Abstract: | It is an unfortunate fact that current general-purpose middleware implementations cannot support large-scale, wide-area, real-time, high-throughput applications using distributed objects. This paper discusses the background to middleware and explores the need for research into techniques that support high-capacity implementationsIn order to place the role of this work in context, an overview is given of the simple applications-middleware-infrastructure three-layered model that has been adopted. The paper also gives some background on how applications may specify the non-functional characteristics that they demand of supporting middleware, including invocation latency, which is a widely recognised, measured and reported metric, and invocation throughput rate, which is not.The paper establishes that there is a growing demand for high-invocation-throughput applications, such as the World Wide Web (WWW), collaborative virtual reality (CVR) and intelligent networks (INs), and also that there is increasing availability of suitable high-capacity infrastructure including networks, hosts and operating systems. It also examines the reasons for a lack of implemented high-capacity capacity middleware and reviews current directions in distributed object (DO) research. There is insufficient but growing activity in high-throughput techniques, such as weak-consistency approaches to object replication, that would enable applications to take advantage of high-capacity infrastructure. |