CAJ | 학술논문

针对当前Hadoop集群自带的任务级调度分配方法在实际处理作业时存在资源分配不均的问题，提出了一种基于权值的任务调度分配算法。该算法结合节点当前的负载状态、节点物理性能和任务优先级等作为依据，通过权值排序当前的作业队列并将空闲资源优先分配给权值高的任务，从而实现运行过程中作业任务的自适应动态调度。实验结果表明，改进算法相比原来的FIFO算法有30％的性能提升。
침대당전Hadoop집군자대적임무급조도분배방법재실제처리작업시존재자원분배불균적문제，제출료일충기우권치적임무조도분배산법。해산법결합절점당전적부재상태、절점물이성능화임무우선급등작위의거，통과권치배서당전적작업대렬병장공한자원우선분배급권치고적임무，종이실현운행과정중작업임무적자괄응동태조도。실험결과표명，개진산법상비원래적FIFO산법유30％적성능제승。
To address the problem of unfair resource assignment in task scheduling method on Hadoop platform, this paper proposed an improved scheduling algorithm based on weight .The algorithm considered the current node state, node physical properties as well as job priority to queue the job , and then sorted the free resource to the first job task based on the weight to implement adaptive dynamic task scheduling during system operation.The experimental results show that it gives about 30% performance better than FIFO algorithm.