Novas: Tackling Online Dynamic Video Analytics with Service Adaptation at Mobile Edge Servers

Abstract

Video analytics at mobile edge servers offers signif-icant benefits like reduced response time and enhanced privacy.However, guaranteeing various quality-of-service (QoS) require-ments of dynamic video analysis requests on heterogeneous edgedevices remains challenging. In this paper, we propose a scalableonline video analytics scheme, called Novas, which automaticallymakes precise service configuration adjustments upon constantvideo content changes. Specifically, Novas leverages the filteredconfidence sum and a two-window t-test to online detect accuracyfluctuations without ground truth information. In such cases, Novas efficiently estimates the performance of all potentialservice configurations through a singular value decomposition(SVD)-based collaborative filtering method. Finally, given theNP-hardness of the optimal scheduling problem, a heuristicscheduling strategy that maximizes the minimum remainingresources is devised to schedule the most suitable configurationsto servers for execution. We evaluate the effectiveness of Novasthrough extensive hybrid experiments conducted on a dedicatedtestbed. Results show that Novas can achieve a substantialover 27×improvement in satisfying the accuracy requirementscompared with existing methods adopting fixed configurations,while ensuring latency requirements. Moreover, Novas improvesthe goodput of the system by an average of 37.86% comparedto existing state-of-the-art scheduling solutions.