Dual-Layer Locality-Aware Optical Interconnection Architecture for Latency-Critical Resource Disaggregation Environments
Abstract
Significant research efforts, both industrial and academic, have been committed in the direction of Rack-scale computing through resource disaggregation, that aims to increase resource utilization at a reduced energy and cost envelope. However, the realization of resource disaggregation necessitates an underlying network infrastructure that can compete with a challenging set of requirements including low-latency performance and high-port count connectivity, as well as high data-rate operation. At the same time, it is crucial for the interconnection architecture to be able to accommodate efficient delivery of traffic with different locality characteristics. We propose a dual-layer locality-aware optical interconnection architecture for disaggregated Data Centers by combining the STREAMS silicon-based on-board communication paradigm with the disaggregation-oriented Hipoλaos high-port count switch. Simulation evaluation of a 256-node disaggregated system, comprising 32 optically-interconnected 8-socket boards, revealed up to 100% throughput and mean, p99 latencies not higher than 335 nsec and 610 nsec, respectively, when a 50:50 ratio between on- and off-board traffic is employed. Evaluation of the same layout with 75:25 on-/off-board traffic yields even lower mean and p99 latency at 210 ns and 553 ns, respectively.
Origin | Files produced by the author(s) |
---|