Risk informed service level agreement for cloud brokerage

• Main Author: Li, Bin
• Published: University of Surrey 2012
• Subjects: 004.360218
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.580347

The use of distributed computing systems, including Grids and Compute Utilities, and now Clouds, becomes a consideration for businesses hoping to manage start-up costs and times, as well as reduce the physical and environmental footprint of infrastructures. Instead of purchasing and maintaining hardware and software, organisations and individuals can take advantage of pay-per- use (utility) models that relate directly to their requirements of infrastructures, platforms and software. However, such metered services are not widely adopted yet due to the lack of assurance of Quality of Service (QoS). It is suggested that such systems will only attain greater acceptance by a larger audience of commercial consumers if binding Service Level Agreements (SLAs) are provided that encompass service descriptions, costs of provision and, importantly, assurances on availability, performance, and liability. Prediction, quantification of risk, and consideration of liability in case of underperformance are considered essential for the future provision of Computer (Cloud) Economics - in particular, for the provision of SLAs through resource brokers, and generally to be more comparable to financial markets. The principal focus of this thesis is on building brokerage and related services for supporting growth of Cloud and contributing to future computational economics. A brokerage should provide negotiation mechanisms between consumers and providers, and perhaps manage available computer resources, to realise the goals of both parties. SLAs are key to this, where each SLA details price, risk, performance and QoS parameters, amongst others. This thesis presents a novel approach that supports the creation and management of Service Level Agreements, aimed towards improved uptake of cornmoditised computational infrastructures, platforms and software services. By analysing issues within current SLAs, it summarises necessary characteristics to be addressed in Cloud SLAs. Inspired by financial portfolio analysis and in particular by credit derivatives, this work demonstrates how the proposed Cloud Collateralised SLA Obligations (CSO), analogous to synthetic collateralised debt obligations (CDO), can be used to mitigate risk of failure or underperformance through diversification of compute resource portfolios. The CSO prices risk integrates into service insurance, and builds in penalties, and in contrast to well-known Cloud price models, relates variable performance to variable price. This performance-price relationship would also be necessary for the appropriate use of other financial models. Through Value-at-Risk (VaR) style analysis, the probability of failure (risk of underperformance) can be related to a confidence level for each SLA offer - the confidence of meeting the SLA. The thesis further identifies how performance tranches support an autonomic aspect in attempting to ensure satisfaction of higher-value SLAs as a trade off against higher-risk, lower-value SLAs. The approach can readily integrate with any SLA framework that supports real- time dynamic characteristics. Outcomes are broadly relevant to Cloud Computing, and more specifically to Infrastructure as a Service Clouds.