Burstiness and Uncertainty Aware Service Level Planning for Enterprise Clouds

As enterprises begin to increasingly adopt the cloud paradigm, Cloud Service Providers (SPs) need tools to help them plan their infrastructure capacity and decide on Service Level Agreements (SLAs) with customers prior to deploying their customers' applications. Service Level Objectives (SLOs) are specified for customers' applications as part of customers' SLAs with cloud SPs. A cloud SP need Service Level Planning (SLP) tools that consider the workloads of the applications deployed on the cloud to determine the adequate capacity required to satisfy the applications' SLOs. Existing SLP approaches have not considered important challenges such as workload burstiness, workload uncertainty, and scalability to large number of applications. This thesis presents an SLP framework that addresses the above challenges simultaneously. The framework implements a novel Resource Allocation Planning (RAP) method to identify a time varying allocation of resources to applications to satisfy their bursts. RAP is a heuristic optimization technique that in conjunction with a trace-driven performance prediction technique estimates the near minimal degree of service level violations that the cloud SP can incur with a given cloud resource capacity. RAP works in consort with a Monte Carlo simulation technique, which allows cloud SPs to systematically consider the impact of workload uncertainty in SLP. Finally, a new burstiness-aware workload clustering algorithm is proposed to increase the scalability of the SLP framework while preserving workload burstiness. Detailed simulation results are presented to characterize the behaviour of the proposed SLP framework. The results show that the proposed RAP variants can identify optimal or near optimal resource allocation plans without exhaustively generating all possible plans. Secondly, the results show that RAP can permit cloud SPs to more accurately determine the capacity required for delivering specified SLOs compared to other competing techniques especially for bursty workloads. Thirdly, the results demonstrate that the proposed Monte Carlo simulation technique enables cloud SPs to accurately estimate the impact of workload uncertainty in their SLP exercises without exhaustively traversing all combinations of application workload scenarios. Finally, the results show that the proposed workload clustering algorithm reduces the number of computations needed to support SLP exercises without significantly impacting accuracy.

Name thesis:degree_name

Doctor of Philosophy (PhD)

Discipline thesis:degree_discipline

Electrical and Computer Engineering

Grantor dc:publisher.institution

Graduate Studies

Year dc:date.issued

2014

Author dc:creator

• Youssef, Anas

Advisor dc:contributor.advisor

• Krishnamurthy, Diwakar

• Computer Science
• Engineering--Operations Research

Statement dc:rights

• University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.

Language dc:language.iso

eng

DOI

http://dx.doi.org/10.11575/PRISM/25188

OAI identifier oai:identifier

oai:ucalgary.scholaris.ca:11023/1318