A Dynamic and Complexity Aware cloud scheduling algorithm for video transcoding

Abstract

Cloud video processing and streaming services has to be delivered under heterogeneous network and device environments. Scalable video coding and transcoding are required to serve heterogeneous users. As the task scheduling algorithm pre-configures a Hadoop MapReduce platform with the assumption of homogeneous node processing capability and task complexity, it cannot well accommodate the practical heterogeneous resources and tasks. In this research, we proposed a Dynamic Adjustment Slot and Complexity Aware Scheduler (DASCAS) algorithm to assign tasks under heterogeneous resources and tasks environments. Complexities of decomposed video segments are evaluated for setting task priority. The scheduling algorithm utilizes a speculative mechanism to detect potential late tasks to re-assign to other nodes for fast processing. It also monitors processing status of the distributed computer cluster and dynamically adjust the number of slots for load balance operations. Experiments show that the proposed method can reduce the transcoding time to 14%∼24% smaller and improve the resource utilization rates to 2%∼12% higher.