ARIMA for Traffic Load Prediction in Software Defined Networks

2016

A multi-step neural network based predictive model is developed and trained offline, deployed in the cloud for real-time scoring, and used to predict network traffic. The information is then used to avoid congestion from occurring in a distributed real-time environment. The model predicts network traffic with high accuracy under varying scenarios representing low, medium and high load conditions. The accuracy is consistently maintained both inside and outside training ranges of the selected parameters which impact congestion. The R-square value ranges between 89.8% and 95.7% across all loads. The results prove robustness of the proposed model and illustrate that it can capture fluctuations, handle spikes, and adapt to varying load conditions. The results are obtained in real-time and used for taking proactive measures.

2020

Predicting the bandwidth utilization on network links can be extremely useful for detecting congestion in order to correct them before they occur. In this paper, we present a solution to predict the bandwidth utilization between different network links with a very high accuracy. A simulated network is created to collect data related to the performance of the network links on every interface. These data are processed and expanded with feature engineering in order to create a training set. We evaluate and compare three types of machine learning algorithms, namely ARIMA (AutoRegressive Integrated Moving Average), MLP (Multi Layer Perceptron) and LSTM (Long Short-Term Memory), in order to predict the future bandwidth consumption. The LSTM outperforms ARIMA and MLP with very accurate predictions, rarely exceeding a 3% error (40% for ARIMA and 20% for the MLP). We then show that the proposed solution can be used in real time with a reaction managed by a Software-Defined Networking (SDN) platform.

In computer networks, diverse applications generate network traffic with different characteristics. Network traffic classification is significant to manage networks better, improve service quality and ensure security. Software-Defined Networks (SDN) provides flexible and adaptable techniques for traffic classification with its programmable structure. SDN flows naturally exhibit particular characteristics of network applications and protocols. Therefore, it can be said that SDN can present significant opportunities in traffic classification using machine learning. This study proposes a traffic classification approach using machine learning models in SDN. In this study, DNS, Telnet, Ping and Voice traffic flows were created on the SDN using the Distributed Internet Traffic Generator (D-ITG) tool. Twelve-features representing these traffic flows (the number of packets transmitted, average transmission time, the number of instantly transmitted packets, etc.) were determined, and over the SDN controller in the physical network, a real-time dataset was created by collecting data depending on the features. Later, the performance of k Nearest Neighbor (k-NN), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Decision Tree (DT) and Naive Bayes (NB) machine learning models were tested for traffic classification on this dataset. When the k-NN model was tested on this real-time dataset, its classification accuracy was obtained as the maximum with 99.4%. Therefore, this model has been determined as a machine learning giving the highest classification performance with the lowest cost flow features in traffic classification in SDN.

Human-Centric Intelligent Systems

These days, Internet coverage and technologies are growing rapidly, hence, it makes the network more complex and heterogeneous. Software defined network (SDN) revolutionized the network architecture and simplified the network by separating the control and data plane. On the other hand, machine learning (ML) and its derivations have made the systems more intelligent. Many pieces of research papers have addressed ML and SDN. In this survey, we collected the papers published in Springer, Elsevier, IEEE, and ACM and addressed SDN and ML between 2016 and 2023. The research papers are organized based on the solutions, evaluation parameters, and evaluation environments to help those working on SDN and ML for improving the target functional or non-functional parameters. The research papers will be analyzed to extract the solutions, evaluation parameters and environments. The extracted solutions, evaluation parameters and environments will be clustered in this review paper. The research gap ...

Applied Sciences

Rapid development in the field of computer networking is now demanding the application of Machine Learning (ML) techniques in the traditional settings to improve the efficiency and bring automation to these networks. The application of ML to existing networks brings a lot of challenges and use-cases. In this context, we investigate different ML techniques to estimate resource requirements of complex network entities such as Virtual Network Functions (VNFs) deployed in Software Defined Networks (SDN) environment. In particular, we focus on the resource requirements of the VNFs in terms of Central Processing Unit (CPU) consumption, when input traffic represented by features is processed by them. We propose supervised ML models, Multiple Linear Regression (MLR) and Support Vector Regression (SVR), which are compared and analyzed against state of the art and use Fitting Neural Networks (FNN), to answer the resource requirement problem for VNF. Our experiments demonstrated that the behav...

2021 IEEE 22nd International Conference on High Performance Switching and Routing (HPSR), 2021

Over the last few years, networks' infrastructures are experiencing a profound change initiated by Software Defined Networking (SDN) and Network Function Virtualization (NFV). In such networks, avoiding the risk of service degradation increasingly involves predicting the evolution of metrics impacting the Quality of Service (QoS), in order to implement appropriate preventive actions. Recurrent neural networks, in particular Long Short Term Memory (LSTM) networks, already demonstrated their efficiency in predicting time series, in particular in networking, thanks to their ability to memorize long sequences of data. In this paper, we propose an improvement that increases their accuracy by combining them with filters, especially the Fast Fourier Transform (FFT), in order to better extract the characteristics of the time series to be predicted. The proposed approach allows improving prediction performance significantly, while presenting an extremely low computational complexity at run-time compared to classical techniques such as Auto-Regressive Integrated Moving Average (ARIMA), which requires costly online operations.

International Journal of Communication Systems, 2005

Fast and accurate methods for predicting traffic properties and trend are essential for dynamic network resource management and congestion control. With the aim of performing online and feasible prediction of network traffic, this paper proposes a novel time series model, named adaptive autoregressive (AAR). This model is built upon an adaptive memory-shortening technique and an adaptive-order selection method originally developed by this study. Compared to the conventional one-step ahead prediction using traditional Box-Jenkins time series models (e.g. AR, MA, ARMA, ARIMA and ARFIMA), performance results obtained from actual Internet traffic traces have demonstrated that the proposed AAR model is able to support online prediction of dynamic network traffic with reasonable accuracy and relatively low computation complexity.

Journal of Networks, 2009

Short period prediction is a relevant task for many network applications. Tuning the parameters of the prediction model is very crucial to achieve accurate prediction. This work focuses on the design, the empirical evaluation and the analysis of the behavior of training-based models for predicting the throughput of a single link i.e. the incoming input rate in Megabit per second. In this work, a neurofuzzy model (α SNF), the AutoRegressive Moving Average (ARMA) model and the Integrated AutoRegressive Moving Average (ARIMA) model are used for predicting. Via experimentation on real network traffic of different links, we study the effect of some parameters on the prediction performance in terms of error. These parameters are the amount of data needed to identify the model (i.e. training set), the number of last observations of the throughput (i.e. lag) needed as inputs for the model, the data granularity, variance and packet size distribution. We also investigate the use of the number of packets or sampled data as inputs for the prediction model. Experimental results show that training-based models, identified with small training set and using only one lag, can provide accurate prediction. We show that counts of packets and especially large packets can be used to efficiently predict the throughput.

International Journal of Advanced Computer Science and Applications

Nowadays Internet does not provide an exchange of information between applications and networks, which may results in poor application performance. Concepts such as application-aware networking or network-aware application programming try to overcome these limitations. The introduction of Software-Defined Networking (SDN) opens a path towards the realization of an enhanced interaction between networks and applications. SDN is an innovative and programmable networking architecture, representing the direction of the future network evolution. Accurate traffic classification over SDN is of fundamental importance to numerous other network activities, from security monitoring to accounting, and from Quality of Service (QoS) to providing operators with useful forecasts for long-term provisioning. In this paper, four variants of Neural Network estimator are used to categorize traffic by application. The proposed method is evaluated in the four scenarios: feedforward; Multilayer Perceptron (MLP); NARX (Levenberg-Marquardt) and NARX (Naïve Bayes). These scenarios respectively provide accuracy of 95.6%, 97%, 97% and 97.6%.

2008

Many network resource management solutions typically employ traffic prediction algorithms to improve the performance of a network. In this paper we extend a newly developed method for prediction with confidence to time series data and apply it to the network traffic demand prediction problem. We investigate the performance of the proposed algorithm on a number of publicly available network traffic demand datasets. The experimental results are very promising.