Parallel Clustering for Interference Alignment with Max-Min Fairness in Hybrid Data Center Networks

Data center networks (DCNs) are the essential infrastructure for cloud data centers, while the contradiction between the fixed topology and nonuniform distribution of traffic are causing the coexisting of network congestions and low network utilization in traditional wired DCNs. Thus the hybrid DCNs emerge as an alternative solution to construct flexible wireless topologies according to the dynamic traffic demands. By using wireless links, the DCN’s resource oversubscription problem can be alleviated and the network energy efficiency can be greatly improved. However, due to the dense deployment of wireless access points (APs) in hybrid DCNs, the interference among APs is an obstacle to the wireless network capacity. In this paper a novel clustering scheme for interference alignment (IA) is proposed to guarantee the max-min fairness of APs in the hybrid DCN. Each AP pair is allowed to join in multiple parallel IA clusters, if those clusters are allocated with orthogonal segmented resource, such as orthogonal frequency bands or time slots. The allocated resource proportion for each parallel cluster is determined by solving a linear max-min optimization problem. For the fully connected J user M × M MIMO IC, the max-min normalized degree of freedom (DoF) by parallel clustering is proven to be (2M-1)/J, while for the partially connected IC, we focused on the 2 × 2 case and proposed a heuristic algorithm to find all potential clusters for resource allocation optimization. Theoretical analysis and simulation results have validated that the proposed parallel clustering scheme can achieve better maxmin normalized DoF performance than the single user MIMO and traditional singular IA clustering schemes.