A meta-predictor framework for prefetching in object-based DSMs

作     者：Beyler, Jean Christophe Klemm, Michael Clauss, Philippe Philippsen, Michael 

作者机构：Univ Erlangen Nurnberg Comp Sci Dept 2 D-91058 Erlangen Germany Univ Delaware Dept ECE Newark DE 19711 USA Univ Louis Pasteur Strasbourg ICPS LSIIT F-67400 Illkirch Graffenstaden France 

出 版 物：《CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE》 (并行学和计算：实践与经验)

年 卷 期：2009年第21卷第14期

页      面：1789-1803页

核心收录：

学科分类：08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：prefetching distributed shared memory cluster computing 

摘      要：Dynamic optimizers modify the binary code of programs at runtime by profiling and optimizing certain aspects of the execution. We present a completely software-based framework that dynamically optimizes programs for object-based distributed shared memory (DSM) systems on clusters. In DSM systems, reducing the number of messages between cluster nodes is crucial. Prefetching transfers data in advance from the storage node to the local node so that communication is minimized. Our framework uses a profiler and a dynamic binary rewriter that monitor the access behavior of the application and place prefetches where they are beneficial to speed up the application. In addition, we use two distinct predictors to handle different types of access patterns. A meta-predictor analyzes the memory access behavior and dynamically enables one of the predictors. Our system also adapts the number of prefetches per request to best fit the application s behavior. The evaluation shows that the performance of our system is better than the manual prefetching. The number of messages sent decreases by up to 90%. Performance gains of up to 80% can be observed on benchmarks. Copyright (C) 2009 John Wiley & Sons, Ltd.