Joint optimization of energy saving and load balancing for data center networks based on software defined networks

Abstract

Summary To meet the surging demand for artificial intelligence and cloud service, data centers have been expanding rapidly on recent years. Therefore, data center networks have received great attention recently and more challenges gradually emerged. The exiting technology of data center networks (DCNs) has presented two problems: high energy consumption and network load imbalance. Traditionally, the middle‐box hardware is dedicated and complexly merged, such as network load balancer and network energy optimizer. As an emerging architecture, software defined networks (SDNs) brings an opportunity to accomplish load balancing and energy optimization simultaneously with its characteristics. In this article, we propose a traffic flow management strategy which jointly considers energy optimization and load balancing. The strategy forwards traffic flows with maximum available bandwidth multipath routing to balance network load. We minimize activated links and switches to save energy by scheduling traffic flows. We jointly formulate these as an integer linear programming (ILP) problem. We propose a heuristic algorithm to handle the problem. The full simulation results reveal the high efficiency of our algorithm and the coexistence of network energy optimization and load balancing.

Modeling the correlation between the workload and the power consumed by a server using stochastic and non‐parametric approaches

“…Currently, there is a plenty of approaches aiming at ensuring energy efficiency, including scheduling, 9,10 consolidation, 11,12 and load balancing 13 . These techniques require a comprehension of the different cost aspects.…”

“…7,8 Currently, there is a plenty of approaches aiming at ensuring energy efficiency, including scheduling, 9,10 consolidation, 11,12 and load balancing. 13 These techniques require a comprehension of the different cost aspects. A major factor is the estimation of the energy consumption of the VMs or the server according to the workload to be performed.…”