Low pause incremental compaction in a generational garbage collection

Abstract

In recent years Java™ Virtual Machine (JVM) is widely used for server applications, such as web services and web application servers (WASes). Server workloads like these have different characteristics from other workloads. They are large applications that need large number of threads, processors, sockets, and a large amount of memory. They are interactive in that they should respond to the requests of clients in short time. And they are long running and the working sets increase as time goes on. To improve the performance of server workloads, a new design of the garbage collection (GC) component of JVM that meets the characteristics of them is required. At first, generational garbage collection algorithm can increase the throughput of garbage collection. And to reduce the working set of server workloads and the memory fragmentation of long running servers, compacting garbage collection algorithm is necessary. But compaction needs copying of a lot of objects and occurs while program activity is stopped. So it is a major contributor to garbage collection pause times. We present a low pause incremental compaction algorithm in a generational garbage collection. It increases the throughput of garbage collection and reduces pause times by compacting a part of the heap occasionally when young generation garbage collection occurs. Experiments show that our algorithm reduces maximum garbage collection pause time stably up to 38%. And the increase of total run time is minimal compared to previous algorithm.