2009 Poster Sessions : ElasticTree: Reducing Energy in Data Center Networks

Student Name : Brandon Heller
Advisor : Nick McKeown
Research Areas: Computer Systems
Abstract:
Today's data center networks consume roughly the same amount of energy when idle as they do when every host is sending full-blast. This flat energy profile highlights a missed opportunity to save energy. The desired behavior, called energy proportionality, is achieved when energy costs vary linearly with load. ElasticTree is a network-wide power optimizer that dynamically adjusts the set of active network elements -- links and switches -- to satisfy changing traffic loads. ElasticTree yields savings of up to 60% of network power in a 28,000-node data center, without requiring hardware changes. The prototype leverages the network path control provided by OpenFlow-enabled switches to optimize network power, while maintaining performance and reliability. Our visualization allows anyone to try out new data center traffic patterns with ElasticTree, live.


Bio:
Brandon Heller is a Stanford PhD student in CS, focusing on energy and networking. His research includes projects to measure energy with networks, reduce the energy consumed by data center networks, and use networks as a tool to enable new efficiency optimizations. Brandon holds an MS and BS in Computer Engineering from Washington University in St. Louis.