(no subject)

Brian TAM (btam@cs.ucsd.edu)
Thu, 4 May 2000 05:43:07 -0700 (PDT)

Brian Tam
2000 spring qtr.
paper evaluations

"The Distributed V Kernel and its Performance for Diskless Workstations"

The V kernel is an effort to build an efficient
operating system for a distributed environment
having workstations without secondary storage. This
rather unusual approach was taken because it was
felt that implementing local file systems introduced
significant overhead in system performance. To
compensate for this, the message-based interprocess
communication facility was optimized for fast remote
access, mainly by stripping off the network transport
layer interface to make communication synchronous.
Experiments generally demonstrated that while remote
operations took longer than local actions, performance
was competitive with other distributed systems.

I was impressed with the measurements performed
on this system, which makes this paper seem like a
research paper more than most papers we've read so
far. One thing that struck me was that the V kernel
does not seem to guarantee reliability like other
distributed systems. A major factor in the slower
performance of other systems compared with V, I would
guess, is the mechanism used to provide reliabilty
in those systems. Unavoidably, these mechanisms also
introduced overhead in the IPC of those systems.
More work needs to be done to see if some guarantees
on reliability can also be implemented on a diskless

"The Sprite Network Operating System"

A major objective behind Sprite was to gain
efficient performance from a distributed environment
of time-sharing (multiprocessor) workstations. This
was done with several techniques: building a
transparent network file system, optimizing remote
procedure call, sharing memory between processes,
making processes be able to migrate between nodes,
and caching memory locally.

Like the V paper, the authors here have done
experiments to assess system performance. Aside
from that, this paper was a pleasure to read. It
was well-organized and clear to read, a refreshing
change from previous papers we've had to read.