面向互联网计算资源共享的自适应调度模型

Adaptive-parallelism scheduling for Internet-based volunteer computing system

在面向互联网的计算资源共享平台中,如何把服务器端的子任务均匀地调度给大规模互联网环境下的志愿机运算是一个重要的研究问题。描述了该平台下的一个自适应并行调度模型。调度器处于服务器端与志愿机之间,缓解服务器端的访问瓶颈;服务器端首先根据调度器的负载对子任务进行第一次分派,在调度器端根据下属的志愿机的软硬信息再分配子任务。通过运行典型的Banch Mark并行程序,把该调度策略与其他策略进行比较,验证了该调度模型针对粗粒度并行的主从(Master-Slave)风格并行应用可以获得较好的性能。

In Internet-oriented computing resource sharing platform,one of the most important topics is how to distribute the subtasks to the volunteers equally.In this paper,a novel scheduling policy is proposed which enables adaptive-parallelism scheduling to be placed between the server and volunteers.The server allocates coarse-grained subtasks to dispatcher alone;volunteers download subtasks from a dispatcher not from the distant server which can decrease the server contention.This approach is also efficient which is strongly supported by its computing capability-based scheduling policy in the volunteers.In order to demonstrate the effectiveness of this approach,two parallel simulation experiments are done.The results obtained from performance analysis show that Master-Slave style parallel applications can benefit a lot from this approach.