RThreads - a uniform interface for parallel and distributed programming
Bernd Dreier, Markus Zahn, Theo Ungerer · 1996
Introduction Today, computer networks grow in size and importance. Workstations coupled by a highspeed network represent an efficient parallel virtual machine. A workstation cluster is often composed of multiprocessor workstations. The processors within a multiprocessor workstation share global memory, whereas a workstation cluster is coupled by a standard network, e.g. Ethernet, FDDI, or ATM. Synchronization of parallel activities among the processors within a workstation is done by access to global variables. In a workstation cluster messages are sent via UNIX ports. Communication within a workstation is much faster than between the workstations of a cluster. As a consequence, medium-grained parallelism can be successfully exploited within a multiprocessor workstation, but only very coarse-grained parallel activities should be distributed over the workstations of a cluster. However, a network of dozens or hundreds of workstations is much more powerful than a multip