GrowthTracker: Diagnosing Unbounded Heap Growth in C++ Software

Abstract

Unbounded growth of heap memory degrades performance and eventually causes program failure. Traditional memory leaks are the most commonly recognized, but not the only cause of this issue. Large software systems use many aggregate data structures that can grow arbitrarily, and application behavior that produces unbounded growth of these structures is common. This growth can remain undetected by both memory leak and staleness detection tools. In this paper, we present an approach for reliably identifying aggregate data structures that can grow without bound over the lifetime of a program. Our solution tracks all aggregates over the lifetime of the program and utilizes heuristics to identify non-convergent growth. Our diagnostic method continuously reduces false positives and false negatives during execution, producing more accurate reports for as long as it is allowed to continue execution. In addition, we present techniques to utilize this method in large, pre-existing C++ software without requiring extensive code modification. Our tool identified data structures with this issue in Google's Chrome web browser and Apple's Safari browser among others.