Characterizing Aging Phenomena of the Java Virtual Machine

Abstract

In this work we investigate software aging phenomena inside the Java Virtual Machine (JVM). Starting from an experimental campaign on real world testbeds, this work isolates the contribution of the JVM to the overall aging trend, and identifies, through statistical methods, which workload parameters are more relevant to aging dynamics. Experimental results show that the Sun Hotpost JVM experiences software aging phenomena. A consistent memory depletion trend (up to 50 KB/min) has been observed during periods of low garbage collector activity; the Just-In-Time compiler is also responsible for a lighter, but not negligible, memory depletion trend; finally, a consistent throughput loss (up to 24 KB/min) has been observed.