Coordinated Power Management in Data Center Networks

In this paper, we present a coordinated power management technique for data center networks (DCN) which we have implemented in the popular network simulator NS3. The proposed mechanism includes three key entities: (a) a local controller (LC) that manages traffic at each DCN node (switches and routers), (b) a global controller (GC) that has a global view of the DCN and provides hints to the LC's based on this knowledge, and (c) a topology aware user request assignment controller (RAC) that controls placement of the external requests on the endpoint hosts based on hints from the GC. The main goal of the coordination from the energy management perspective is to properly direct and consolidate network traffic to maximize low power (or “sleep”) opportunities for the network interfaces while avoiding link congestion. In the process of testing out our mechanisms, we have enhanced the popular NS3 package for network modeling. The key enhancements are in the accounting of energy consumption at various levels and enhanced traffic routing mechanisms. We show that these mechanisms can reduce the power consumption by up to ∼40% in the common fat-tree based DCNs using the low power idle (LPI) feature of the Ethernet. We also apply our mechanism to the Hypercube network and show that it too can save a significant amount of energy.