Selective Forwarding Unit Placement in Edge Computing Based on DQN for Real-Time Communications

Abstract

In recent years, there has been a dramatic increase in demand for telepresence such as online education and video conferencing. These applications typically make use of a cloud- based selective forwarding unit (SFU) for video distribution. With the assistance of network virtualization technologies, the SFU s can also be placed on edge computing (EC) servers, benefiting bandwidth relief and latency reduction. In this study, the SFU placement policy is modeled as a joint optimization problem for achieving high Quality of Service (QoS) that integrates end-to-end delay and bit rate as QoS metrics. The limitation and disparity in compute and network resources among EC servers are considered. To tackle the above optimization problem, an EC server control method based on Deep Reinforcement Learning (DRL) is proposed. Simulations with and without single-hop SFU cascade demonstrate that the proposed method can effectively improve the bit rate and reduce latency, individually or simultaneously. Additionally, novel modelings of round-trip time and stream processing time are introduced.