CAJ | 학술논문

在机群尤其是共享式多用户机群环境中提高非规则问题的并行计算效率是一项具有挑战性的工作.给出了一种适于多核处理器节点构成的机群的层次Work-stealing负载均衡策略(简称HWS),它能够有效利用节点的动态变化的可用计算资源.分析了分治计算问题中的动态性与非规则性,并使用HWS策略对分治计算问题进行了动态负载均衡(DLB)实验.实验表明,HWS在加速比、CPU利用率和负载分布等性能指标上均优于Work-sharing和Satin-CRS策略.
재궤군우기시공향식다용호궤군배경중제고비규칙문제적병행계산효솔시일항구유도전성적공작.급출료일충괄우다핵처리기절점구성적궤군적층차Work-stealing부재균형책략(간칭HWS),타능구유효이용절점적동태변화적가용계산자원.분석료분치계산문제중적동태성여비규칙성,병사용HWS책략대분치계산문제진행료동태부재균형(DLB)실험.실험표명,HWS재가속비、CPU이용솔화부재분포등성능지표상균우우Work-sharing화Satin-CRS책략.
It is a challenging problem for solving non-uniform problems to improve the efficiency of parallel computing in PC cluster system,especially in non-dedicated and multi-user cluster environment,among which dynamic load balancing (DLB) is essential for the efficient use of parallel systems.A hierarchical work-stealing strategy for load balancing is proposed for PC clusters based on multi-core processors,which can make the most of available resources,even if such availability will change dynamically.Some dynamic and non-uniform characteristics of divide-and-conquer computation are discussed,and experiments with scheduling divide-and-conquer programs by work-stealing strategy for load balancing are carried out.Experimental results are provided,and the results show that HWS outperforms work-sharing and Statin-CRS algorithms in a number of performance metrics such as speedup,CPU utilization,load distribution and so on.