A two-phase multipathing scheme based on genetic algorithm for data center networking

Data centers for cloud computing should allocate services with different traffic patterns, provide high data transfer capacity and link fault tolerance. Data center network topologies provide physical connection redundancy, which forwarding mechanisms avail to generate multiple paths. In this paper, we divide multipathing into two phases: (i) Configuration phase based on genetic algorithms to minimize path lengths and maximize link usage diversity; (ii) Path selection phase based on heuristics to minimize path reuse. The proposed multipathing scheme implements minimal modification in infrastructure. Our proposal only requires common network devices features and it avoids any tenant modification. We develop a flow simulator to evaluate multipathing techniques. The simulations model flow behaviors in different data center scenarios and compares the proposed scheme with multipathing techniques in literature. The results show the proposed scheme enhances transmission rates, even in the highest network utilization scenarios.