home · mobile · calendar · bactac · 1999-2000 · 

BACTAC - Seidl

Improving the Reference Locality of Dynamically Allocated Objects
Matthew Seidl
Grad Student, Computer Science

Heap-allocated objects play an important role in many modern programs. My investigations of dynamically allocated objects have shown that there is a strong skew in object reference behavior: a small percentage of the objects receive a large percentage of the references. In this talk, I describe Avon, a memory allocation system that uses the skew in reference density between objects to improve program performance.

The Avon system is designed to increase the reference density of large programs that make heavy use of dynamically allocated memory by segregating objects into separate heap areas based on their reference behavior and lifetime. In my talk, I will show how it is possible to predict object reference behavior at allocation time. My system does this by using profile data to observe both the information available to the allocator when it creates an object and the resulting behavior of that object. My talk will also show how to build an allocator for a high performance program that can efficiently gather the allocation time information necessary to predict object behavior. As a demonstration of the Avon system, I will conclude by presenting an analysis of the performance improvements possible in a number of memory intensive programs.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:24)