Research On The Mechanism Of Mutiprocessors-Oriented Parallel Garbage Collection

The pervasiveness of multiprocessor and multicore hardware and the rising level of available parallelism are radically changing the computing landscape. Multithreaded applications with multigigabyte heaps running on modern servers provide new challenges for garbage collection.The challenges for"server-oriented"GC include: ensuring short pause times on a multigigabyte heaps while minimizing throughput penalty, and good scaling on multiprocessor hardware.This paper introduces the basic study of Java garbage collection. It includes researches and an analysis of the main factors, which have an impact on the efficiency of garbage collectors. Then, there are a summay and comparison of current scheduling schemes for garbage collection in multi-processor environment. The goals of garbage collection that work for multi-threaded applications on modern servers cover the time convergence, spatial convergence, scalability, load balance,and other features.With a combination of conjecture and experimental demonstration, the relationship between the heap size of applications and the number of garbage collections is discoverd, which improves our understanding of memory management. If all other factors remain equal, the larger the heap size, then the fewer GCs required, leaving more processor time for applications. This paper puts forward a tax-and-spend scheduling model for server-oriented GC with its consumption rules and replenishment rules, which is inspired by bandwidth-preserving server scheduling algorithms.In the environment of service-oriented computing on multi-processor, according to the differences among application requirements, applications can be classifed into time-critical and space-critical. The efficiency of garbage collection, in essence, is a time-space relationship between the storage space and garbage collection time. With the understanding above, space-time complementary scheduling scheme based on control of heap size growth is brought forward, which makes a consideration of three elements, that is, the workload of applications on real world systems, storage resources, and scheduling algorithms.Finally, there is an analysis of time complexity, and a upper bound for the amount of work to finish a full GC cycle is presented. The test data with benchmarks shows that it makes an effective use of parallel processing power and idle processing time on multi-processor systems, shorterning the jitter time of applications largely. In addition, the scheduling scheme obtains good scalability.