On Improving DREAM Framework with Estimations and ProgME

Title: On Improving DREAM Framework with Estimations and ProgME
Authors: Hernandez Remedios, Rene
Date: 2017
Abstract: Software Defined Networking (SDN) is an emerging architecture that is dynamic, manageable, cost-effective and adaptable, making it ideal for the high-bandwidth, dynamic nature of today’s applications. Using SDN, networks can enable a variety of concurrent, dynamically instantiated measurement tasks, that provide fine-grain visibility into network traffic by configuring Ternary Content Address Memory (TCAM) counters in hardware switches. However, TCAM memory is limited, thus the accuracy of measurement tasks depends on the number of resources devoted to them on each switch. In this thesis, we propose a solution that improves Dynamic Resource Allocation for Software-defined Measurements (DREAM), a framework with an adaptive step size search that achieves a desired level of accuracy for measurement tasks. We have enabled prediction capabilities in the framework to generate better counters configurations using previous network traffic information. We implement four estimation techniques (EWMA-based Prediction, Polynomial Curve Fitting, KMeans++ Cluster and Pseudo Linear Extrapolation) that have been tested with simulations running three types of measurement tasks (heavy hitters, hierarchical heavy hitters and traffic change detection) that show the proposed techniques improve task accuracy and tasks concurrency in DREAM. Existing traffic measurements tools usually rely on some predetermined concept of flows to collect traffic statistics. Thus, they usually have issues in adapting to changes in traffic condition and present scalability issues with respect to the number of flows and the heterogeneity of the monitoring applications. We propose an integration of the Programmable MEasurements (ProgME) paradigm, which defines a novel approach to defined measurement tasks in a programmable way using the concept of flowsets, on top of the DREAM framework. This enables better scalability for measurement tasks that deal with large amounts of traffic flows on DREAM while reducing the required number of counters allocations for the tasks.
URL: http://hdl.handle.net/10393/36013
CollectionThèses, 2011 - // Theses, 2011 -