This study evaluates several state-of-the-art network stacks for running compute-intensive 5G control plane VNFs such as AMF and AUSF. Results show that popular userspace kernel bypass network stacks outperform Linux kernel stack on IO intensive VNFs but the gap narrows for CPU-intensive VNFs. The work emphasizes the importance of traffic partitioning at subscriber granularity for scalability. It also profiles the CPU usage breakdown revealing network stack overheads are relatively low compared to VNF computation.
Summary: This paper compares various state synchronization strategies in a distributed LTE Evolved Packet Core (EPC) implemented as virtualized network functions. It studies designs ranging from no synchronization, synchronization at session boundaries, to synchronization after every message, analyzing throughput and latency trade-offs. Synchronization after every message has prohibitive overhead, while session granularity offers a balance between fault tolerance and performance. The authors provide open-source implementations and experimental results guiding EPC design choices.
Building on earlier work, this technical report investigates sophisticated control algorithms to manage web request queues under extreme overload, using user behavior modeling, simulation, and guidelines for real-world deployment in scalable web services.
The paper proposes models for devolving and redeeming mobile data usage, allowing users to trade data balance efficiently. Analytical and empirical results show how such systems can optimize network resource usage, reduce cost, and increase user flexibility in mobile data plans.
