CSR--PDOS: Memory Efficient Garbage Collection Framework for Java Server Applications

CSR--PDOS：Java 服务器应用程序的内存高效垃圾收集框架

• 批准号: 0720757
• 负责人: Witawas Srisa-an
• 金额: $ 30万
• 依托单位: University of Nebraska-Lincoln
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2007
• 资助国家: 美国
• 起止时间: 2007-09-01 至 2010-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0720757&HistoricalAwards=false
• 关键词: CSR; PDOS; Memory; Efficient; Garbage

Applications utilizing garbage collection often require up to five times more heap memory than similar applications using explicit memory management. In large server applications, such poor memory utilization can cause the throughput performance to degrade ungracefully, leading to unstable systems and unexpected failures.The goal of this project is to develop a memory-efficient garbage collection framework for these servers. The framework has three components not existing in today's commercial garbage collectors: compartmental heaps, phase-based garbage collection invocation policy, and priority page eviction policy. Collectively, these three components allow any garbage collectors utilizing our framework to perform more efficiently while requiring less memory. The ultimate result is an increase in the throughput performance and a predictable and graceful throughput degradation behavior of these servers under extreme memory demands.The success of our proposed work will help promote the transfer of the technologies developed into practice through the industry collaborations and into the classroom through the various related courses. In a broader context, the proposed work will improve the performance and robustness of server applications including application servers and web services, which are the foundations of web based applications serving millions of users each day.

使用垃圾收集的应用程序通常需要比使用显式内存管理的类似应用程序多五倍的堆内存。在大型服务器应用程序中，如此糟糕的内存利用率可能会导致吞吐量性能急剧下降，从而导致系统不稳定和意外故障。该项目的目标是为这些服务器开发内存高效的垃圾收集框架。该框架具有当今商业垃圾收集器中不存在的三个组件：分区堆、基于阶段的垃圾收集调用策略和优先级页面驱逐策略。总的来说，这三个组件允许任何垃圾收集器利用我们的框架更有效地执行，同时需要更少的内存。最终结果是吞吐量性能的提高以及这些服务器在极端内存需求下的可预测且优雅的吞吐量下降行为。我们提出的工作的成功将有助于促进通过行业合作将开发的技术转移到实践中，并通过各种相关课程转移到课堂中。在更广泛的背景下，拟议的工作将提高服务器应用程序的性能和稳健性，包括应用程序服务器和 Web 服务，它们是每天为数百万用户提供服务的基于 Web 的应用程序的基础。