Abstract
Owing to the demand of high performance computing applications, network throughput becomes a critical bottleneck when data are stored in high I/O performance storage devices, such as non-volatile memory, solid-state disk, and RAM-based disk. To increase network throughput, a brand new technology called Software-defined networking (SDN) has been proposed for high bandwidth applications. Many data centers therefore upgrade their network infrastructures to support the SDN technology. However, the best performance of the SDN technology only can be gained when the SDN system employs an appropriate packet-flow scheduler to serve the applications in the SDN environment. Finding the best settings of the packet-flow scheduler in the SDN environment, this study constructs a large-scale test-bed to simulate the packet flow in the SDN environment. In the test-bed, we investigate the capability of a packet-flow scheduler in the SDN controller while the scheduler employs different scheduling algorithms under different network workloads. The experimental results suggest what kind of scheduling algorithms should be utilized under the network workloads.