Linear Programming Approaches for Power Savings in Software-defined Networks (The Extended Version)

TL;DR

This paper introduces four linear programming algorithms for traffic scheduling in SDN that optimize power savings while maintaining quality of service, outperforming traditional shortest path methods.

Contribution

It proposes novel linear programming approaches tailored for SDN traffic scheduling to balance power efficiency and performance, with demonstrated superior results over baseline algorithms.

Findings

All algorithms outperform shortest path scheduling in power savings.

Switches in FatTree networks can save up to 60% power in sleep mode.

Algorithms can achieve at least 15% power savings with minimal performance impact.

Abstract

Software-defined networks have been proposed as a viable solution to decrease the power consumption of the networking component in data center networks. Still the question remains on which scheduling algorithms are most suited to achieve this goal. We propose 4 different linear programming approaches that schedule requested traffic flows on SDN switches according to different objectives. Depending on pre-defined software quality requirements such as delay and performance, a single variation or a combination of variations can be selected to optimize the power saving and the performance metrics. Our simulation results demonstrate that all our algorithm variations outperform the shortest path scheduling algorithm, our baseline on power savings, less or more strongly depending on the power model chosen. We show that in FatTree networks, where switches can save up to 60% of power in sleeping mode, we can achieve 15% minimum improvement assuming a one-to-one traffic scenario. Two of our algorithm variations privilege performance over power saving and still provide around 45% of the maximum achievable savings.