CAREER: Principles and Techniques for Automated Middleware Specializations in Distributed Systems

Vanderbilt University, Nashville TN

Investigators

Abstract

Standardized general-purpose middleware remain an attractive option to develop performance-sensitive and safety-critical distributed systems due to their support for commonly required features that simplify distributed computing. The feature-richness of these middleware and coarse-grained design, however, often adversely impact memory footprint and performance of the applications. Proprietary solutions and handcrafted optimizations to standardized middleware are infeasible due to high development and maintenance costs. This research is addressing these problems via automated specialization of general-purpose middleware (i.e., pruning unwanted features, and adding and customizing the necessary ones). The intellectual merit of this research lies in how the automated specializations are realized by (a) identifying and exploiting the algebraic structure of middleware that helps map the specialization problem into a feature-oriented software development problem; (b) developing a new theory for feature composition, refactoring, and interactions across the lifecycle stages of applications; and (c) developing a well-defined specialization process based on generative programming. The broader impact of this CAREER research is observed through the (a) development of new principles to design middleware that make them inherently amenable to specialization; (b) documentation of specialization patterns, which reduces software maintenance; (c) enhanced developer productivity and system correctness; and (d) transitioning of the research artifacts to education and scientific community through an open source tool suite called GAMMA (Generative Aspects for the Manipulation of Middleware Architectures).