PVM supports heterogeneity at the application, machine, and network level. At the application level, subtasks can exploit the architecture best suited to the their solution. At the machine level, computers with different data formats are supported as well as different serial, vector, and parallel architectures. At the network level, different network types can make up a Parallel Virtual Machine, for example, Ethernet, FDDI, token ring, etc. Under PVM, a user-defined collection of serial, parallel, and vector computers appears as one large distributed-memory computer; we use the term virtual machine to designate this logical distributed-memory computer. The hardware that composes the user's personal PVM may be any Unix-based machine on which the user has a valid login and that is accessible over some network.
Using PVM, users can also configure their own parallel virtual machine, which can overlap with other users' virtual machines. Configuring a personal parallel virtual machine involves simply listing the names of the machines in a file that is read when PVM is started. Applications, which can be written in Fortran 77 or C, can be parallelized by using message-passing constructs common to most distributed-memory computers. By sending and receiving messages, multiple tasks of an application can cooperate to solve a problem in parallel.
PVM supplies the functions to automatically start up tasks on the virtual machine and allows the tasks to communicate and synchronize with each other. In particular, PVM handles all message conversion that may be required if two computers use different data representations. PVM also includes many control and debugging features in its user-friendly interface. For instance, PVM ensures that error messages generated on some remote computer get displayed on the user's local screen.