Resource Prediction Service for Efficient Execution of Bioinformatics Workflows in Federated Cloud with Machine Learning

Cloud federation emerged to extend the resources available between different interconnected cloud providers for transparent and unlimited availability to the end-user. Cloud orchestration platforms have become a way to centralize demands for high computational power in applications such as Bioinformatics workflows. The large quantity of resources available among several providers in a federation makes it challenging to choose a suitable one for particular workflows. This work proposes a Machine Learning Resource Prediction Service called sPCRAM. sPCRAM uses a machine learning model combined with a GRASP metaheuristic to transparently and adequately dimension the resources, determining the monetary cost and the runtime before the workflow execution. sPCRAM interactively allows the user to set the execution type, calibrate time and cost. Such executions can have, for example, long duration and low cost, as well as a shorter duration and a higher cost. The results demonstrate that sPCRAM can appropriately estimate runtime and cost for cloud federation resources on average 97,70% faster than the brute force technique for resource selection.