Autonomous garbage collection: resolving memory leaks in long-running server applications

Abstract

We demonstrate the benefits of a garbage collection technique that requires neither programmer assistance nor rebuilding (compiling or linking) of target applications. That is, our approach effectively mitigates performance degeneration due to memory leaks in applications when the source and object code are unavailable. Our technique is an extension of the garbage collection method known as conservative garbage collection. We refer to our approach as autonomous garbage collection. Autonomous garbage collection is especially useful for long running server applications in large-scale information processing server environments. Our prototype demonstrates that this garbage collection technique is feasible. Our experimental results show that this technique is more general and easier to use than many of the previous garbage collection proposals targeted at resolving memory leaks in non-cooperative server applications.