Parallel programming on loosely coupled distributed systems is becoming a viable approach with the rapid increase in network speeds and availability of large amounts of unused CPU capacity on individual workstations. Parallel programs are often written for a specific configuration of the distributed system such as the number of nodes, their relative speeds and their network connections. These programs perform poorly when there is a change in the configuration or when they are taken to a system other than what they are intended for. We propose a new paradigm which can express the scale of a program as a part of the program itself. The scale of the program is specified in an abstract manner for an arbitrary number of nodes, their relative speeds and their network connections. The runtime system uses this information to decide the actual scale of program on a given distributed system.