Understanding the overhead of large layer 2 data center networking: Measurement and analysis of TRILL as an example

Highly virtualized data center have placed many new and unique requirements on the networking fabric. Conventional network protocols limit the scale, latency, throughput of cloud networks. To solve these problems, the large layer 2 technology is proposed. The large layer 2 technology such as TRILL and SPB has obvious advantages in scalability, convergence and resource utilization compared with the traditional layer 2 technology. However, the disadvantages of employing the large layer 2 technology are barely addressed. To study the overhead of large layer 2 networking, we setup a data center environment with 6 core switches and conduct experiments to compare the performance of the representative large layer 2 protocol-TRILL and the typical layer 2 technology-STP. We also evaluate the intra-VLAN forwarding and inter-VLAN forwarding of TRILL. From our experiments, it is observed that the forwarding efficiency of TRILL is 30.16% lower than STP on average. It is also observed that the inter-VLAN traffic forwarding of TRILL averagely costs about 54% more time and consumes about 5% more CPU than intra-VLAN with a fixed packet generation rate. This is the first study towards unveiling the inefficiency of large layer 2 approach in data center networks as far as we know.