AggFlow: Scalable and Efficient Network Address Virtualization on Software Defined Networking

Abstract

In this paper, we propose AggFlow, a new address virtualization scheme for high scalability and resource efficiency. AggFlow is intended to provide complete address virtualization while incurring low overhead for physical switches and the control channel. To reduce overhead of address virtualization, we propose mapping-less address virtualization. We additionally introduce hop-by-hop-based forwarding, which aggregates flow rules installed in physical switches. This leads to efficient use of ternary content addressable memory and bandwidth of the control channel. Our evaluation using Mininet with simple linear and tree topologies, the flow table size of the core switch is reduced up to 1/16 for both topologies, and the control channel traffic decreases by 51% and 29% respectively. Furthermore, we measure the overhead incurred from AggFlow. Compared to OpenVirteX, AggFlow increases control plane delay by 0.1 ms, only 0.4% increase of CPU usage, and no change of throughput for UDP traffic.