjoint from both UNIX and the DCE RPC mechanism. Thus, the prototype HR RPCs can be easily ported to any UNIX machine supporting DCE RPC