All Times ET
Virtual
WITHDRAWN: Detection and Mitigation of Anomalous Traffic Spikes in Communication Networks (304210)
*Mrinmoy Bhattacharjee, NokiaAditya Gudal, Nokia
Veena B. Mendiratta, Nokia Bell Labs
Keywords: anomaly detection, concept drift, time series forecasting, machine learning, analytics
In this talk we demonstrate how time series control data from network servers in the field are used to address anomaly detection and mitigation in large communication networks. The 4G core network plays a central role in connecting millions of user devices (mobile phones) to the internet and handling user mobility. The requirement from service providers is to find the user device within a few seconds. A Mobility Manager contacts an eNodeB to find the user device, successively increasing the number of eNodeBs contacted to maximize the chances of successfully finding the user device. Due to anomalies occurring in a network, a Mobility Manager may have to expand the search area for a large number of user devices thereby causing network congestion which can lead to outages. Using streaming control data from Mobility Management servers, we will present the analysis and algorithms used to detect such anomalies with the goal of: creating an alert when an anomaly occurs; and triggering the Mobility Manager to take automatic corrective measures such as reducing the search area to Last Seen eNodeB. The models we designed using 4G network data are equally applicable to emerging 5G networks.