Meaningful Type Names as a Basis for Object Lifetime Prediction

Object lifetime prediction can identify short-lived and longlived objects at or before their heap-allocation point. This prediction is an increasingly important technique to enable optimization of runtime decisions for high-performance garbage collection. This paper shows that extracting information from object type names can give a reliable and efficient way to make such object lifetime predictions. We use this approach to optimize a generational garbage collector in the Jikes RVM system.