SoftMoW

Abstract

The current 4G LTE network architecture is organized into very large and rigid regions, each with an access edge consisting of only base stations and an Internet edge comprised of centralized packet gateways (PGWs) that enforce almost all network policies. In this architecture, there is minimal interaction among regions other than interference management at the edge, and all users’ outgoing traffic must traverse a PGW and possibly go through the Internet, even if the other endpoint is served by a close base station in the neighboring regions. The centralized policy enforcement, rigid organization of large regions, and a lack of inter-region interaction make the current cellular network architecture incredibly inflexible and inefficient. First, a recent study [1] shows that the lack of a sufficiently close PGW is a major cause of path inflation, suboptimal routing, and QoS degradation in large carriers. Second, there is no support or simple solutions for IP-based mobility between regions (“inter-PGWs”). Thus, users crossing regions experience service interruption [2]. Third, the sheer amount of traffic and centralized policy enforcement at PGWs, and the inability to directly route traffic between regions take a heavy toll on the scalability and reliability of PGWs and the cellular architecture as a whole. Fourth, with the exponential growth of mobile data and rapidly changing traffic patterns, the current architecture is ill-suited to adapt to the rise of new applications such as machine-to-machine (e.g., connected vehicles, telehealth) and bandwidth-intensive applications. Rather than organizing mobile wide area networks as rigid regions with no direct traffic transit, we argue that the cellular networks should have a fully connected core topology, small logical regions, and more egress points. In addition, operators should leverage software defined networking to manage the entire network with a logically-centralized controller. The controller directs traffic through efficient network paths that might cross region boundaries, supports and optimizes inter-region handoffs, and dynamically adapts to traffic patterns with efficient inter-region traffic engineering. Such an architecture raises unique scalability challenges in comparison with data-center and enterprise networks due to the geographically distributed nature of mobile WANs. Indeed, a logically-centralized controller in one pointof-presence with a flat architecture quickly becomes infeasible, if the mobile WAN spans a large region. This is due to the high latency between the controller and the data plane switches, the amount of signaling load from mobile users, and the very high number cellular handoffs.