Exploring the impact of node failures on the resource allocation for parallel jobs