You have a business-critical workload running on a fixed set of Compute Engine instances for several months. The workload is stable with precisely allocated resources. How can you reduce costs for this workload without impacting performance?