Software Defined Networking (SDN) utilises the concept of centralised command and control to significantly improve network manageability. However, this is at the expense of performance, with the controller(s) now being a bottleneck. This paper further develops the authors' concept of using a Preemptive Flow Installation Mechanism (PFIM) to improve SDN performance for periodic traffic by limiting the number of controller look-ups through predicting when a flow is likely to need a particular rule. Unlike previous work in this area, the paper uses real SDN hardware and a series of experiments containing varying amounts of temporally predictable and random flows, along with a modified version of the POX controller. The work goes on to discuss issues found pairing the controller with hardware, before presenting results that demonstrate that the system can detect majority of temporally periodic flows and preemptively install a flow rule, facilitating the performance comparable to layer two switching speeds. The paper also suggests that the performance of the modified controller is approximately equivalent to the unmodified version when dealing with non-periodic traffic.
KEYWORDSflow preemption, SDN controller, SDN testbed, software defined networking
INTRODUCTIONSoftware Defined Networking (SDN) differs from the traditional approach to network design by abstracting the control element 1 away from individual data forwarding devices 2 or switches. Control functions are delivered to each device via centralised controllers by means of flow rules. As data is forwarded by switch devices that are essentially dumb, the costly, often proprietary, on-board control and management facilities can be dispensed with. This not only allows a more abstracted and controllable network, but has the potential to promote multi-vendor networks and limits the dominance of proprietary standards from any one manufacturer. Such advantages notwithstanding, this centralisation can lead to performance issues, due mainly to the controller being a locus for control functions. Such issues will be the focus of this work.This paper follows on from the works of Bull et al 3,4 in investigating the implications of Preemptive Flow Installation Mechanism (PFIM) in SDN environments. This paper extends this work by:1. Summarising previous performance improvement techniques used in SDN.2. Implementing an improved PFIM algorithm using SDN hardware, rather than in a simulator.3. Discussing the issues of moving from a simulated environment to real SDN hardware.4. Investigating the performance of the PFIM algorithm on real SDN hardware using a number of scenarios, comparing it to a standard SDN controller and a traditional layer two switch. Concurrency Computat Pract Exper. 2020;32:e4531. wileyonlinelibrary.com/journal/cpe FIGURE 1 The Architecture of an SDN
BACKGROUNDTo obtain data forwarding decisions, SDN switches must communicate with centralised controllers. Such controllers transmit instructions back to the switch in the form of flow rules via the southbound i...