Dependability, Abstraction, and Programming

Abstract. In this paper, we look at what is required to produce programs that are dependable. Dependability requires more than just high availability. Rather, a program needs to be ―right ‖ as well, solving the problem for which it was designed. This requires a program development infrastructure that can, by means of appropriate abstractions, permit the programmer to focus on his problem, and not be distracted by ―systems issues ‖ that arise when high availability is required. We discuss the attributes of good abstractions. We then illustrate this in the programming of dependable systems. Our ―abstraction ‖ is a transparently persistent stateful programming model for use in the web enterprise setting where exactly-once execution is required. Work on this abstraction is reviewed. The new technical meat of the paper is in (1) describing how to reduce the performance cost of using the abstraction; (2) extending the flexibility of using this abstraction; (3) and showing how to exploit it to achieve dependability