Research Projects

Policy-driven Data Centre Resource Management

In modern Cloud Data Centres, correct implementation of network policies is crucial to provide secure, efficient and high performance services for tenants. It is reported that the inefficient management of network policies accounts for 78% of DC downtime, challenged by the dynamically changing network characteristics and by the effects of dynamic Virtual Machine (VM) consolidation. While there has been significant research in policy and VM management, they have so far been treated as disjoint research problems.

In this branch of research, we explore the simultaneous, dynamic VM and policy consolidation problem. We then propose Sync and PLAN, efficient and synergistic schemes to jointly consolidate network policies and virtual machines. Extensive evaluation results and a testbed implementation of our controller show that policy and VM migration under these schemes significantly reduces flow end-to-end delay by nearly 40%, and network-wide communication cost by 50% within few seconds, while adhering strictly to the requirements of network policies.

  • Lin Cui , Fung Po Tso, Dimitrios P. Pezaros, Weijia Jia, Wei Zhao (2016): PLAN: Joint Policy- and Network-Aware VM Management for Cloud Data Centers. In: IEEE Transactions on Parallel and Distributed Systems, 2016, ISSN: 1045-9219.

    BibTeX (Download)

    @article{Cui2016Plan,
    title = {PLAN: Joint Policy- and Network-Aware VM Management for Cloud Data Centers},
    author = {Lin Cui , Fung Po Tso, Dimitrios P. Pezaros, Weijia Jia, Wei Zhao},
    url = {http://www.poscotso.com/wp-content/uploads/2016/09/policy_aware_vm_migration.pdf},
    issn = {1045-9219},
    year  = {2016},
    date = {2016-08-19},
    journal = {IEEE Transactions on Parallel and Distributed Systems},
    keywords = {policy},
    pubstate = {published},
    tppubtype = {article}
    }
    
  • Fung Po Tso, Lizhuo Zhang, Weijia Jia (2007): Video Surveillance Patrol Robot System in 3G Internet and Sensor Networks. ACM Conference on Embedded Networked Sensor Systems (SenSys ), 2007.

    BibTeX (Download)

    @conference{Tso2007,
    title = {Video Surveillance Patrol Robot System in 3G Internet and Sensor Networks},
    author = {Fung Po Tso and Lizhuo Zhang and Weijia Jia},
    url = {http://www.poscotso.com/wp-content/uploads/2016/03/p395-tso.pdf},
    year  = {2007},
    date = {2007-06-09},
    booktitle = {ACM Conference on Embedded Networked Sensor Systems (SenSys )},
    keywords = {publications},
    pubstate = {published},
    tppubtype = {conference}
    }
    
  • (): . .
  • (): . .
  • (): . .
  • (): . .

Glasgow Raspberry Pi Cloud

We are currently building a scale mode of cloud data centre with many Raspberry Pis, Lego bricks and an awful lot of cables! Please visit our project blog for more details: http://raspberrypicloud.wordpress.com

  • (): . .

    BibTeX (Download)

    @{,
    keywords = {policy, policy, policy, policy, policy, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications},
    pubstate = {},
    tppubtype = {}
    }
    

Distributed and Traffic-Aware Virtual Machine Management for Cloud Data Centres

A limited number of studies have proposed traffic-aware VM management schemes that try to minimise the impact of virtualisation on the DC network. However, the proposed algorithms are either centralised and therefore do not scale well to the full size of today’s DCs, concern the initial placement of VMs and do not deal with maintaining steady-state throughout the system’s evolution, or they only consider bandwidth allocation at the lower host-to-network layers overlooking congestion that happens in a significant fraction of the core links even when parts of the DC infrastructure remain underutilised. In this project, we developed scalable communication cost reduction scheme for intra-DC workloads that dynamically re-allocates VMs through live migration and minimises the communication cost incurred by the resulting traffic dynamics in an always-on manner, while adhering to server-side resource capacity boundaries. We provided testbed implementation based on software defined networking (SDN).

  • (): . .

    BibTeX (Download)

    @{,
    keywords = {policy, policy, policy, policy, policy, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications},
    pubstate = {},
    tppubtype = {}
    }
    
  • (): . .
  • (): . .

    BibTeX (Download)

    @{,
    keywords = {policy, policy, policy, policy, policy, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications},
    pubstate = {},
    tppubtype = {}
    }
    

Longer is Better!

Traffic agonistic network protocols have failed to leverage rich path redundancy in today’s Cloud data centre network, resulting in overall network utilisation whilst still seeing episodes of network congestions. However, in order for Cloud computing to be profitable, the operators have to keep their network utilisation as high as possible. In this project we are researching into an adaptive flow scheduling scheme which proactively measure network link utilisation and then opportunistically schedule flows over least utilised links even when paths are slightly longer. We are also constructing a testbed consists of programmale switches – NetFPGA boxes and OpenFlows switches.

  • (): . .

    BibTeX (Download)

    @{,
    keywords = {policy, policy, policy, policy, policy, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications},
    pubstate = {},
    tppubtype = {}
    }
    
  • (): . .
  • (): . .

    BibTeX (Download)

    @{,
    keywords = {policy, policy, policy, policy, policy, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications},
    pubstate = {},
    tppubtype = {}
    }
    
  • (): . .

DragonNet – Robust Internet System for Trains

With the tremendous popularity of Wi-Fi capable devices such as laptops, netbooks, and smartphones, rail operators are rushing to deploy high-speed wireless networks in a bid to lure potential passengers to travel by railway. Existing infrastructures for providing Wi-Fi to Internet access are realised by relaying WLAN traffic via a cellular network, satellite, trackside WiMAX, or leaky coaxial cable (LCX) to the backbone network. However, there are still some barriers that hinder the use of these technologies. Satellite communications are not ideal for high-speed access to trains since satellite links have limited bandwidth and long roundtrip times (RTT). WiMAX access creates an enormous financial burden for the large-scale installation of trackside WiMAX APs and equipment maintenance thereafter, and so does the LCX. On the contrary, the cellular-based infrastructure takes advantage of an existing cellular architecture for reducing the deployment cost. However, handoffs between base stations and drastic fading phenomena can easily cause severe deterioration in signal strength of a certain client device to an unacceptable level, resulting in degraded network performance. In this project, we developed DragonNet, a gateway 500 meters long that aims at handling single-point failure gracefully. DragonNet is formed by a DragonNet router (or D-router; we will use D-router and node interchangeably thereafter in this paper) chain running through the whole length of LDTs. Keywords: Long-distance train, mobile Internet, random failure, cascading failure, DragonNet, multipath routing.

  • (): . .

    BibTeX (Download)

    @{,
    keywords = {policy, policy, policy, policy, policy, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications},
    pubstate = {},
    tppubtype = {}
    }
    
  • (): . .

    BibTeX (Download)

    @{,
    keywords = {policy, policy, policy, policy, policy, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications, publications},
    pubstate = {},
    tppubtype = {}
    }
    

Student Projects

The Raspberry Pi Cloud: Development of interactive e-learning system for Cloud computing

The objective of this project is to develop an experimental evaluation platform based on the PiCloud to facilitate pioneer Cloud computing teaching at the School of Computing Science. The project will focus on building a software system to accept and execute experiments, and to return feedback results to students. It will consist of a web-based management front-end and a back-end engine able to run an extended set of algorithms. We will implement a low-end MapReduce engine, and a distributed consensus algorithm animator. Students will then be able to submit and execute experiments with different input, and then view explanatory results and feedback through the web interface.

Network Measurement as a Service

The objective of this project is to develop a software defined networking (SDN) approach that collects real-time network statistics and offer them as a service for other applications. This project rides on both the sFlow standard which specifies instrumentation in the forwarding table hardware that provides real-time, network-wide visibility into traffic flowing across the network, and the OpenFlow protocol allows controller software running on a server to configure the hardware forwarding tables in a network of switches. Combined, sFlow and OpenFlow can be used to construct feedback control systems that optimise performance, automatically adapting the network to meet changing demands, offer exciting opportunities for delivering breakthrough data centre and cloud networking performance.

Big Data for Mini Computers?

The ARM-based servers are gaining their popularity in Cloud data centre environment. Intuitively, one potential scenario in the future development of Cloud Computing is having computer clusters that are built entirely out of ARM-based servers. The Raspberry Pi Cloud, a miniature Cloud infrastructure built out of 56 Raspberry Pis, is just such a pioneer architecture. The most prominent advantage of using low-cost processors is their power-efficiency which is believe to be beneficial to sustainable and greener computing. Nevertheless, their suitability for big data paradigm still remains largely unknown. Towards this end, the goal of this project is to design and implement a new hadoop operational architecture that will harness collective number resource constraint mini computers for big data applications.

Is Raspberry Pi Cloud Cheaper?

The goal of this project is to benchmark Raspberry Pi Cloud’s performance metrics, e.g., CPU utilisation, I/O speed and utilisation and power consumption, etc., with respect to real jobs; and then compare results with x86 architecture to demonstrate the “real cost” of Raspberry Pi Cloud.