Micro-burst Analysis and Mitigation for 5G Low-latency Communication Services

Micro-bursts are a common problem in data center networks, and many studies have analyzed ways to mitigate micro-burst traffic. However, little attention has been paid to micro-burst impacts in mobile networks even though fifth generation (5G) mobile networks have stringent performance requirements such as low-latency communications. For example, third generation partnership project (3GPP) defines the minimum requirement of end-to-end latency as 5 milliseconds. We need to consider how to avoid micro-bursts in a data center of 5G networks because the end-to-end definition includes not only radio access networks and mobile backhaul but also data center networks deploying user plane function (UPF). 5G networks have some differences from networks in past micro-burst studies. Applications requiring low-latency communications are not backend systems assumed in the previous studies. Augmented reality (AR) and vehicle to everything (V2X) messages are typical delay sensitive applications in 5G networks. Furthermore, the traditional approaches for absorbing micro-burst traffic such as a deep buffer and buffer allocation scheme cannot be used in 5G networks because packet buffering increases queueing latency. In this paper, we analyze micro-burst characteristics in 5G networks in order to clarify what types of applications and what network architecture cause micro-bursts. For mitigating micro-bursts, we evaluate quality of service (QoS) schedulers and window size control of transmission control protocol (TCP). The results show that the TCP window size control can reduce queuing latency to be sub-millisecond without any additional mechanisms or equipment while priority queueing (PQ) and weighted round robin (WRR) schedulers have some limitations.