An operating system's scheduling and memory allocation policies attempt to provide fairness among resource principals, as well as graceful behavior of the system under various load conditions. Most operating systems treat a process, or a thread within a process, as the schedulable entity. The process is also the ``chargeable'' entity for the allocation of resources, such as CPU time and memory.
A basic design premise of such process-centric systems is that a process is the unit that constitutes an independent activity. This give the process abstraction a dual function: it serves both as a protection domain and as a resource principal. As protection domains, processes provide isolation between applications. As resource principals, processes provide the operating system's resource management subsystem with accountable entities, between which the system's resources are shared.
We argue that this equivalence between protection domains and resource principals, however, is not always appropriate. We will examine several scenarios in which the natural boundaries of resource principals do not coincide with either processes or threads.