Dynamic resource allocation for cloud-based media processing
Krisantus Sembiring, Andreas Beyer · 2013
As an economic and scalable solution of providing interactive and adaptive media content across different devices, cloud-based media processing has recently attracted lots of attention from both academic and industry. Within a media cloud, a large number of media processing tasks will be dynamically spawn to run in parallel and then share resource with each other in the cloud. In such a dynamic and shared environment, how to appropriately provision cloud resource to different tasks with regards to both system efficiency and quality of service remains a challenging issue. Existing resource allocation algorithms consider generic task and only focus on maximizing system efficiency, but failed to meet the QoS requirement of media tasks. In this paper, we investigate the relationship between the properities of media tasks and their resource comsumption and accordingly propose a dynamic resource allocation solution. Our solution leverages machine learning technique to predict resource requirement and survival functions to prevent QoS degradation.