Cloud computing provides a highly scalable, distributed infrastructure on which applications and data can be hosted. Hosted applications must perform within specified constraints, despite running over an extremely dynamic infrastructure. This dynamic behaviour is created, on one hand, by the applications running within virtual machines hosted on heterogeneous servers, and on the other by the constantly fluctuating demand on the network link between the virtual machine and the user of the application. This elasticity of infrastructure, and fluctuating demand placed on it, makes it difficult to ensure QoS for the application. It is no longer sufficient to measure the performance of various host servers within a data centre - it is also now required to measure the performance of the application where it is most critical with the user of the application. The efficiency of application hosting is also difficult to ensure. Optimisation of resource allocation should minimise the hosting cost while ensuring the delivery of the application to the end-user is acceptable. This thesis details a programme of research aimed at designing, implementing and using a distributed, highly-scalable system for monitoring and ensuring end-to-end, SLA-compliant performance of virtual private cloud applications, while minimising the overall hosting cost for the cloud. The research undertaken and described within this document reviews cloud computing, the underlying virtualisation technologies and optimisation techniques, and presents three models with completed prototypes. The first model is a cloud resource monitoring methodology, which acquires real- time metrics from live applications and hosts running in a virtual private cloud test- bed environment. This model is an adaptation of the Analytical Hierarchy Process (AHP), aimed at providing a SLA-compliant scoring mechanism for all criteria to be considered when measuring application performance. The model presented finally allows an overall score to be calculated by which the application's performance is assessed at any potential host where it may be placed. The second model is a cloud-centric network monitoring methodology, which analyses the quality of the network link between any two I.P. addresses. This system provides network quality information to be used in the assessment of the application's (potential) performance between any given host location and the end-user. The third model is an optimiser for SLA-compliant virtual machine placement within virtual private clouds, where the hosting cost is minimised. In summary, the key outputs of this thesis are: 1) a prototype vendor-agnostic monitoring system, for virtual private clouds, which provides the ability to acquire any metric, relevant to the successful delivery of an application to the end-user, and provide information regarding real-time SLA-compliance based on performance thresholds set for those metrics 2) a live system for projecting SLA-compliant application performance on any given host 3) a hierarchical framework for dynamically defining application SLAs, with any conceivable, acquirable metric 4) an optimisation engine with the goal of optimally placing applications on host servers with the goal of reducing the hosting cost across all host servers. The hosting costs for each application on each server is provided by the monitoring system 5) a software-based, virtualised tool for automatic monitoring of network performance, with a remote data logging component 6) An agent-based system for automatically and periodically acquiring VM and host metrics from a VMWare virtual private cloud. The thesis presents an overall fully-functional system incorporating all these aspects.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:588494 |
| Date | January 2012 |
| Creators | McConnell, Aaron |
| Publisher | University of Ulster |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.0017 seconds