Predicting References to Dynamically Allocated Objects University of Colorado at Boulder Predicting References to Dynamically Allocated Objects

Dynamic storage allocation has become increasingly important in many applications, in part due to the use of the object-oriented paradigm. At the same time, processor architectures are including deeper cache memory hierarchies to hide the increasing latency to main memory. In this paper, we investigate eeorts to predict which heap objects will be highly referenced at the time they are allocated. Our approach uses proole-based optimization, and considers a variety of diierent information sources present at the time of object allocation to predict the object's reference frequency. Our results, based on measurements of four allocation intensive programs, show that program references to heap objects are highly skewed and that our prediction methods can successfully predict the program objects that will be the most highly referenced. We show that using this technique, a large fraction of a program's heap references can be directed at a part of the heap that is small enough to t in the cache (e.g., 8{64 kilobytes).