Workload scheduling for Nokkhum video surveillance as a service

Abstract

Video Surveillance Systems (VSS) on the Internet as known as Video Surveillance as a Service (VSaaS) or Cloud based Video Surveillance (CVS) systems. Video processing workload analysis has usually employed only one category of static video processing attribute, such as a frame rate with a single frame size on the same computing node specification, but VSaaS must handle a variety of video processing attributes. Also, in a static workload, it is difficult to identify the resource consumption of video processing attributes, especially involving a combination of frame rates and sizes on different computing nodes on virtual or physical machines. Consequently, it is difficult to place a task on a computing node if the resource usage information is unknown to the scheduler. In this paper, the video processing workload characteristics utilize various parameters, such as the type of video processing task, frame rate, frame size, and compute node specification. The analysis results have helped us to design a scheduler that supports different computing node specifications. We explore video processing workload for testing resource usage capacity in several computing nodes, and collect information for the scheduler’s estimation. This paper also proposes a resource estimation module for predicting the video processing resource usage for a new video processing task when there is no matching or close estimation. Furthermore, we suggest scheduler criteria for optimizing system resource usage.