Abstract

In this work we formulate strategic decision models describing when and how many reserved instances should be bought when outsourcing workload to an IaaS provider. Current IaaS providers offer various pricing options for leasing computing resources. When decision makers are faced with the choice and most importantly with uneven workloads, the decision at which time and with which type of computing resource to work is no longer trivial. We present case studies taken from the online services industry and present solution models to solve the various use case problems and compare them. Following a thorough numerical analysis using both real, as well as augmented workload traces in simulations, we found that it is cost efficient to (1) have a balanced portfolio of resource options and (2) avoiding commitments in the form of upfront payments when faced with uncertainty. Compared to a simple IaaS benchmark, this allows cutting costs by 20%.

Share

COinS
 

Strategic Decision Support for Smart-Leasing Infrastructure-as-a-Service

In this work we formulate strategic decision models describing when and how many reserved instances should be bought when outsourcing workload to an IaaS provider. Current IaaS providers offer various pricing options for leasing computing resources. When decision makers are faced with the choice and most importantly with uneven workloads, the decision at which time and with which type of computing resource to work is no longer trivial. We present case studies taken from the online services industry and present solution models to solve the various use case problems and compare them. Following a thorough numerical analysis using both real, as well as augmented workload traces in simulations, we found that it is cost efficient to (1) have a balanced portfolio of resource options and (2) avoiding commitments in the form of upfront payments when faced with uncertainty. Compared to a simple IaaS benchmark, this allows cutting costs by 20%.