“A Consistent and Fault-Tolerant Data Store for Software Defined Networks”

From Navigators

Jump to: navigation, search

Fabio Botelho (advised by Alysson Bessani, Fernando Ramos)

Master’s thesis, Mestrado em Segurança Informática, Departamento de Informática, Faculdade de Ciências da Universidade de Lisboa, Feb. 2014

Abstract: Even if traditional data networks are very successful, they exhibit considerable complexity manifested in the configuration of network devices, and development of network protocols. Researchers argue that this complexity derives from the fact that network devices are responsible for both processing control functions such as distributed routing protocols and forwarding packets. This work is motivated by the emergent network architecture of Software Defined Networks where the control functionality is removed from the network devices and delegated to a server (usually called controller) that is responsible for dynamically configuring the network devices present in the infrastructure. The controller has the advantage of logically centralizing the network state in contrast to the previous model where state was distributed across the network devices. Despite of this logical centralization, the control plane (where the controller operates) must be distributed in order to avoid being a single point of failure. However, this distribution introduces several challenges due to the heterogeneous, asynchronous, and faulty environment where the controller operates. Current distributed controllers lack transparency due to the eventual consistency properties employed in the distribution of the controller. This results in a complex programming model for the development of network control applications. This work proposes a fault-tolerant distributed controller with strong consistency properties that allows a transparent distribution of the control plane. The drawback of this approach is the increase in overhead and delay, which limits responsiveness and scalability. However, despite being fault-tolerant and strongly consistent, we show that this controller is able to provide performance results (in some cases) superior to those available in the literature.

Export citation



Research line(s): Fault and Intrusion Tolerance in Open Distributed Systems (FIT)

Personal tools
Navigators toolbox