Towards real-time anomalies monitoring for QoE indicators

Abstract

Anomaly detection and characterization is a main topic for network managers. Although qualityof-service (QoS) indicators can help to infer problem occurrence, they do not provide immediate insight on the user's perceived quality. Evolved service-level agreements (SLA) will likely be established in terms of quality of experience (QoE) indicators. QoS metrics composing the QoE indicators need to be monitored on a real-time basis by the SLA management tools in order to detect anomalies driving to contract violations. Monitoring SLA contracts may involve the surveillance of individual application sessions for several users. In this work, we address the problem of anomaly detection with impact on a relatively large number of users, either on one or on several types of applications simultaneously. We propose a method to characterize the state of the network, representing QoE indicators as time series and reducing the dimension of the data set. The singular spectrum analysis (SSA) method, using a combination of geometric and statistical methods, is proposed as an analysis tool in order to detect anomalies on QoE indicator evolution.

The effects of Packet Delay Variation on the perceptual quality of video

“…The forget factor, the recency effect [15] and sudden change of PDV during voting are possible reasons. Including the remaining effects of the previously obtained outputs into the calculation of the current output is possible by the EWMA approach [7]. When obtaining the UR, the EWMA is used for computing the correlation of instantaneous user perception against the QoS metrics.…”

On the testing of network cyber threat detection methods on spam example

“…As the maximal nProbe 9 throughput is not significantly limited by the spam plugin, a rough performance estimate may be based on data provided with this software and used to answer such a question. According to the performance data published by ntop 10 when run on dual core hardware, the probe is easily capable of handling traffic above 1, 000 [Kpps] without dropping packets.…”

“…Finally, an end-to-end testing strategy was proposed and performed. This took into account the feasibility of the entire threat detection process 9 In the tests, nProbe version 6.9.4 was used 10 http://www.ntop.org/nprobe/tuning-nprobe-6-4-scalability-and-per formance/ including traffic capturing, calculation of flow features and analysis by the proposed threat detection algorithm.…”

“…In order to be able to detect events in such a context, the data set to be analysed should be reduced [9]. Therefore, only the most important features should be selected to describe the collected records.…”

Studying the challenges in assessing the perceived quality of mobile-phone based video

“…The study of autocorrelation of perceived instantaneous quality shows that the perceived instantaneous quality is influenced by the latest received quality [20]. We investigated this through the presentation with visual stimuli, i.e., repeated video sequences, to the participants and studied the fluctuation of MOS values on the same video of a particular user and the memory effect to understand the psychological influence factor of previous video sequences.…”