Runtime Object Lifetime Profiler for Latency Sensitive Big Data Applications

Abstract

Latency sensitive services such as credit-card fraud detection and website targeted advertisement rely on Big Data platforms which run on top of memory managed runtimes, such as the Java Virtual Machine (JVM). These platforms, however, suffer from unpredictable and unacceptably high pause times due to inadequate memory management decisions (e.g., allocating objects with very different lifetimes next to each other, resulting in severe memory fragmentation). This leads to frequent and long application pause times, breaking Service Level Agreements (SLAs). This problem has been previously identified, and results show that current memory management techniques are ill-suited for applications that hold in memory massive amounts of long-lived objects (which is the case for a wide spectrum of Big Data applications). Previous works reduce such application pauses by allocating objects in off-heap, in special allocation regions/generations, or by using ultra-low latency Garbage Collectors (GC). However, all these solutions either require a combination of programmer effort and knowledge, source code access, offline profiling (with clear negative impacts on programmer's productivity), or impose a significant impact on application throughput and/or memory to reduce application pauses. We propose ROLP, a Runtime Object Lifetime Profiler that profiles application code at runtime and helps pretenuring GC algorithms allocating objects with similar lifetimes close to each other so that the overall fragmentation, GC effort, and application pauses are reduced. ROLP is implemented for the OpenJDK 8 and was evaluated with a recently proposed open-source pretenuring collector (NG2C). Results show long tail latencies reductions of up to 51% for Lucene, 85% for GraphChi, and 69% for Cassandra. This is achieved with negligible throughput (