Date of Award
Doctor of Philosophy (PhD)
College of Engineering and Computing
Michael J. Laszlo
Increasing energy costs and environmental issues related to the Internet and wired networks continue to be a major concern. Energy-efficient or power-aware networks continue to gain interest in the research community. Existing energy reduction approaches do not fully address all aspects of the problem. We consider the problem of reducing energy by turning off network links, while achieving acceptable load balance, by adjusting link weights. Changing link weights frequently can cause network oscillation or instability in measuring the resulting traffic load, which is a situation to be avoided. In this research, we optimize two objectives, which are minimizing network power consumption by maximizing utilization of shortest paths, and at the same time achieving load-balance by minimizing network Maximum Link Utilization (MLU).
Research to date has focused on the link level of traffic load balance, to minimize power consumption, while putting less focus on utilizing adaptive strategic techniques that optimize multi-objectives problems. This research developed a new approach that relies on live data collected from wired networks, and performs Multi-Objective Optimization (MOO) using a Non-dominated Sorting Genetic Algorithm (NSGA-II) that applies alternative adaptive strategies in order to optimize those two objectives. We also studied how adding delays between link weights adjustments can alleviate the network oscillation or instability without causing higher network power consumption and imbalanced network traffic.
This work introduced a novel approach to select underutilized links to go to sleep using adaptive strategies of MOO that are aware of traffic changes. Re-computing the algorithm takes less than a minute, while network traffic is frequently updated every few minutes. The hybrid approach that we designed was able to reduce the power consumption by 35.24%, while reducing MLU by 42.86% for specific traffic pattern used in Abilene network topology. For network instability, we introduced sequential_delay and wait_interval delay parameters that are implemented in conjunction with link weight settings. We show reduction of instability measurement from 175% down to 8.6% for Abilene network topology when using a value of 1sec for both sequential_delay and wait_interval delay parameters.
Hatem Yazbek. 2019. Adaptive Strategies of Multi-Objective Optimization for Greener Networks. Doctoral dissertation. Nova Southeastern University. Retrieved from NSUWorks, College of Engineering and Computing. (1096)