Flow-based Dynamic Queue Selector: Guaranteeing service level agreements in a congested IP-network
Authors
Vejlø, Mads Holdgaard ; Andersen, Niels Fristrup
Term
4. term
Education
Publication year
2014
Submitted on
2014-06-02
Pages
78
Abstract
Denne afhandling præsenterer en løsning til at øge udnyttelsen af backbone-netværksforbindelser, samtidig med at serviceaftaler (SLAs) for forskellige tjenestetyper overholdes, ved hjælp af Software-Defined Networking (SDN). Arbejdet bygger på et emuleret netværk, hvor letvægtsvirtualisering af enkelte netværtsværter og virtuelle switche bruges til at efterligne et rigtigt miljø. Trafikken genereres syntetisk for tre tjenestetyper: video, tale og best effort. For hver tjeneste er der SLA-parametre for gennemløb (hvor meget data der leveres), ende-til-ende-forsinkelse (hvor lang tid data er undervejs) og pakketabsrate (hvor meget der går tabt). Løsningen er en kontrolsløjfe med metoder fra kognitive netværk, som overvåger strømme (flows) og tilpasser håndteringen af dem. Den flytter strømme mellem forskellige køer på en switchs udgangsport og kan om nødvendigt droppe strømme for at sikre, at hver enkelt strøm overholder sine SLA-krav. Resultaterne viser, at løsningen kan garantere SLAs per strøm ved linkudnyttelser op til 80 % for de strømme med de mest krævende SLAs.
This thesis presents a way to increase backbone link utilization while meeting Service Level Agreements (SLAs) for different service types, using Software-Defined Networking (SDN). The study uses an emulated network built with lightweight virtualization of individual hosts and virtual switches to mimic a real environment. Synthetic traffic is generated for three service types: video, voice, and best effort. Each has SLA parameters for throughput (how much data is delivered), end-to-end delay (how long data takes to travel), and packet loss rate (how much is lost). The proposed solution is a control loop based on methods from cognitive networks that monitors flows and adapts how they are handled. It moves flows between different queues on a switch's output port and, when necessary, drops flows to keep each flow within its SLA. Results show that the approach can guarantee per-flow SLAs at link utilizations up to 80% for flows with the most demanding SLAs.
[This abstract was generated with the help of AI]
Keywords
SDN ; Controller ; Network ; QoS ; SLA ; IP ; Link Capacity ; Throughput ; OpenFLow ; Performance ; Backbone ; Utilisation ; Smart Droppings ; Software-Defined Networking ; Mininet ; Open vSwitch
Documents