Influence of the trace resolution and length in the cost optimization process in cloud computing

In order to optimize costs in cloud computing deployments, Virtual Machine (VM) allocation strategies are frequently employed. To compare different strategies, workload traces are required. These traces should be of one second resolution (to take into account the per-second billing) and of one year length (to consider reserved VMs properly). However, there are no public traces with these characteristics corresponding to transactional services. To overcome this issue, in this paper we present different synthesis techniques designed to generate a trace of one second resolution based on a trace with lower resolution, usually one hour. The influence of the different synthesis techniques in the cost optimization process is analyzed, concluding that they have an appreciable effect in the optimization process. As an alternative, the execution of VM allocation strategies using traces shorter than a year (a few months, for example) and the extrapolation of the results to the whole year is also analyzed. We found that this mode of operation can generate misleading results.