网格环境中任务调度算法的研究

随着计算机网络技术的成熟,基于网络的高性能计算的进一步发展,出现了网格系统,网格是一种异构的计算环境。首先展示了在异构平台上对于常规的计算和数据,最优分配将是一个NPC问题,寻求有效的静态调度将是困难的,动态策略应当是有前途的,因为机器负载是自我管理,甚至自我均衡,尽管任一机器是异构的。然而动态策略在异构平台上真正的敌人是数据依赖,可能将导致速度被拖为最慢的处理器速度。因此,对于网格系统的异构平台,调度方案应该是动静结合,处理器数据多样性的问题能够通过在每个可识别静态相之间重映像数据和计算来实现,采用ScaLA-PACK的块模式外部乘积算法,将问题抽象为矩形块的分配,给出了完全NP启发式算法的MM乘的最优结果,并且给出研究重映像的一个统一策略。最后,做了一个10个节点的异构平台的实验,展示了结果的实用性。

Job Scheduling Algorithm Research on Grid Environment

Along with the maturity of computer network tec hn ology,the high performance computing based on network is evolving into grid.A grid system is composed geographically distributed resources(include hosts,dat abase,and peripheral devices,etc.)that can be reached over the network.we deal with algorithmic issues on grid.we show that static scheduling and load-balancing strategies are absolutely needed to achieve good performances.The pro blem is to load balance the work with different-speed resources while minimizin g the communication volume.Technically,we state several NP-completeness heur istics results on MMM algorithms that demonstrate the intrinsic difficulty of s tatic load-balancing on heterogeneous.Experiment with grid shows very promis ing results.