Machine learning based control and security of software-defined networks

About this project

Project description

Software-defined networking (SDN) is a new paradigm that allows the use of programmable network devices. A centralised SDN controller has a global view of the network, and thus can have a better control. This facilitates better utilisation of network resources, separates the control and data planes, and eliminates the need for specialised network devices such as routers and switches. In this project, we will solve some key research problems in SDN.

The first problem is designing optimal network control rules for the SDN controller. The control rules will take the network state as input and produce various network control decisions such as flow admission, routing, transmission scheduling. This work will involve both model based (Markov decision process formulation) and data based algorithm development (via deep reinforcement learning).

The second problem will be an extension of the first problem in a multi-controller setting. This is important because the single controller setting is not scalable for the IoT and edge computing setting. This work will involve tools from federated learning.

The third problem will be on security of SDN. In a denial of service attack, a malicious node generates fake traffic to force the SDN controller to deprive a legitimate node of resources. In a multi-controller setting, if one SDN controller is compromised due to hacking, others can detect that by statistical analysis of network traffic. This work will be primarily based on anomaly detection using machine learning tools, and queueing and scheduling in presence of such attacks.

All these problems will be solve in the IoT setting, keeping in mind the key challenges such as large number of nodes, network dynamism, intermittent connectivity, ultra-reliable low latency communication requirement, etc. The problems will be solved theoretically, and the solution algorithms will be tested on publicly available data sets.

Outcomes

  1. Algorithms for control and security of SDN
  2.  Publications in reputed journals and conferences
  3. Patents, if applicable
  4. Tentative: Joint research proposals between the IITD and UQ teams

Information for applicants

Essential capabilities

Networking, machine learning, optimization, queueing theory, reinforcement learning, python programming

Desireable capabilities

Wireless communication, operating systems, industry experience in networking or data analysis

Expected qualifications (Course/Degrees etc.)

B. Tech/M. Tech/MSc(Research) or equivalent in ECE/CS/EE from a top institute with high CGPA and demonstration of research capability

Project supervisors

Principal supervisors

UQ Supervisor

Associate professor Marius Portmann

School of Information Technology and Electrical Engineering
IITD Supervisor

Assistant professor Arpan Chattopadhyay

Department of Electrical Engineering