Building Scalable and Adaptive Garbage Collector for Server Systems

国家科学基金会分布式系统研究CISE/ cns摘要提案号：0411043主要研究者：Srisa-an， WitawasINSTITUTION机构：为服务器系统构建可扩展和自适应的垃圾收集器提案标题：内布拉斯加州大学林肯分校行业观察家预测，随着越来越多的公司部署网络服务，使应用程序或信息在互联网上可用，到2007年底，应用服务器将成为每年210亿美元的业务。面向对象语言是目前开发这些应用程序的首选。面向对象语言，如Java和计算平台，如微软。. NET采用了垃圾回收作为管理动态内存的方式。然而，这种采用至少带来了两个重大挑战：(1)服务器应用程序是多线程的，是为多处理器系统设计的；(2)与桌面应用程序不同，服务器系统可用的工作负载和资源可能差别很大。本研究通过引入一种新的垃圾收集系统来解决这些问题，该系统可以有效地动态响应服务器应用程序的特定需求和挑战。所建议的垃圾收集方案是完全可伸缩和线程安全的。它还可以区分本地和远程对象，以便根据局部性实现不同的收集策略。此外，垃圾收集器是资源感知的，也就是说，它可以根据服务器上资源的可用性调整分析和垃圾收集策略。因此，所提出的工作极大地提高了应用服务器、分布式应用程序和web服务的性能，这些都是每天为数百万用户提供服务的普及计算应用程序的基础。Brett D. fleisch CISE/ cns2004年6月23日

National Science FoundationDistributed Systems Research CISE/CNSABSTRACTPROPOSAL NUMBER: 0411043PRINCIPAL INVESTIGATOR: Srisa-an, WitawasINSTITUTION: Building Scalable and Adaptive Garbage Collector for Server SystemsPROPOSAL TITLE: University of Nebraska-LincolnIndustry observers have predicted that application servers will become a 21 billion dollars per year business by the end of 2007 as more companies deploy web services to make applications or information available across the Internet. Object-oriented languages are currently the choice for the development of those applications. Object-oriented languages such as Java and computing platform such Microsoft .NET have adopted garbage collec-tion as the way to manage dynamic memory. However, this adoption poses at least two significant challenges: (1) server applications are multithreaded and are designed for mul-tiprocessor systems and (2) unlike desktop applications, the workloads and resources available for server systems can vary tremendously.This research addresses these issues by introducing a novel garbage collection system that can efficiently and dynamically respond to the specific needs and challenges of server applications. The proposed garbage collection scheme is fully scalable and thread-safe. It also discriminates between local and remote objects so that differentiated collec-tion strategies can be implemented according to locality. In addition, the garbage collec-tor is resource cognizant, that is, it can adjust the profiling and garbage collection policy based on the availability of resources on the server. As a result, the proposed work greatly improves the performance of application servers, distributed applications, and web services, which are the foundations of pervasive computing applications that already provide services for millions of users each day.Dr. Brett D. FleischProgram Director, CISE/CNSJune 23, 2004..