Fat-trees: universal networks for hardware-efficient supercomputing
IEEE Transactions on Computers
Flattened butterfly: a cost-efficient topology for high-radix networks
Proceedings of the 34th annual international symposium on Computer architecture
MapReduce: simplified data processing on large clusters
Communications of the ACM - 50th anniversary issue: 1958 - 2008
PortLand: a scalable fault-tolerant layer 2 data center network fabric
Proceedings of the ACM SIGCOMM 2009 conference on Data communication
Energy proportional datacenter networks
Proceedings of the 37th annual international symposium on Computer architecture
ElasticTree: saving energy in data center networks
NSDI'10 Proceedings of the 7th USENIX conference on Networked systems design and implementation
Network traffic characteristics of data centers in the wild
IMC '10 Proceedings of the 10th ACM SIGCOMM conference on Internet measurement
High Performance Datacenter Networks: Architectures, Algorithms, & Opportunities
High Performance Datacenter Networks: Architectures, Algorithms, & Opportunities
Minimizing ISP network energy cost: formulation and solutions
IEEE/ACM Transactions on Networking (TON)
Merging traffic to save energy in the enterprise
Proceedings of the 2nd International Conference on Energy-Efficient Computing and Networking
Hi-index | 0.00 |
The problem of reducing energy usage in datacenter networks is an important one. However, we would like to achieve this goal without compromising throughput and loss characteristics of these networks. Studies have shown that data-center networks typically see loads of between 5% -- 25% but the energy draw of these networks is equal to operating them at maximum load. To this end we examine the problem of reducing the energy consumption of datacenter networks by merging traffic. The key idea is that low traffic from N links is merged together to create K ≤ N streams of high traffic. These streams are fed to K switch interfaces which run at maximum rate while the remaining interfaces are switched to the lowest possible rate. We show that this merging can be accomplished with minimal latency and energy costs (less than 0.1W total) while simultaneously allowing us a deterministic way of switching link rates between maximum and minimum. We examine the idea of traffic merging using three different datacenter networks -- flattened butterfly, mesh and hypercube networks. In addition to analysis, we simulate these networks and utilizing previously developed traffic models we show that 49% energy savings are obtained for 5% per-link load while we get 20% savings for a 50% load for the flattened butterfly and somewhat lower savings are obtained for the other two networks. The packet losses are statistically insignificant and the maximum latency increase is less than 3μs. The results show that energy-proportional datacenter networks are indeed possible.