Impact of buffer sizing on energy efficiency and performance

Abstract

Internet router buffers are commonly made very large with the aim of minimising losses and maximising link utilisation, supported by the ever-decreasing cost of memory. However, large packet queues not only increase the energy consumption, but can also adversely affect network performance which has recently been attracting attention under 'bufferbloat'. This study compares the buffer-related energy consumption of two models, that is, 'large-buffer with on-off server' and 'small-buffer with always-on server' in the presence of multiplexed on-off traffic and FTP traffic; and demonstrates that the latter model is more energy efficient. Further analysis is then carried out on the 'small-buffer model with always-on server' for the evaluation of its performance against different levels of energy consumption. Relationships among energy, delay, throughput and loss are studied in a range of scenarios, quantifying the variations in performance metrics in relation to energy consumption. In general, lower buffer sizes which reduce the energy consumption were found to improve delay, but sacrifice throughput by 5-8% at high load. Based on a broad set of results, this study provides guidelines for sizing router buffers in realistic networks which would lead to optimal trade-offs between energy consumption and quality of service.

A Backhaul Adaptation Scheme for IAB Networks Using Deep Reinforcement Learning With Recursive Discrete Choice Model

“…In this way, a buffer cost is defined to reward the system for minimizing queuing delays, thereby protecting against overflows and subsequent packet losses, as well as penalizing every packet that is lost. The buffer cost is defined as the expected sum of holding costs and overflow cost with respect to the traffic arrival and goodput distributions [26], and can be expressed as follows:…”