Consensus algorithms in distributed SDN controllers

Programsko določena omrežja z logično centralizacijo nadzora in abstrakcijo omrežnih virov obljubljajo večjo prilagodljivost, cenovno učinkovitost in lažje upravljanje omrežne infrastrukture, vendar njihovo širšo uporabo ovirajo različni izzivi. V magistrskem delu raziščemo problem zagotavljanje visoke razpoložljivosti krmilnikov programsko določenih omrežij. Gre za slabo raziskan problem, ki je hkrati ključnega pomena pri uvajanju programsko določenih omrežij v produkcijska okolja. Najprej predstavimo koncept programsko določenih omrežij. Pri tem raziščemo, kje ima njihova uporaba največji potencial in identificiramo različne izzive, ki otežujejo njihovo uvedbo. Izpostavimo problem zagotavljanja visoke razpoložljivosti in skalabilnosti omrežja zaradi centralizacije krmilne ravnine ter kot rešitev navedemo implementacijo krmilnika v obliki porazdeljenega sistema odpornega na napake. V nadaljevanju preučimo omejitve pri načrtovanju in implementaciji teh sistemov. Osredotočimo se na algoritme soglasja kot ključno komponento za zagotavljanje visoke razpoložljivosti. Nato predstavimo strategije pri zasnovi porazdeljenih krmilnikov in raziščemo, katere odprtokodne implementacije omogočajo njihovo visoko razpoložljivost. Krmilnik ONOS izberemo kot potencialno najbolj primeren za produkcijsko uporabo in analiziramo njegovo arhitekturo. Z izbranim krmilnikom in simulatorjem programsko določenih omrežij Mininet vzpostavimo pilotno okolje. Razvijemo testno ogrodje, s katerim simuliramo scenarije, ki vključujejo različne odpovedi vozlišč krmilnika in komunikacijskih kanalov, ter pri tem analiziramo obnašanje sistema. Na podlagi rezultatov analize ovrednotimo izbrano implementacijo krmilnika z vidika zagotavljanja visoke razpoložljivosti in podamo predloge za izboljšanje razpoložljivosti rešitve.Software defined networks (SDN) promise greater flexibility, cost efficiency and easier management of network infrastructure by logically centralizing the control and abstracting network resources, but their wider use is inhibited by various challenges. In this master’s thesis, we explore the problem of ensuring high-availability of SDN controllers. It is a mostly unexplored problem which is also crucial with the implementation of software defined networks in production environments. Firstly, we present the concept of software defined networking. Doing so, we explore where their use has the greatest potential and identify various challenges which make their implementation difficult. We emphasize the problem of ensuring high-availability and the scalability of the network because of the centralized control plane and list as a solution the implementation of a controller in the form of a fault-tolerant distributed system. Following this, we study the limitations in design and the implementation of these systems. We focus on consensus algorithms as a key component in ensuring high availability. Following this, we present strategies with developing distributed controllers and explore which open-source implementations allow for their high availability. We choose the ONOS controller as the potentially most suitable for production use and analyze its architecture. With the chosen controller and the simulator of software defined networks Mininet, we establish a pilot environment. We develop a test framework for simulating scenarios that include various controller node and communication channel failures and analyze system behavior while doing so. Based on the results of the analysis we evaluate the chosen controller implementation based on ensuring high availability and give suggestions for improving the availability of the solution