摘要
为实现Hadoop分布式文件系统的负载均衡,并保证较低的负载迁移代价和数据传输代价,提出了确定环境下多阶段多目标(CMM)决策模型.该模型以CPU、内存和磁盘剩余负载能力作为决策条件,以负载均衡效果、负载迁移代价和数据传输代价作为决策目标,依据决策节点间的影响关系构建有向无环图,通过多个决策阶段的决策,并计算方案效用确定最优均衡方案.仿真实验结果表明,基于CMM模型的负载均衡策略能取得较好的负载均衡效果、负载迁移代价和数据传输代价.
In order to balance the load of Hadoop distributed file system with lower load migrating cost and data transmission cost, a certainty multi-stage and multi-object （CMM） decision model was pro- posed. The model is a directed acyclic graph built on decision nodes, which adopts the remaining load capacities of CPU, memory and disk as decision preconditions, and also adopts load balancing effect, load migrating cost and data transmission cost as decision targets. By CMM model, the best balancing plan is determined by selecting results of multiple decision stages and computing the plan usage. Simula- tions show that the CMM based strategy can achieve better load balancing effect, load migrating cost and data transmission cost.
出处
《北京邮电大学学报》
EI
CAS
CSCD
北大核心
2014年第5期20-25,共6页
Journal of Beijing University of Posts and Telecommunications
基金
国家科技重大专项项目(2012ZX03005010-003)
国家高技术研究发展计划项目(2014AA01A706)
关键词
分布式文件系统
负载均衡
决策模型
剩余负载能力
distributed file system
load balancing
decision model
remaining load capacity
