John-Paul Fryckman (fryckman@SDSC.EDU)
Thu, 4 May 2000 07:42:11 -0700 (PDT)


Cheriton and Zwaenepoel write about a kernel for diskless
workstations that yields transparency between local and
network interprocess communication. Diskless workstations
are a low cost more reliable alternative to workstations with
disks. The paper addresses some of the key issues that this
generates with the leading concern being network performance.

The heart of this OS as well as many others is IPC. The V kernel
provides for short message bursts along with data transfer
primitives. Another point is that processes on a network of
computers running V have a shared process identifier space.
The kernel also provides for location abstraction so the processes
do not have to worry about where the receiving processes are
located at. And the authors proceed to establish the filesystem
ontop of IPC.

The last half of the paper describes the performance of the
kernel along with various user operations. The authors are
very complete in their analysis as well as describing their


Sprite is all about giving the performance of a huge time-sharing
mainframe to a network of workstations. The authors claim that
new developing networks, large memories, and multiprocessors
make this possible. Sprite is based on BSD. However it adds
resources for a transparent network file system and mechanisms
for sharing writable memory between processes and process
migration. Another feature that it provides is transparent
procedure calls. If the call is located on another machine
a stub is generator outside of the calling process and the
RPC transport layer carries it to the proper machine in which
the reverse happens. This reinforces their main goal of
merging distributed computers into a single time-sharing like
mainframe. To this end the implemented a system that encourages
sharing of resources and spare CPU cycles that processes can
be migrated to. The authors have done a nice job in defining
all areas of their OS and have successful implemented their