Title | ||
---|---|---|
Energy-efficient optical HPC and datacenter networks using optimized wavelength channel allocation |
Abstract | ||
---|---|---|
This paper presents a novel energy-efficient all-optical switch architecture for high-performance computing (HPC) systems and datacenter networks (DCNs), that employs the arrayed waveguide grating router (AWGR) and wavelength selective switches (WSSs). To enable flexible wavelength allocation (WA) at switch output ports, we propose an adaptive load-balancing wavelength assignment optimization algorithm. The proposed flexible scheme enhances network resource utilization, and thus significantly alleviates the contention conditions in the network. Further, by developing a tractable analytic model to approximate blocking probability between source-destination node pairs, the flexible allocation scheme can optimally allocate optical buffering using fibre delay lines (FDLs). In simulation experiments, it is shown that the dynamic resource allocation scheme supports high load operation and enhances the network performance and energy efficiency significantly, compared to a uniformly-dimensioned switch, under non-uniform hotspot traffic load. For a target packet loss rate of < 10−6, the proposed architecture achieves a power reduction of 25%, thus enabling increased network scalability. With the addition of a small-scale electronic buffer, packet loss rates may be reduced to < 10-10, without a significant increase in energy consumption. |
Year | DOI | Venue |
---|---|---|
2015 | 10.1109/SPECTS.2015.7285286 | SPECTS@SummerSim |
Keywords | Field | DocType |
HPC, Intra-DCN, OPS, WSS, FDLs, Energy-efficient Networking | Efficient energy use,Computer science,Packet loss,Computer network,Real-time computing,Resource allocation,Hotspot (Wi-Fi),Energy consumption,Wavelength,Scalability,Network performance | Conference |
Citations | PageRank | References |
2 | 0.41 | 9 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Jingyan Wang | 1 | 7 | 2.23 |
Conor McArdle | 2 | 31 | 6.39 |
Liam P. Barry | 3 | 23 | 14.10 |