This paper summarizes the main themes of my talk of the same tide at the First ACM Workshop on Optimization of Middleware and Distributed Systems. To begin with, lets discuss what we mean by the term "middleware". Exactly what is middleware in the middle of? There are two similar but slightly different approaches to this question. One approach views middleware as being in the middle of a sea of distributed components, such as applications, directories, databases, etc. Middleware serves as the glue that allows all of these components to interact with one another. Using this approach, middleware is seen as enabling the horizontal composition of components. The second approach views middleware as the stuff in between an application and lower level services offered by the distributed network. In other words, middleware serves as the abstraction layer that allows an application programmer to easily access distributed services without having to worry about the details of connectivity, protocol conversion, data transformation, load balancing etc. Using this approach, middleware is seen as enabling the vertical composition of an application with lower level services. Both approaches are correct and they describe two important functions of middleware: enabling component interaction and simplifying the programming model for application writers. Concretely, when we speak of middleware we refer to things like Web Application Servers, Object Request Brokers (ORBs), and Message Oriented Middleware (MOM). All of these technologies present programming models to the application programmer and facilitate the collaboration between distributed components. It is hard to define the functionality of middleware precisely as every new middleware product release incorporates additional features.
