Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.573405
Title: Exploration of scaling properties in cloud data centres
Author: Sriram, Ilango Leonardo
Awarding Body: University of Bristol
Current Institution: University of Bristol
Date of Award: 2011
Availability of Full Text:
Access from EThOS:
Abstract:
In the past five years, the phrase "cloud computing" has come into common usage to mean remotely-hosted computing services, where the provider of the service relies on one or more highly automated large-scale data centres as the computing infrastructure. With increasing demand for cloud computing, economies of scale are causing a race towards ever-larger data centres. With these increases in size comes a problematic growth in system complexity: Many conventional management techniques that work well when controlling a relatively small number of data-centre nodes become impracticable on larger scales. There is currently very little engineering theory, or experience-based teaching, that can be brought to bear on the design and management of large-scale data centres. One major issue facing traditional academic and industrial research facilities is that, contrary to the way data centre designers used to work, in the case of cloud data centres the systems they are able to use for detailed development and testing are always going to be much smaller than the final systems that go into production. For this reason, it is fair to characterise much of the current development work as being more art than science, and this imprecision can lead to costly errors. In almost all of current engineering practices, predictive simulation studies are used for rapid exploration and evaluation of design alternatives before they go into production. This helps avoiding costly mistakes. Despite this well-established tradition of computational modelling and simulation, there are currently no comparable tools for cloud-scale computing data centres. The research work described in this thesis is motivated by exactly that problem. I argue that there is a need for tools that allow owners and man- agers of cloud computing infrastructure to evaluate alternative designs, and to answer "what-if" questions. In many other areas of engineering, predictive computer simulation systems allow engineers to explore aspects of a design for some artefacts without that artefact actually having to be constructed. I have developed SPECI (Simulation Program for Elastic Cloud Infrastructures) and released it to the open-source community as a first step towards meeting this need. The research presented in this thesis covers several disciplines, presenting the cloud middleware model of component policy subscription updates that are used to manage services in the system, introducing simulation to this model, and using recent advances from complex network theory to model subscription distributions. It is a first step towards developing adaptive data-centre management policies that "intelligently" and dynamically organise and reorganise the network of components that work together within the data-centre in light of changing demands.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.573405  DOI: Not available
Share: