A service workflow including ordered services is received, and a heuristic utility value is calculated for each service. A best node having a smallest heuristic utility value for a service is selected, and a best node identifier is placed in a node list. If the best node includes a parallel sub-workflow, potential next nodes are identified by generating potential next nodes from a data center that can perform a service associated with the best node with a minimum run-time value. Otherwise, potential next nodes are generated based on a data center associated with the service. A heuristic utility value is determined for each potential next node, and a new best node is selected based on the heuristic utility values. The identifying, determining, and selecting operations are repeated until the best node contains only the last ordered service. Data centers for each ordered service are identified based on the best node.