17.op/TEA.ghormley.rodrigues
.ls 2
.na
.LP
Operating System Support for Using a Network of Workstations
as a Supercomputer

Doug Ghormley and Steve Rodrigues

(Professor T. E. Anderson)

AT&T Bell Labs, Digital Equipment, MICRO, and (NSF) CCR-92-57974 (NYI)

For many years, workstations have been gaining in performance more
rapidly than supercomputers. Networks of workstations have advanced
to the point where they can be considered as candidates for the next
generation of supercomputers. Relative to traditional supercomputers,
a network of workstations offers the potential for better
cost-effectiveness, increased scalability, and decreased development
time. Realizing the full performance of a network of workstations,
however, requires attention to 
communication costs and scheduling.

Unfortunately, operating systems for current workstations were
designed for autonomous operation in an untrusted environment and do
not pass the full performance of the hardware available on to the
application. In an environment of trusted, cooperating workstations,
many of the operating system mechanisms currently in use are
unnecessary and even detrimental to performance. Our research is
aimed at providing the network of workstations with the operating
system support appropriate to the environment, that is, low-latency
communication and cooperative scheduling, while still maintaining the
inter-process protection that users have come to expect from the
single-workstation environment.