Autonomic Workload Management
Wherever services are provided, service users have expectations; Service Level Agreements (SLAs) make explicit what expectations users can realistically place on a service provider, and may be associated with a charging model that determines the remuneration associated with certain Qualities of Service (QoS). Whether or not formal agreements are in place, decisions must nonetheless be made that influence the behaviours users experience, and service providers must put in place mechanisms that make such decisions. In dynamic and complex environments, such as clouds or distributed systems, mechanisms that make such decisions must be under autonomic control, to accommodate potentially rapidly changing workloads.
As such, meeting QoS expectations involves continuously reconsidering decisions as to where specific software services are deployed, what resources are allocated to the different deployments, what requests are allocated to different resources, how resources are configured to support specific types of request, etc. Furthermore, the changes to the behaviour of each component need to be carried out in a way that doesn’t undermine the performance of any of the others. As such, autonomic decision making can potentially affect many different aspects of many different software platforms. Unfortunately, research to date on autonomic systems falls well short of being able to address such challenges. Current research typically gives rise to isolated islands of autonomic behaviour, with little understanding of how to coordinate alternative responses either within or across autonomic components. As a result, unpredictable aggregate behaviour may result from a single platform providing several different, and not entirely independent, responses to changes in circumstances. Indeed, different autonomic systems operating in a shared environment may change their behaviours in mutually disadvantageous ways. This PhD proposal is make autonomic systems design and development more systematic and effective by improving understanding of individual decision making techniques, and by investigating techniques for coordinating multiple autonomic mechanisms.
Research in Autonomic Information Management at Manchester has explored autonomic workload management in distributed query processing and workflow systems using heuristic and utility-based techniques to address specific autonomic tasks. The aim of the proposed project is to investigate decision making in autonomic computing, with a view to improving understanding of how and where specific policies can be applied effectively, and to demonstrate how autonomic components can be coordinated to provide globally desirable behaviours. Thus the proposed research seeks to understand features of specific workload management problems, and to investigate empirically which autonomic decision making techniques (in particular heuristic, utility-based, learning-based and economic) are most suitable for specific problem types and why.
This School has two PhD programmes: the Centre for Doctoral Training (CDT) 4-year programme and a conventional 3-year PhD programme.
School and University funding is available on a competitive basis.
For further details, please see our funding pages here: http://www.cs.manchester.ac.uk/study/postgraduate-research/programmes/phd/funding/
The minimum requirements to get a place in our PhD programme are available from:
How good is research at University of Manchester in Computer Science and Informatics?
FTE Category A staff submitted: 44.86
Research output data provided by the Research Excellence Framework (REF)
Click here to see the results for all UK universities