Tensor-Based Rule-Space Management System in SDN

This paper presents a tensor-based rule-space management (TERM) system for improving the available capacity of switches in software defined networking (SDN). Limited storage capacity of switches is a key challenge in SDN as the switches use ternary content addressable memories having very low capacity. Low rule storage capacity eventually leads to the high number of Packet-In messages and control plane overloading. The challenge is to design a dynamic scheme to store a large number of heterogeneous flow-rules in SDN switches and reduce the number of Packet-In messages. To address this problem, we apply the concept of tensor decomposition in order to aggregate flow-rules. In addition, we employ a rule caching mechanism for better throughput. Simulation results show the efficiency of TERM in terms of reduction in the number of Packet-In messages. TERM reduces the Packet-In message count by
${{57.78\%}}$
than the flow aggregation approach proposed in the existing literature.