A method and mechanism for a distributed on-demand computing system. The system automatically provisions distributed computing servers with customer application programs. The parameters of each customer application program are taken into account when a server is selected for hosting the program. The system monitors the status and performance of each distributed computing server. The system provisions additional servers when traffic levels exceed a predetermined level for a customer's application program and, as traffic demand decreases to a predetermined level, servers can be un-provisioned and returned back to a server pool for later provisioning. The system tries to fill up one server at a time with customer application programs before dispatching new requests to another server. The customer is charged a fee based on the usage of the distributed computing servers.