mBalloon: enabling elastic memory management for big data processing

Abstract

Big Data processing often suffers from significant memory pressure, resulting in excessive garbage collection (GC) and out-of-memory (OOM) errors, harming system performance and reliability. Therefore, users tend to give an excessive heap size to applications to avoid job failure, causing low cluster utilization. In this paper, we demonstrate that lightweight virtualization, such as OS containers, opens up opportunities to address memory pressure: 1) tasks running in a container can be set to a large heap size to avoid OOM errors without worrying about thrashing the host machine; 2) tasks that are under memory pressure and incur significant GC activities can be temporarily "suspended" by depriving the hosting container's resources, and can be "resumed" later when other tasks complete and release their resources. We propose and develop mBalloon, an elastic memory manager, that leverages containers to flexibly and precisely control the memory usage of big data tasks. Applications running with mBalloon can survive from memory pressure, incur less GC overhead and help improve cluster utilization.