Dynamic in-network classification for service function chaining ready SDN networks

Service Function Chaining (SFC) paradigm consists in steering traffic flows through an ordered set of Service Functions (SFs) so that to realize complex end to end services. SFC architecture introduces all the logical functions that need to be developed in order to provide the required service. The SFC overlay infrastructure can be built on top of many different underlay network technologies. The high flexibility and centrally controlled feature of Software Defined Networking (SDN), make SDN networks to be a perfect underlay to build the SFC architecture. Due to Ternary Content Address Memory (TCAM) limited size, SDN switches have a limitation in the number of flow rules that can be hosted. This constraint is particularly penalizing in case of the SFC classifier function, since it requires to manage a high number of different flows. The limitation imposed by the TCAM size on the SFC classifier can be a bottleneck for the number of SFC requests that the SDN-based SFC architecture can handle. In this paper we define the Dynamic Chain Request Classification Offloading (D-CRCO) problem, as the one of maximizing the number of accepted SFC requests, having the possibility of: i) implement the SFC classifier also in a node that is internal to the SDN-based SFC domain, and ii) install classification rules in a reactive fashion. Furthermore, we propose the Dynamic Nearest Node (DNN) heuristic to solve the D-CRCO problem. Performance evaluation shows that by using DNN heuristic it is possible to triple the number of accepted requests, with respect to existing solutions.