A sociality-aware online AP association algorithm based on spectral clustering

Imbalanced load distribution among multiple APs (access points) has become a serious problem in enterprise WLANs. Load imbalance causes inefficiency of network usage and unreasonable resource allocation. We find that users tend to come and leave together, which can lead to fluctuation of network load. We propose a novel online AP allocation scheme (ASC) to keep load balanced when users come and leave together. We study the homophily of users in social network, where users with tight social relationships will have more similar network usage and more opportunity to leave in union. When users come, we collect user data first, then we adopt the spectral clustering to classify users according to the collected user data. The controller distribute users from the same cluster to different APs. Through the simulation, we find that our scheme improve the overall balancing performance by at least 22% compared with the least load first selection algorithm (LLF) and 58% compared with random selection algorithm (RS).