TOVAC: A Two-Timescale Framework for QoS-Aware Video Analytics With Hierarchical Edge Computing

Abstract

As one of the most representative applications of edge computing, video analytics typically involves multiple vision components in the pipeline, which together determine the quality of service (QoS) for users. By exploiting diverse resource demands of different components, a fine-grained hierarchical orchestration with QoS-aware configuration adaptation can further boost the efficiency of heterogeneous resources in device-edge-cloud architecture. To judiciously match the temporal granularity between the dynamics of video traffic and resource price, we propose TOVAC, a Two-timescale Optimization framework for Video Analytics with hierarchical cross-edge Collaboration. The core idea of TOVAC is to jointly optimize long-term resource costs and QoS (including accuracy and latency) via resource allocations and fine-grained traffic control across long and short timescales. We first decouple the intractable optimization problem based on traffic conservation constraints and propose two-timescale online optimization algorithms under perfect prediction. To further enhance the performance and robustness, we extend our framework to more general cases with imperfect predictions and develop a refined Predictive Resource Allocation algorithm. We rigorously analyze the performance guarantee of our proposed algorithms under any prediction using a unified parameterized competition ratio, and further verify the empirical performance of TOVAC through extensive trace-driven experiments.