Assessing the Cloud-RAN in the Linux Kernel: Sharing Computing and Network Resources

Area of Science:

• Computer Networking
• Telecommunications Engineering
• Edge Computing

Background:

• Cloud-based Radio Access Network (Cloud-RAN) utilizes virtualization for economic and operational efficiency.
• Co-locating multiple virtual Base Band Units (vBBUs) on edge computers can lead to performance degradation due to resource contention.

Purpose of the Study:

• To empirically analyze the performance of vBBUs on a Linux RT-Kernel.
• To investigate the impact of resource sharing on vBBU performance.
• To evaluate CPU management strategies, specifically CPU affinity and isolation, as solutions for performance challenges.

Main Methods:

• Empirical analysis of vBBU performance on a Linux RT-Kernel.
• Evaluation of CPU affinity and CPU isolation techniques.
• Measurement of throughput variability, NACK ratios, and scheduling latency.

Main Results:

• CPU affinity significantly reduced throughput variability by up to 40%.
• CPU affinity led to decreased vBBU's NACK ratios and reduced vBBU scheduling latency.
• Resource sharing with user-space tasks and Kernel threads impacts vBBU performance.

Conclusions:

• CPU management strategies, particularly CPU affinity, can enhance vBBU performance in Cloud-RAN.
• Implementing CPU affinity leads to more efficient and stable network operations.
• The findings affirm the practicality and efficacy of CPU affinity for optimizing Cloud-RAN deployments.