Maximum Flow Minimum Energy routing for ExaScale Cloud Computing systems

This paper explores the feasibility of Exascale Cloud Computing over the Nobel-EU (European Union) IP backbone network. The EU represents an attractive environment for Exascale Cloud Computing, due to its well-established IP infrastructure with relatively low edge distances. An Exascale Cloud Computing system linking hundreds of data-centers can interconnect potentially millions of servers, considerably larger than the world's most powerful supercomputer. However, the US Dept. of Energy has recently concluded that cloud computing is too expensive and suitable only for scientific applications with minimal communications requirements. This paper explores a Future-Internet network to address these concerns. The network combines several technologies; (i) a QoS-aware router scheduling algorithm, (ii) a flow resource-reservation signalling algorithm, and (iii) flow aggregation, to establish highly-efficient guaranteed-rate connections between data-centers. A Maximum-Flow Minimum-Energy routing algorithm is used to route high-capacity ”trunks” between data-centers with minimum energy requirements. The traffic flows between data-centers are aggregated onto the trunks, to achieve exceptionally low latencies with improved link utilizations and energy-efficiencies. The high packet loss rates and large queueing delays of traditional Best-Effort Internet are eliminated, and the latency over the cloud is reduced to the fiber latency. The average fiber latencies over the Nobel-EU network are a few milliseconds, and they can be mitigated with multithreading. The bisection bandwidth of the cloud can be improved to match the demand by activating dark fiber. By lowering latencies, the utilization of the Internet data-centers will also increase, further improving overall energy efficiency. The proposed Future Internet network can improve the utilization and energy efficiency of both the communications and computations in Cloud Computing systems.