The broadening of the Grid architecture to include commercial as well as scientific workloads raises the possibility of commercial datacentres allocating computational resources on demand. The Utility Datacentre provides a large pool of computational resources that are allocated to service applications on request. When a service application consists of several components, these may be allocated physical computational resources which are distributed across the datacentre. The quality of placement determines how well or badly the application consumes finite global resources, such as bandwidth in the network backbone. This paper examines the placement problem, develops metrics for placement quality, and provides a decentralised approach to improving placement quality based on rounds of resource trading.