Flexible, highly scalable and cost-effective network structures for data centers

The ever-increasing number of online services and tremendous data volume make the data center networks expand rapidly. The data center network must be flexible in scalability and cost-effective in order to respond quickly to the ever changing service requirements. However, the costs of switch-centric topologies are high, while the scalabilities of server-centric topologies are not good. How to build a flexible, highly scalable and cost-effective data center network structure is becoming a very challenging problem. Based on the Cartesian product graph, this paper proposes a new type of data center network structure named SDCCP (Scalable Data Center network structure based on Cartesian Product graphs). It is constructed by using commodity m-port switches and 2-port servers. Based on the m-degree Cartesian product graphs, we can construct different kinds of SDCCP structures which can be expanded in different scales by using the same type of m-port switches. In order to achieve incremental scalability, we further propose an incomplete SDCCP structure. Servers can be gradually added into the incomplete SDCCP structure while maintaining their topological properties unchanged. Compared with Fat-Tree and BCube, the SDCCP reduces the cost by about 30% and 60%, respectively; and reduces the energy consumption by about 27% and 35%, respectively. The performance analyzes and experimental results demonstrate that SDCCP strikes a good balance in terms of flexibility, scalability, cost-effectiveness, and low energy consumption in contrast to the typical data center network structures.