Availability-Aware Virtual Resource Provisioning for Infrastructure Service Agreements in the Cloud

Service availability is a key construct in Service Level Agreements (SLA) between a cloud service provider and a client. The provider typically allocates backup resources to mitigate the risk of violating the SLA-specified uptime guarantee. However, initial backups may need to be adjusted in response to real-time failure and recovery events. In this study, we first develop a recurrent intervention at fixed intervals (RIFI) strategy that allows the provider to adjust the allocation of backup resources such that the expected total cost is minimized. Next, we focus on the limit to number of interventions, starting from single intervention strategy, as frequent reallocations may be operationally disruptive. Particularly, we provide a cost minimization approach to guide the service providers in their virtual resources management, and a specific downtime minimization approach for more mission-critical applications as a more aggressive alternative. We present computational results exploring the impact of intervention on the likelihood of SLA violation for the rest of the contract period, and evaluate parameters such as time and quantum of resource level adjustment, penalty levels desired by clients, and their influences on the backup resource provisioning strategies. We also validate our models through the analysis of use cases from Amazon Elastic Compute Cloud. Finally, we summarize this study by providing key practical managerial implications for resource deployment in the availability-aware cloud.

Automated summary

This study focuses on the availability of cloud services and proposes a strategy for adjusting the allocation of backup resources to minimize expected total cost. The study also considers the limitation on the number of interventions and provides cost minimization and downtime minimization approaches for resource management. The findings are validated through analysis of real-world use cases and provide practical managerial implications for resource deployment in availability-aware cloud environments.