Loading...

Use and Evaluation of Predictive and Speculative Techniques in Software-Defined Networks (SDN) processor

Dorosti, Zahra | 2017

444 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 50594 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Jahangir, Amir Hossein
  7. Abstract:
  8. Software Defined Networking (SDN) is an emerging paradigm which makes the network programmable by separating the control plane from data plane and makes both planes to work independently. There is a centralized controller and a programmable data plane in these networks architecture and forwarding data packets is realized by programming the data plane via an open interface called OpenFlow. OpenFlow is a communication protocol between control and data planes. The centralized architecture of these networks provides a global view of the underlying network to upper applications and brings numerous advantages such as routing, traffic engineering and QoS control. Despite these advantages there are some concerns about SDN scalability. A central controller does not scale up when the network grows and in some cases the controller fails to handle all incoming requests.This thesis shows that two main problems of SDN are performance and scalability. The key solution for this problems is to add some prediction ability to switches and also to reduce the load on the controller. Some previous works used distributed mechanisms to divide the load on multiple controllers, but they have limitation on number of events due to consistency overheads. Some others introduce new functionalities in switches to suppress frequent events and keep the traffic in the data plane, but these methods are complex. In this study the performance and scalability by different ways are enhaced. We try to decrease end to end delay and the controller load via predicting the necessary rules in the network and forwarding packets before controller response arrives. By using predictive and speculative techniques for predicting rules, end to end delay, the number of OpenFlow switch messages to the controller and flow table size decreased on average 44, 35 and 52 percent respectively
  9. Keywords:
  10. Prediction ; Network Processors ; Performance ; Software Defined Networks (SDN) ; OpenFlow Protocol

 Digital Object List

 Bookmark

No TOC