The PVM system, a software framework for heterogeneous concurrent comp
uting in networked environments, has evolved in the past several years
into a viable technology for distributed and parallel processing in a
variety of disciplines. PVM supports a straightforward but functional
ly complete message passing model, and is capable of harnessing the co
mbined resources of typically heterogeneous networked computing platfo
rms to deliver high levels of performance and functionality. In this p
aper, we describe the architecture of PVM system, and discuss its comp
uting model, the programming interface it supports, auxiliary faciliti
es for process groups and MPP support, and some of the internal implem
entation techniques employed. Performance issues, dealing primarily wi
th communication overheads, are analyzed, and recent findings as well
as experimental enhancements are presented. In order to demonstrate th
e viability of PVM for large scale scientific supercomputing, the pape
r includes representative case studies in materials science, environme
ntal science, and climate modeling. We conclude with a discussion of r
elated projects and future directions, and comment on near and long-te
rm potential for network computing with the PVM system.