Investigating the Performance of Link Aggregation on OpenFlow Switches

OpenFlow technology is emerging rapidly with the potential of increasing the flexibility in network management and operation. The technology introduces the OpenFlow switch, which is a cost-effective but well-functional switch implementation. The switch enables network programmability as well as various ways of forwarding packets. Since the version 1.1, the specification of OpenFlow switch introduces Link Aggregation (LA), which is a basic and important function of conventional switches. Theoretically, LA automatically distributes and balances the traffic across the links within a Link Aggregation Group (LAG), hence improves network resilience and throughput. However, the performance of LA on OpenFlow switches has not been well investigated. In this paper, we extensively explore the operation of LA on OpenFlow switches in comparison to the LA in conventional switches. The comparison of two LA implementations has been conducted in a real testbed under the UDP and TCP traffic loads. The testbed includes Pica8 P-3925 switches, which support two modes: an OpenFlow switch (i.e., using Open vSwitch) and a conventional switch (i.e., using the operating system called XorPlus). The evaluation results show that two LA implementations achieve similar performance in improving throughput. However, the XorPlus’s implementation provides a better resilience than the other. Specifically, the LA implementation on XorPlus spends less than 1.49538 seconds to switch the TCP traffic on the faulty link to the other links of a LAG while the switchover time is four times longer on the Open vSwitch. In the case of UDP traffic, the maximum switchover time on the Open vSwitch is twice the one on XorPlus.