机群系统中一种新的自适应结点分配算法:受限负载平衡结点分配(LLBA)

A NEW ADAPTIVE PROCESSOR ALLOCATION ALGORITHM FOR CLUSTER:LIMITED LOAD BALANCING ALLOCATION (LLBA)

在机群系统中结点分配策略根据一定的原则为作业确定运行结点是提高系统性能的关键．通过对机群结点分配策略的研究，作者发现当前基于负载平衡自适应结点分配策略为并行作业选择负载最轻的结点，这不利于系统性能的充分发挥．作者提出了一种新的自适应负载平衡结点分配算法：受限负载平衡结点分配（ＬＬＢＡ）．它的基本思想是为并行作业选取负载相近的计算结点，使得并行作业分配在各结点上的运算部分具有相同的运算速度，减少运算结点的空闲等待时间，提高系统的吞吐率．通过模拟实验，将ＬＬＢＡ和传统的负载平衡自适应结点分配进行比较，ＬＬＢＡ能够明显改善机群系统的性能．

In a cluster system, allocation algorithm chooses processors for jobs. Efficient processor allocation is one of essential components of cluster management system. Through study to the previous researches, it is found that the underlying adaptive allocation policies always select the most lightly loaded processors for parallel jobs. The phenomenon weakens the performances of a cluster system.This paper proposes a novel adaptive allocation algorithm:Limited Load Balancing Allocation (LLBA).LLBA selects the processors with approximately equal processing speed for a parallel job. It can reduce the idle time of processors and increase the throughput of a cluster.Through simulated experiences,it compared LLBA with the traditional load balancing allocation and substituted the conclusion that LLBA can improve the performance of a cluster obviously.