Edge Computing for Real-Time Video Data Analysis: An Online Resource Allocation Solution

The real-time analysis of video data obtained from mobile devices requires a significant amount of computing resources. By leveraging edge computing, we can transfer computationally intensive tasks to nearby edge servers, thereby reducing the computational burden on resource-limited endpoint devices and decreasing the long latency associated with transferring data to the cloud. Additionally, we can introduce software-defined networking and network function virtualization technologies to restructure the edge computing network. When transferring video data from multiple endpoint devices to edge servers, the allocation of bandwidth and computing resources significantly affects key performance metrics, such as the amount of data successfully processed in real-time tasks. The time-varying channel state also has a significant impact on decision-making. Therefore, we propose to design an online resource allocation solution aimed at maximizing the long-term average successful processing of real-time tasks. This problem is modeled within the framework of Markov decision processes and an online solution using the Asynchronous Advantage Actor-Critic (A3C) algorithm is proposed. Extensive simulation results demonstrate that the proposed solution outperforms four other baseline methods, achieving higher real-time task success rates.