Analysis of Queued Message Communication Between Processors

Abstract

Computing systems, from single user workstations to mainframes, are really interconnections of multiple processors. Normally only one of those processors is exposed as the programmable processor. With multiple processors sharing the work, it is interesting to consider the implications of different communication mechanisms between them. Classically we have used a communications channel to send messages between processors. Examples range from Local Area Networks (LAN) to the channels on an IBM 3090 computing system. In these cases the receiving system is involved in reading and storing the input messages. An alternative exists in which the sending system has the ability to place the message directly into a queue in memory which is accessible to the other processor. This paper explores the performance benefits of using a shared queue for communication between processors.