Scalable Mobile Edge Computing: A Two-Tier Multi-Site Multi-Server Architecture with Autoscaling and Offloading

Mobile Edge Computing (MEC) provides computation resources within 5G networks hosting applications that are close to a user equipment (UE). For scalability, MEC servers can be placed behind the base stations of an access network (AN) and also inside the core network (CN) of a cellular system, which results in a two-tier architecture. A scalable MEC system reveals a management problem because keeping all servers on as traffic fluctuates wastes operational expenditure. On the other hand, traffic can become unbalanced, with hotspots in some base stations. This work proposes a two-tier multi-site multi-server architecture and integrates Latency Satisfaction Aware Autoscaling (LSAA) and Dynamic Weight Offloading (DWO) to address the above two problems. Offloading is a short-term solution to hotspot traffic, while autoscaling is a long-term solution to traffic fluctuation. A two-tier MEC testbed was implemented in the framework of OpenNESS with 3GPP integration, with experimental comparisons of one-tier vs. two-tier, uniform vs. hotspot, transient vs. persistent hotspot traffic, with or without offloading and autoscaling. Under heavy hotspot traffic, two-tier MEC satisfies 86 percent, 73 percent, and 21 percent traffic with both offloading and autoscaling, offloading only, and without offloading and autoscaling, respectively, while one-tier MEC only satisfies 32 percent, 32 percent, and 21 percent traffic.