Prioritized Routing in SDN

Project Title

Prioritized Routing in SDN

Project Area of Specialization Computer ScienceProject Summary

 Software Defined Network (SDN) provides flexibility, agility, programmability and a centralized control to the network. A network admin can mold the network according to the system and application requirements. Deep learning provides the system ability to learn and improve with experience without any explicit instructions, that is it provides intelligence to computer which comes naturally to human being, such as image, sound and audio recognition This is done by training Neural Networks using large labeled data-sets). This research project will focus on combining these domains to design an environment in which videos will be classified and given priority through Deep learning methods. The classified videos will be converted into packets with a flag of respective priorities and the packets will be sent towards the destination through socket programming. The packets will then get to the openflow switch initially there would not be a flow-match entry in the switch, the packets will be sent to controller via PACKET-IN message, the controller will then read the header to inspect the priority defined and define the next hop on the switch. The Fat Tree Topology with around 100 hosts and 25 switches (virtual) will be used to test the environment and in order to make the network non linear, the bandwidth will vary. Two types of connections will be used; Ethernet (provides traditional LAN connection) and Infiniband (provides high speed connection with Low latency). To check performance of the prototype, it will be compared with the default SDN network based on Dijkstra algorithm.

Project Objectives

To prioritize critical surveillance videos over other video traffic, by classifying them using deep learning and defining flows which would be best for their specifications.

• Building Deep Learning Models:
• Collecting video datasets for three different classes.
• Training Models.
• Testing Models. 
• Developing the SDN network:
• Building a Fat-tree topology on every system.
• Testing the connection.3.
• Experimentation:
• Optimizing Video Classifier.
• Bench-marking the prototype against default Floodlight Controller.

Project Implementation Method

The project is Quantative. Precisely, it is Applied Research, and Deductive in nature.The primary method of research is through Literature Reviews as well as Conceptual Modelling. The team will attempt to solve the task by creating prototypes and verifying their accuracy. All findings will be logged and the prototype will be observed during testing.
Several Research papers have pertained to the study of SDNs, but the implementation of ML to provide priority of packets was seldom seen. Literature Review of such documents will assist in the implementation of the project.
 Classifiers will be trained in this research to detect relevant video types. These classifiers will be modified and improved until it is reliable enough to be used to mark video priority, in an attempt to reduce transfer time for higher priority videos.

Benefits of the Project

Internet is a Network of Networks. Here, each application running on various hosts have different requirements, such as high bandwidth and low latency. Less important applications might congest the network which affects the quality of service of Surveillance systems and other critical application. In a traditional network, it becomes difficult to cater these necessities, as there is no centralized control over the network and the switches operate in an independent way. Although, SDNs provides a centrailized control and make it easier to manage network resources, but it does not implicitly deal with priority of packets based on Applications.

Technical Details of Final Deliverable

This end result of the research will be three video classifiers that will be able to classify videos according to their importance. A modified floodlight controller that will be able to forward the video packets according to their priority as classified by the Neural Network. 

• Nvidia GTX 1080 or similar GPU will be used for training the classifier to detect Fire in a video.
• Mininet, a network virtualization tool will be used to simulate the SDN testbed.
• Infiniband will be used as a high speed link between two systems.

Final Deliverable of the Project HW/SW integrated systemCore Industry ITOther IndustriesCore Technology Big DataOther TechnologiesSustainable Development Goals Partnerships to achieve the GoalRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
Nvidia GTX 1080	Equipment	1	70000	70000