Scalable classification of QoS for real-time interactive applications from IP traffic measurements

Abstract

Measurement of network Quality of Service (QoS) has attracted considerable research effort over the last two decades. The recent trend towards Internet Service Providers (ISP's) offering application-specific QoS is creating possibilities for more sophisticated QoS metrics to be offered by ISP's in service level agreements. This in turn could be used for the purposes of improved network optimization and billing according to application specific QoS guarantees. We report a scalable near real-time approach using passively logging IP traffic data for classification of application latency and packet loss across a range of real-time interactive applications. We run six experiments involving Minecraft, Quake 3 Urban Terror, VLC video streaming and the commercial Wirofon VOIP application. We use a mixture of laboratory and realworld deployments, with run times ranging from hours to days, and observe a combination of real and simulated ISP latency and packet loss events. Our binary classification (i.e. classes 'OK' or 'lag') 10-fold cross validation F1 scores are between 0.80 and 0.93 depending on application type. Our multi-class classification (i.e. classes representing discrete packet loss or latency ranges) 10-fold cross validation F1 scores for Minecraft are 0.89 for latency and 0.90 for packet loss. With new business models between ISP's and application developers being actively considered this work represents a significant contribution to the debate by providing scientific evidence relating to a novel approach to scalable QoS measurement

Towards the Deployment of Machine Learning Solutions in Network Traffic Classification: A Systematic Survey

“…This characteristic makes them suitable for feature Some of them are computed under the assumption that the properties values are normally distributed, which might not be true for some cases. [110], [95], [96], [97], [111], [112], [113], [114], [115], [116], [117], [118], [119], [120], [121], [122], [123], [124], [125], [126], [127], [128], [129], [96], [130], [131], [132], [133], [134], [106], [135], [136], [137], [138], [105], [139], [140], [107], [141], [142], [143] Graph based features Internet interactions are modeled as graphs and valuable features can be extracted from these representations They are ideal for understanding communication patterns…”

“…The work in [123] groups flows into application categories (video streaming, VoIP, etc) to improve the QoS. Several experiments were carried on varying the condition in which the data is collected (e.g., packet loss and high or low latency).…”

“…Therefore, self-learning, evolving or retraining strategies must be taken into account. In general, [123], [95], [149], [70], [116], [157], Decision trees [171], [54], Support vector machines (SVMs) [126], [104], [96], Expectation Maximization [141], Laplacian SVM [127], Bayesian Networks (BNs), Decision Trees (DTs) and Multilayer Perceptrons (MLPs) [117],K Nearest Neighbour (KNN) [137], [119], Naive Bayes [138], [105], [142], [130], [172]. Hidden Markov Model(HMM) [144], [173], [93], Gaussian model [147], [158] Multiclassification and ensemble approach It aims at creating a combination of several classifiers…”

Analysis of promising approaches and research on traffic flow classification for maintaining QoS by ML methods in SDN networks

“…Основные сферы применения методов ML в SDN-сетях: классификация потоков трафика [26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43]; динамическое определение оптимального маршрута с учетом состояния соединений и сетевых элементов [25,44]; прогнозирование параметров QoS/ QoE (Quality of Experience) на основе имеющейся статистической информации [45][46][47][48]; управление ресурсами [44,[49][50][51]; управление политиками безопасности, в т.ч. обнаружение вторжений [52][53]; проведение статистических исследований, предсказание занятости каких-либо каналов связи или использование определенных приложений [40,44].…”

Time-Aware Learning Framework for Over-The-Top Consumer Classification Based on Machine- and Deep-Learning Capabilities

“…In turn, multiple studies have been conducted to determine how network providers can leverage machine learning to analyze user traffic. Middleton and Modafferi [24] exploited machine learning to classify IP traffic in support of QoS guarantees. Yang et al [25] proposed the classification of Chinese mobile internet users into heavy and high-mobility users by analyzing the network traffic of 2G and 3G services.…”