Live heap space analysis for languages with garbage collection

The peak heap consumption of a program is the maximum size of the live data on the heap during the execution of the program, i.e., the minimum amount of heap space needed to run the program without exhausting the memory. It is well-known that garbage collection (GC) makes the problem of predicting the memory required to run a program difficult. This paper presents, the best of our knowledge, the first live heap space analysis for garbage-collected languages which infers accurate upper bounds on the peak heap usage of a program's execution that are not restricted to any complexity class, i.e., we can infer exponential, logarithmic, polynomial, etc., bounds. Our analysis is developed for an (sequential) object-oriented bytecode language with a scoped-memory manager that reclaims unreachable memory when methods return. We also show how our analysis can accommodate other GC schemes which are closer to the ideal GC which collects objects as soon as they become unreachable. The practicality of our approach is experimentally evaluated on a prototype implementation.We demonstrate that it is fully automatic, reasonably accurate and efficient by inferring live heap space bounds for a standardized set of benchmarks, the JOlden suite.