Login
The UniverSelf project is proud to announce the final release of the technical report on use case requirements:
Deliverable 3.7 - Adaptation of learning and operation methods to specific needs of future networks and services
The report can be accessed under the Dissemination/Technical Reports section, or directly at the following link: http://www.univerself-project.eu/technical-reports
Summary:
This deliverable builds upon the work in task T3.3 previously presented in the deliverable D3.2, and focuses on the various techniques within the realm of observation and action. In addition to presenting the progress made in the techniques reported in the first year, the deliverable places the techniques together in the context of integration within the Unified Management Framework (UMF), the framework and integration structure developed in WP2 for unifying management functions that targets the embedding of autonomic paradigms in networks. As the work that has been done in T3.3 relates to quite diverse application contexts and is as such unsuitable to be combined into a single overall storyline, this deliverable is structured in terms of use cases as defined in WP4.
Thus, the chapters 2 to 7 each pertain to a specific WP4 use case. Each chapter lays out the details of the work done on the different techniques (also called Network Empowerment Mechanisms or NEMs, for short) related to the given use case. The context, content and merits of each of the NEMs are given. Emphasis was put on substantial evaluation results and the progress beyond previously reported work. At the end of each chapter, a use case wide discussion is presented that relates the NEMs of a use case in terms of potential interworking, and explains the usefulness of UMF for such a collaborative environment.
The presented techniques cover a significant spectrum of observation related domains (discovery, aggregation, diagnosis, mining, and (self-) modelling) and methodologies (fuzzy logic, self-organizing maps, reinforcement learning, case-based reasoning, Bayesian networks and statistical analysis). In Use Case 1 (“Self-diagnosis and self-healing for IMS VoIP and VPN services”), methods for root cause analysis (with much improved accuracy), anomaly detection (with high effectiveness while avoiding manual parameter tuning), congestion prediction (highly expressive and proactive instead of reactive), flow self-diagnosis of QoS (with high success rate) and context acquisition (with significantly compressed yet representative data representation) are described. Use Case 2 (“Networks' Stability and Performance”) comprises methods for performance improvement of TCP Vegas (by classifying the cause of round trip time changes), orchestration of control loops (by automatically analyzing the configuration parameter space of the individual sub-loops) and vulnerability management (allowing autonomic agents to assess their security exposure in an automated way). Use Case 3 (“Dynamic Virtualization and Migration of Contents and Servers”) covers items such as context discovery improvements (by means of rule-based reasoning on raw data) and more efficient video delivery (reducing buffer starvation while minimizing capacity needs). In Use Case 4 (“SON and SON collaboration according to operator policies”), capacity enhancements of in-band relay links (via learning-based self-adaptation) and coverage optimization in dense base station deployments (via an intelligent reward distribution scheme) are covered. Use Case 6 (“Operator-governed, end-to-end, autonomic, joint network and service management”) addresses two load estimation mechanisms for RAN cells (one based on Self-Organizing Maps with very good accuracy, one based on statistical analysis) and an admission control scheme based on Explicit Congestion Notification (where the latter’s parameters are automatically adjusted without knowledge about the underlying traffic aggregate). Finally, Use Case 7 (“Network and Service Governance”) presents a new work on FTTH related fault analysis with very high accuracy.
Their effectiveness in operation was demonstrated mostly through simulation or mathematical analysis. In addition, we also consider the effectiveness of the techniques when used together in the same network, which reinforces the need for coordination and exchange of information between these techniques, and identify the location and nature of the integration that is required through the UMF in order for the techniques to draw a maximum benefit out of the collaboration.
One of the objectives of this deliverable was to identify what method is best-suited for a given observation and action related problem. This particular objective has been addressed in milestone MS31, which provides a systematic methodology to assess this question by means of a questionnaire, and the results will be conclusively presented in the handbook deliverable D3.9.
