Bryan Ng

Senior Lecturer School of Engineering and Computer Science

Teaching in 2017

Bryan Ng profile picture

Research Interests

The two main axis of Bryan's research are computer networking and statistics. Current research interests lie in the intersection between the areas of software defined networking (SDN) and analytics/big data, primarily using tools such as regression analysis, Markov chains, decision processes, generalised linear models and machine learning.

Background

Bryan completed his Bachelor's and Master's in the area of computer networking and Ph.D., in the areas of networking and statistics. He held teaching and research positions in Malaysia, Japan and France in addition to attachments to commercial research laboratories such as Intel, Motorola and Panasonic. Prior to his return to academia, he held leadership positions spearheading commercial analytics developments at AC Nielsen, Orange and PanaHome.

Research Topics

Software Defined Networking (SDN)

SDN in a nutshell is a maturing paradigm in networking which espouses the separation between the control plane (think of it as the "brain") and the data plane (imagine the "musculoskeletal system"). The separation of the control and data planes creates plenty of challenges in the ever evolving data networking environment. These challenges are also opportunities, and the next few years will see many emerging startups take on the big wigs for a slice of the SDN pie.

I manage the SDN research centre with Ian Welch. Our wonderful team of engineers & researchers has grown from two in 2014 to ten in 2016. I have championed the approach of validation on a production system and adopting best practices from the networking & software engineering industry. This efforts have seen the SDN centre progress from individuals working on single scripts to teams adopting agile practice on production level platforms within two short years. There has been swirling rumours that we may spin-off a startup based on SDN ..... any takers ?

My interest in SDN is anchored around modelling and performance evaluation to answer real world questions. Some of my recent work on SDN :
  1. Analytical models for SDN - In this work we develop stochastic models using for analysing network performance. The models help network architects manage capital expenditure (CAPEX) and operational expenditure (OPEX) via optimal dimensioning and provisioning to meet customer demands or a prescribed service level of agreement (SLA). There is also significant interest in using models for dealing with regulatory compliance issues. I primarily use queueing theory, Markov chains, generalised linear models, dynamic systems and variable selection methods as fundamental tools for analytical models. Software used: Python, R, Mathematica, C++, QualNet, ns-2 and ns-3, CUDA.
  2. Network analytics - The work in network analytics concerns deriving economic benefits via data analysis in computer networks. Analytics in computer networks differ from census data or retail sales data in that it possesses rich metadata and provenance. I employ fundamental statistical models to data collected from SDN for deriving actionable insights. In this fast emerging area, the SDN research centre has an excellent track record of developing models for helping decision makers achieve their desired results. Software/platforms used: Python, R, SAS, Prometheus, Hive, SparkR, Hadoop, Jenkins, AWS, Ansible, Puppet and SQL.
Sample work:

See publications and software section in SDN Research Centre page.

Analytics/Big data

In this area of research, I attempt to use data to drive discoveries or emerging behaviours. My past experience has been largely with commercial organisations focusing on census, retail measurements, cloud services optimisation, network tomography and construction management. I am particularly interested in variable selection methods, stream analytics, recommendation systems, and classification systems to yield data informed insights. Some topics that are of interest are:
  1. Retail measurement analytics - This topic involves helping decision makers make strategic and tactical marketing decisions based on periodic measurements of marketing variables. We have some projects for individuals interested in churn prediction, universe estimation, sampling design, fraud detection and customer relationship management (CRM). The statistical concepts frequently used are: outlier detection, variable selection, k-means, PCA, CCA, boosting and bagging. Programming languages/platform: R, SAS, python, clipper, Apache kafka, SparkR, FBLearner Flow, WhizzML.
  2. Traffic classification - Traffic classification in this context refers to making classifications based on data traffic movements across the network. The challenge herein lies in making sense of the sheer volume and speed of the data generated. With the completion of a recent project in collaboration with the researchers at WiNe, VUW can boast the world's first software defined traffic classification system - as a bonus we have a freely available prototype of the system on the Internet. Programming languages/platform: R, python, C++, Hadoop, Hive, Ansible, BigML, AutoML.
Read about the details of analytical models in action for real world use cases in (i) wireless networks and (ii) cloud.