并行分布计算中随机任务负载的分配

Distribution of Random Workloads in Parallel and Distributed Computing

为了提高并行分布计算的执行效率,必须尽量减少处理结点间的通讯和平衡处理结点间的计算成本,这两个目标看上去是冲突的。从减少通讯量角度出发,应把所有任务放在同一处理结点上,但很明显,这样会造成系统资源的极大浪费而且各处理结点计算成本极不平衡;从求得处理结点间的计算成本平衡及提高资源利用率出发,应把任务均匀分配到各处理结点上,但这样有可能大大增加处理结点间的通讯。因此,并行分布系统中仟务分配的目标就是寻找一个任务分配方案,使得整个任务负载的执行时间最短。

As one of the hard problems in Parallel Distributed Computing (PDC) ,task distribution has great influence on the execution efficiency of PDC.In this paper,we first discuss the distribution model of random workloads ,which is engaged to make trade-off between the balancing of the execution costs among processors and the reducing of communication and synchronization overheads. Finally we analyze the PDC meaning of the model and discuss the way of assigning these random workloads to processors.