Enabling Adaptive Task Dispatching for Real-time Video Analytics in Uncertain Edge Environments

Recently the edge-assisted method has been proposed as a promising technique to deliver fast video analytics by partitioning frames into multiple blocks for parallel execution in edge servers. However, the problem of dispatching tasks over edge servers becomes challenging in uncertain edge environments provided by different operators, considering that resource availability is not prior knowledge and changes in unpredictable ways. This paper proposes an Adaptive Task Dispatching Scheme (ATDS) for coping with the above problems. We formulate the task dispatching problem as an online learning problem with personalized goals. We propose a Multi-armed bandit-based algorithm to solve it. Simulation results show the effectiveness of the proposed method.