In the past, researchers have conducted experiments linking workstations that provide on the order of 1 to 10 MIPS. Such experiments have included remote execution, computer farms, and migration of computations.
More recently, experiments have focused on linking higher-performance workstations (those providing on the order of 10 to 100 MFLOPS) together with multiprocessors and conventional supercomputers.
To fully exploit these multiple computer configurations, researchers have developed various software packages that enable scientists to write truly heterogeneous programs. Examples of such software packages include Express, P4, Linda, and PVM. Each package is layered over the native operating systems, exploits distributed concurrent processing, and is flexible and general-purpose; all exhibit comparable performance. Their differences lie in their programming model, their implementation schemes, and their efficiency.
Recent conferences as well as informal discussion media seem to indicate that most attention is focused on the four systems mentioned above- Express, P4, Linda, and PVM-in terms of use by the high performance scientific computing community. In the remainder of this section, we present brief outlines of each of the first three, with a detailed description of PVM following in Section 3. We wish to emphasize however, (1) that these systems are by no means the only software packages in existence; and (2) the descriptions that follow are not detailed and formal critiques, but rather brief synopses abstracted from our understanding of the systems and the developers' own articles or communications.