Horde:面向多核集群的并行编程框架  

Horde:A framework for parallel programming on multi-core clusters

作　　者：薛巍[1] 张凯[1] 陈康[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《清华大学学报（自然科学版）》2011年第11期1663-1667,1672,共6页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(NSFC60973143);国家"九七三"重点基础研究发展计划(2007CB310900);国家"八六三"高技术项目(2008AA01Z112)

摘　　要：并行程序可以充分发掘硬件计算能力并提高程序性能,但是在多核集群环境中编写并行程序十分复杂。该文提出了面向多核集群的并行编程框架,Horde。Horde提供了一组简单易用的消息传递接口和事件驱动(event-driven)编程模型,用以帮助程序员表达算法逻辑中潜在的并行性,将计算分解与底层硬件结构去耦合,从而简化编写并行程序的复杂度,灵活地在不同的底层结构的集群上进行映射并能保持良好的性能。此外,Horde也提供了有效的任务对象迁移机制,可以实现动态负载均衡与在线容错。在128核集群上的实验表明:Horde可以有效执行并行程序,并且可以实现高效的任务对象迁移。Parallel programming hardware to improve performance. utilizes the capacity of parallel However, parallel applications are difficult to program on multi-core clusters. This paper presents a framework for parallel programming on mult？core clusters called Horde. This framework provides a set of easy to use message-passing interfaces and an event driven programming model while helps programmers express parallelisms in the application level and decouple the computational decomposition strategy from the hardware architecture. As such, Horde releases programmers from the difficulties of building complex parallel programs and accommodates different infrastructures while maintaining reasonable performance. Horde also provides task-object migration, which is the key technology for dynamic load balancing and fault tolerance. Tests on a 128-core cluster demonstrate that this system enables high performance parallel programs as well as effective job migration.

关 键 词：分布式操作系统 并行编程 多核集群 容错 

分 类 号：TP316.4[自动化与计算机技术—计算机软件与理论]