Department of Computer Science and Technology

Systems Research Group – NetOS

ACS Projects (2018—2019)

NetOS

This page collects together various ACS project suggestions from the Network and Operating Systems part of the Systems Research Group. In all cases there is a contact e-mail address given; please get in touch if you want more information about the project.

Under construction: please keep checking back, as more ideas will hopefully be added to this page during the coming weeks.


1. Power Efficient Autonomous Flight for Area Surveying

Contact: Cecilia Mascolo, Alessandro Montanari (email)

Area surveying is a costly and often difficult process if performed through human observation: road access, danger from wildlife and simply unknowns make this very hard. Drone based area surveying has the aim to inform the development of new and effective infrastructure by providing timely and fine-grained information on the terrain and area details.

The aim of the project is to explore the power consumption aspects of autonomous flight and research methods to achieve long endurance flight. The project will include a first investigation to identify the most efficient aircraft’s form factor (e.g. multirotor, fixed wing or vertical take-off and landing aircraft) to support endurance flying. Subsequently, the student will devise methods to plan and execute the most efficient trajectories to survey a certain area. The control algorithms will be implemented on popular drone firmwares (e.g. Ardupilot). We expect to test and validate the system with simulations as well as with real-world experiments.

The outcome of this project will constitute a building block of a larger autonomous survey system used for research in the following months.


2. Efficient Area Surveying through Context Aware Drone Mobility

Contact: Cecilia Mascolo, Alessandro Montanari (email)

Area surveying is a costly and often difficult process if performed through human observation: road access, danger from wildlife and simply unknowns make this very hard. Drone based area surveying has the aim to inform the development of new and effective infrastructure by providing timely and fine-grained information on the terrain and area details.

This project aims at optimizing drone surveying flying paths using onboard context sensing and computation. The intuition is that areas with certain characteristics (e.g., presence of human settlements, roads, rivers, etc.) require more time for the survey while others can be completed quicker. Rather than adopting a statically defined path the drone will adapt its path based on the environment in order to achieve longer flight times and cover larger areas.

Different machine learning algorithms will be evaluated in order to find the one that best matches the requirements of low latency and resource consumption. Given that the system will mainly rely on the video feed captured during flight particular emphasis is required on techniques able to process video streams (e.g. deep learning). However other sensors could potentially be included. We expect to test and validate the system with simulations as well as with real-world experiments.

The outcome of this project will constitute a building block of a larger autonomous survey system used for research in the following months.


3. Machine Learning for Network Traffic Classification

Contact: Andrew Moore, Noa Zilberman (email)

In an original piece of research in 2005, by Denis Zuev (Part III) and Andrew W. Moore of the Computer Laboratory, showed how Bayes methods of machine-learning were, when combined with high-quality ground-truth data, able to provide a useful Internet application-identification. In the intervening years there has been a revolution in Internet applications, in machine-learning methodologies, and in applications toward which such information may be put.

This proposal would be to consider extended versions of the original data sets and new data-sets using a number of modern methodologies for training and testing Internet traffic. This would permit us to evaluate such methods in new environments, e.g., the data-center, and to consider the actual value of new methodologies when compared with long-standing mechanisms for constructing Bayesian priors.

A soup-to-nuts proposal machine-learning proposal, the candidate would look at the creating of data-sets for machine learning, apply new (and comparison methods) to this data, and assess the methodologies creating data, features, classes and applying classification scheme for a number of purposes.

As a result, practically this work will start by tagging and preprocessing new datasets and making a baseline comparison with the datasets of the 2005/6/7 work. This will enable a direct comparison of the value of previous algorithms on new datasets and in particular to explore the opportunity for continuous refinement of the prior. Potentially this will also include an exploration of applying recent ML algorithms to both new and old datasets.

Specific details for your project, as applicable, for example:

Background in machine learning - an advantage
Background in computer networks - an requirement
Background in statistics - an advantage

Background material

[1] Internet Traffic Classification Using Bayesian Analysis Techniques Moore AW, Zuev D, ACM SIGMETRICS 2005

[2]Bayesian Neural Networks for Internet Traffic Classification Auld T, Moore AW, Gull SF, IEEE Trans on NN

[3]Probabilistic Graphical Models for Semi-Supervised Traffic Classification Rotsos C, VanGael J, Moore AW, Ghahramani z

Building on some work performed over the suymmer of 2017; the outcome of this project will constitute critiical building blocks of wider-randing research into machine-learning based network control in the following months.


4. Making the right Big Data

Contact: Andrew Moore, Noa Zilberman (email)

Despite the rise and rise of the data-center, the research community has often worked at arms length from real systems. This work will contribute a significant, trace-driven analysis of datacenter operations. Data will be collected for a number of standard operations in DCs and, utilising uncommon levels of instrumentation, measurements will provide insight into the relationship between network operation and data-center application behaviours.

Operational data has provided important insight for architectural decisions in current and future networks. This data is often highly propriatory and rarely made publicly available. Yet if data was comonly understood entire classes of network control and manipulation become available. This work will run-alongside efforts in the SRG to apply machine-learning to network control problems with appropriate opportunities for joint publication arising.

This work will seek high quality data as the first order outcome and as such seeks to identify and achieve data of the highest repeatability. Significant effort will be placed on ensuring the quality of data and recording the provenance of data - alongside ensuring the repreatability of the frameworks used therein.

Specific details for your project, as applicable, for example:

Background in computer networks - an requirement
Background in statistics - an advantage

Background material

[1] Zilberman N, et al. "Where has my time gone?" PAM 2017

[2] Benson T. et al. Network traffic characteristics of data centers in the wild", ACM SIGCOMM IMC 2010

[3] "Internet Traffic Classification Using Bayesian Analysis Techniques" Moore AW, Zuev D, ACM SIGMETRICS 2005

Working longside researchers in the Systems Research Group; the outcome of this project will constitute critiical building blocks of wider-randing research into machine-learning based network control in the following months.


5. Accelerating ML in the network

Contact: Andrew Moore, Noa Zilberman (email)


In-network computing is an emerging research area in systems and networking, where applications traditionally running on the host are offloaded to the network hardware (e.g., switch, NIC). Examples of applications offloaded in the past include consensus protocols, caching and key-value store application.
The objective of this project is to study the applicability of In-Network Computing to accelerating machine learning applications, identify limitations, and - if deemed possible - implement a machine learning function (or part of it) in a network device.

References
[1] P4
[2] Sapio A. et al."In Network Computation is a Dumb Idea Whose Time Has Come." HotNets, pp. 150-156. ACM, 2017.
[3] Bosshart P. et al. "Forwarding metamorphosis: Fast programmable match-action processing in hardware for SDN." SIGCOMM Computer Communication Review, vol. 43, no. 4, pp. 99-110. ACM, 2013.


6. Large-scale Data Processing and Optimisation Projects

Contact: Eiko Yoneki (email)

List of projects on large-scale data processing and optimisation - See here!


7. Trusted Networking Projects

Contact: jon crowcroft (email)

List of projects on trusted networking - See here!