Abstract: On a special data-intensive application scenario, users have obvious priority. Based on resource reserve and resource preemption, this paper proposes a scheduling algorithm on user level. This algorithm can improve resource utilization rate, quickly respond to high-priority users’ request, reduce resource preemption cost, and avoid the job with lower priority never being dealt with. Based on dynamic replica creation, this paper proposes another scheduling algorithm on task level. This algorithm can realize data performing locally, slove the load balancing in data access, reduce the response time of job execution.

Key words: resource scheduling, resource reserve, resource preemption, dynamic replica creation, MapReduce

摘要： 针对用户之间具有明显优先级的数据密集型应用场景，提出了基于资源预留和资源抢占的用户级调度算法，该算法在提高资源利用率的前提下，能够快速响应高优先级用户的作业请求，降低资源抢占代价，并避免低优先级作业长时间得不到调度。提出了基于动态副本创建的任务调度算法，该算法实现了数据本地化执行，并解决了热点数据访问带来的负载均衡问题，降低了作业的响应时间。

关键词: 资源调度, 资源预留, 资源抢占, 动态副本创建, MapReduce