Evaluation of Linux Scheduler Algorithms for Low Latency Control Plane Applications

Sammanfattning

This thesis investigates the performance of Linux scheduling algorithms, with a focus on the "Earliest Eligible Virtual Deadline First" (EEVDF) algorithm. The study is conducted within the context of low-latency control plane applications in 5G cellular networks, which are tasked with managing a high volume of control signals without disrupting the data flow. EEVDF was recently integrated into the Linux kernel as the default, general-purpose scheduling policy starting from version 6.6, with the promise of enhancing performance for latency-sensitive tasks. This thesis aims to evaluate whether EEVDF can meet the low-latency requirements of Radio Access Networks (RANs), comparing its performance against other scheduling algorithms such as Completely Fair Scheduler (CFS), Round Robin (RR), and First In First Out (FIFO). The performance evaluation was carried out using both synthetic benchmarks and a real-world application, with a specific focus on latency. To simulate a 5G control plane application, a dedicated test platform was developed, and multiple test scenarios were measured on different Linux kernel versions using different scheduling policies. Results indicate that the newer kernel version, featuring EEVDF, provides improvements in both the synthetic benchmark and the real-world application, particularly in terms of tail latencies. However, in certain cases, median latencies exhibited minor regressions. Overall, this thesis demonstrates the effectiveness of EEVDF in reducing latency for control plane applications in 5G networks and presents it as a viable alternative to real-time schedulers for this application. Additionally, this thesis lays the foundation for further research on tuning EEVDF for specific application requirements.