A Case for NOW (Networks of Workstations)

			   David E. Culler
		      Computer Science Division
		  University of California, Berkeley

The natural evolution of the "killer micro" combined with the
emergence of the "killer network" is likely to position networks of
workstations (NOWs) as the primary computing infrastructure for
science and engineering.  They will provide a vast pool of
computational and storage resources for demanding applications too
large for the desktop, as well as the vehicle for interactive
computing.  In addition to the migration of parallel computing to a
mainstream infrastructure, this transformation brings the opportunity
to dramatically improve virtual memory and file system performance by
using the aggregate DRAM of a NOW as a giant cache for disk and by
using collections of workstation disks as a soft RAID.  We describe
the technical challenges in exploiting these opportunities, including
efficient communication hardware and software, global coordination of
multiple workstation operating systems, scalable management of
metadata, and the tension with commodity system implementations.  We
discuss progress in the Berkeley NOW project toward a 100-node NOW
prototype that demonstrates practical solutions to these challenges.