Towards an Automatic Parameter-Tuning Framework for Cost Optimization on Video Encoding Cloud

Abstract

The emergence of cloud encoding services facilitates many content owners, such as the online video vendors, to transcode their digital videos without infrastructure setup. Such service provider charges the customers only based on their resource consumption. For both the service provider and customers, lowering the resource consumption while maintaining the quality is valuable and desirable. Thus, to choose a cost-effective encoding parameter, configuration is essential and challenging due to the tradeoff between bitrate, encoding speed, and resulting quality. In this paper, we explore the feasibility of an automatic parameter-tuning framework, based on which the above objective can be achieved. We introduce a simple service model, which combines the bitrate and encoding speed into a single value: encoding cost. Then, we conduct an empirical study to examine the relationship between the encoding cost and various parameter settings. Our experiment is based on the one-pass Constant Rate Factor method in x264, which can achieve relatively stable perceptive quality, and we vary each parameter we choose to observe how the encoding cost changes. The experiment results show that the tested parameters can be independently tuned to minimize the encoding cost, which makes the automatic parameter-tuning framework feasible and promising for optimizing the cost on video encoding cloud.