Flexible Network Functions in Edge Clouds: Enhancing Processing Capabilities With StateProc

Mobile Edge Computing (MEC) brings computation close to end-users to support latency-sensitive applications, such as tele-manipulation and autonomous driving. While Network Function Virtualization (NFV) facilitates the deployment of these services, MEC applications must still satisfy strict requirements related to mobility, fault tolerance, scalability, and low-latency processing. Fulfilling these requirements often compels Network Functions (NFs) to incorporate custom processing capabilities beyond their initial design. However, existing approaches frequently fall short in supporting these additional custom processing capabilities in NFs, particularly when such capabilities rely on the NF states and their transfer, which often requires extensive modifications to source code. To address these challenges, we propose StateProc, a flexible framework that empowers stateful NFs with custom processing capabilities through optimized, low-latency state transfer. StateProc introduces a controller-side processing manager that coordinates the distributed custom processing consistency and a portable shared library that is embedded within the data plane NFs to execute custom processing with minimal code intrusion. Furthermore, StateProc minimizes the state transfer latency by integrating state priming, aggregation, and control thinning. We evaluate StateProc on a practical testbed using two representative MEC use cases: robot collaboration and network-assisted autonomous driving. In the robotics use case, employing selective state transfer with adaptive deadbands as custom processing improves the time-average of the linear prediction accuracy up to five-fold compared to conventional approaches. In the autonomous driving use case, applying delta-encoding via custom processing reduces the state transfer time by 40–60% compared to the baseline.