Investigating High Memory Churn via Object Lifetime Analysis to Improve Software Performance

Abstract

High memory churn occurs when many temporary objects are created and shortly thereafter collected by the garbage collector. Such excessive dynamic allocations negatively impact an application’s performance because (1) a great number of objects has to be allocated on the heap and (2) an increased number of garbage collections is required to collect them. In this paper, we present ongoing research on how to support developers in detecting, understanding and resolving high memory churn in order to improve their application’s performance. Based on a recorded memory trace, an algorithm automatically searches for memory churn hotspots and calculates the age at which objects die within it, since objects that die young are the major contributors to memory churn. Information about these objects, for example their types and allocation sites, can then be inspected in order to locate the problematic code locations. To demonstrate the feasibility and applicability of our approach, we implemented and present a new memory churn analysis feature in AntTracks, our trace-based memory monitoring tool.