Evaluating Energy Efficiency of Gigabit Ethernet and Infiniband Software Stacks in Data Centres

Reducing energy consumption has become a key issue for data centres, not only because of economical benefits but also for environmental and marketing reasons. Many approaches tackle this problem from the point of view of different hardware components, such as CPUs, storage and network interface cards (NIC). To this date, few works focused on the energy consumption of network transfers at the software level comprising their complete stacks with different energy characteristics, and the way the NIC selection impacts the energy consumption of applications. Since data centres often install multiple NICs on each node, investigating and comparing them at the software level has high potential to enhance the energy efficiency of applications on Cloud infrastructures. We present a comparative analysis of the energy consumption of the software stack of two of today's most used NICs in data centres, Ethernet and Infiniband. We carefully design for this purpose a set of benchmark experiments to assess the impact of different traffic patterns and interface settings on energy consumption. Using our benchmark results, we derive an energy consumption model for network transfers and evaluate its accuracy for a virtual machine migration scenario. Finally, we propose guidelines for NIC selection from an energy efficiency perspective for different application classes.(VLID)2215294Accepted versio