This project develops optimizations based on object behavior prediction to improve the performance of manual and automatic storage reclamation algorithms. Programs written in languages such as C, C++, or Fortran that use manual garbage collection, and program written in language such as Common Lisp, ML, Smalltalk, or Modula-3 which uses automatic storage reclamation benefit from the techniques developed in this project. The project uses profile-based optimizations to predict properties of allocated objects at the time they are allocated. The properties include object lifetime and object reference behavior. By correctly predicting these properties, the space usage, CPU overhead, and reference locality of programs using either manual or automatic storage reclamation is improved. The project first investigate empirical properties of program in languages using manual and automatic storage reclamation and determines if behavior prediction is feasible. Then the project develops and evaluates alternative algorithms that exploit observed predictable behaviors. Finally, the project prototypes the most promising algorithms and measures their performance.
