A time‐efficient shrinkage algorithm for the Fourier‐based prediction enabling proactive optimisation in software‐defined networks

Abstract

SummaryThis paper focuses on the problem of time‐efficient traffic prediction. The prediction enables the proactive and globally scoped optimisation in software‐defined networks (SDNs). We propose the shrinkage and selection heuristic method for the trigonometric Fourier‐based traffic models in SDNs. The proposed solution allows us to optimise the network for an upcoming time window by installing flow entries in SDN nodes before the first packet of a new flow arrives. As the mechanism is designed to be a part of a sophisticated routing‐support system, several critical constraints are considered and taken into account. Specifically, the system is traffic‐ and topology‐agnostic, thus the prediction mechanism must be applicable to the networks with highly variable traffic loads (e.g., observed inside intra‐DCNs: datacentre networks). Furthermore, the system must effectively optimise routing in large‐scale SDNs comprised of numerous nodes and handling millions of flows of a dynamic nature. Therefore, the prediction must be simultaneously accurate as well as being time efficient and scalable. These requirements are met by our Fourier‐based solution, which subtracts consecutive harmonics from the original signal and compares the result with an adaptive threshold adjusted to the signal's standard deviation. The evaluation is performed by comparing the proposed heuristic with the well‐known Lasso method of proven accuracy. The results show that our solution is able to retain prediction accuracy at a comparable level. Moreover, in accordance with our main aim, we operate in a manner which is always significantly faster. In some cases, computation times are reduced by as much as 50 times.

Fine-grained Network Traffic Prediction from Coarse Data

“…Typically the seasonal effect is not harmonic and it is approximated by a sum of multiple Fourier components for both weekly and daily seasonal effects (see Rzym et al (2020) for details on how to choose the number of components). Finally, further temporal dependences are captured via a one-dimensional autoregressive (AR) process…”

Long-Term Traffic Forecasting in Optical Networks Using Machine Learning

“…The network long-term traffic forecasting problem is not new in the literature and has been widely studied in many papers. Typically, it is formulated as a TS problem and solved using approaches based on ARIMA and its numerous variations, as well as ML techniques [15]. Authors of [16] compared ARIMA, Holt-Winters, and neural network algorithms for forecasting the amount of traffic in TCP/IP-based networks.…”

Machine Learning Classification and Regression Approaches for Optical Network Traffic Prediction

“…Typically, the problem of forecasting network traffic is formulated as a time series problem. A large majority of works in the field use approaches based on an autoregressive integrated moving average (ARIMA) and its numerous variations, as well as ML techniques [17], to solve this. Authors in [18] compared ARIMA, Holt-Winters, and neural network algorithms for forecasting the amount of traffic in TCP/IP-based networks.…”