Toward Power-Efficient Backbone Routers
Jianyuan Lu,Liang Liu,Jun (Jim) Xu,Bin Liu
DOI: https://doi.org/10.1145/3003977.3004003
2016-01-01
Abstract:Recently, a new design framework, called GreenRouter, has been proposed to reduce the power consumption of backbone routers. In a GreenRouter, a line card is partitioned into two functional parts, namely, an InTerFace (ITF) part that is relatively much lighter and a Processing Engine (PE) part that is relatively much heavier, in power consumption. This partitioning allows ITFs to share the collective processing capability of PEs, which in turn allows a significant percentage of PEs to be put into sleep mode (to save energy) during periods of light link utilizations. In this paper, we study how ITFs can distribute traffic flows to the PEs so that the offered loads on all active PEs are near-perfectly balanced over time, and kept close to a target load (say 90%), so that the number of active PEs can be minimized. Since Green- Router's original solution to this problem is quite crude,we propose a principled solution that has much lower system overheads and achieves better load-balance. Throughboth simulation studies and rigorous analyses, we show our solution can, with high probability, rapidly restore the nearperfect load balance among active PEs, after each PE overload event.