A research report submitted to the Faculty of Engineering and the Built
Environment of the University of the Witwatersrand in partial fulfilment
of the requirements for the degree of
Master of Science in Engineering
February 2017 / Software-defined networking (SDN) is the physical separation of the control and data
planes in networking devices. A logically centralised controller plane which uses a
network-wide view data structure to control several data plane devices is another
defining attribute of SDN. The centralised controllers and the network-wide view data
structure are difficult to scale as the network and the data it carries grow. Solutions
which have been proposed to combat this challenge in SDN lack the use of the statistical
properties of the workload or network traffic seen by SDN controllers. Hence, the
objective of this research is twofold: Firstly, the statistical properties of the controller
workload are investigated. Secondly, Autoregressive Integrated Moving Average Models
(ARIMA) and Artificial Neural Network (ANN) models are investigated to establish
the feasibility of forecasting the controller workload signal. Representations of the state
of the controller plane in the network-wide view in the form of forecasts of the controller
workload will enable control applications to detect dwindling controller resources and
therefore alleviate controller congestion. On the other hand, realistic statistical traffic
models of the controller workload variable are sought for the design and evaluation of
SDN controllers. A data center network prototype is created by making use of an SDN
network emulator called Mininet and an SDN controller called Onos. It was found that
1–2% of flows arrive within 10 s of each other and more than 80% have inter-arrival
times in the range of 10 s–10ms. These inter-arrival times were found to follow a
beta distribution, which is similar to findings made in Machine Type Communications
(MTC). The use of ARIMA and ANN to forecast the controller workload established
that it is feasible to forecast the workload seen by SDN controllers. The accuracy of
these models was found to be comparable for continuously valued time series signals.
The ANN model was found to be applicable even in discretely valued time series data. / MT2017
| Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:wits/oai:wiredspace.wits.ac.za:10539/22955 |
| Date | January 2017 |
| Creators | Nkosi, Emmanuel |
| Source Sets | South African National ETD Portal |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | Online resource (xvii, 142 leaves), application/pdf, application/pdf |
Page generated in 0.0023 seconds