Synthesis of Adaptation Plans for Cloud Infrastructure with Hybrid Cost Models

We present a method and a tool for devising adaptation plans for the deployment of software applications on cloud computing infrastructure comprising a combination of reserved and "pay-as-you-go" resources. Starting from a trace log that represents the expected workload pattern of the application, we exploit queuing theory results to synthesize an adaptation plan that manages the cloud elasticity in a cost-effective way. To this end, the amount of resources to be reserved for a long period is quantified in a first activity of the method. Then, as a second activity, the number and types of pay-as-you-go resources is assessed. The adaptation plan produced for a real web server is shown to be cost effective even when the actual workload differs from that in the trace log by as much as 5 -- 20%.