The trade-off between power consumption and latency in computer networks

Abstract

As the power consumed by computer networks is nearly independent from the load, networks consume much more power than necessary when the load is low. We analyze the influence of disabling network components on power consumption and latency of data transfers. We define two power consumption models and compare the power consumption necessary to achieve given upper limits of latency. Our results show that a trade-off between power consumption and latency exists; it is e.g. possible to conserve 39% of power when accepting a 20% latency increase in the nobel-germany network. We conclude that it is possible for networks to adapt to changing demands to conserve power while the latencies increase only slightly. © 2011 IFIP International Federation for Information Processing.