EV9

qin Zhanhai (qinzhan_hai@hotmail.com)
Thu, 04 May 2000 07:23:29 GMT

________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com
=============================================================================
The Distributed V Kernel and its Performance for Diskless Workstations
=============================================================================
The distributed V kernel is a message-oriented kernel that provides
uniform local and network interprocess communication. The paper is trying
to show us that V kernel is able to be efficiently applied on a high-speed
local network connecting a bunch of diskless workstations that share some
file servers.

In order to achieve the performance of the communication among those
diskless workstations, two more primitives as communication facilities were
added:

ReplyWithSegment(srcpid, dest, src, count) and
ReceiveWithSegment(message, segptr, segsize)

The advantages of these two primitive operations are that they
combine some information while a file read is requested and acknowledged.

By carefully defining the concept of "network penalty", they
measured
the kernal performance, file access performance (page-level file access,
sequential file access, and program loading). And then they concluded that
1) Diskless workstations can access remote files with minimal performance
penalty.
2) The V message facility can be used to access remote files at comparable
cost to any well-tuned specialized file access protocol.

I don't think the paper was organized properly. The context was
too limited so that I can hardly understand why they did this and why they
did that. Instead they put a lot of effort on measuring the performance to
argue that V kernel is good. But I want to see more about how they organized
and implemented the system and a picture of interprocess communication over
LAN. They focused little on this, though.

============================================================================
The Sprite Network Operating System
============================================================================
Sprite is a network operating system, and the spirit of Sprite is
transparency. And sharing is the main strategy for the transparency and high
performance.

The contribution of the work focus on three facilities: a
transparent
network file system, a simple mechanism for sharing writable memory between
processes on a single workstation, and a mechanism for migrating processes
between for migrating processes between workstations to take advantage of
idle machines. They provided two features - multithreaded kernel and Remote
Procedure Calls(RPC) to support multiprocess and network operation.

In stead of using static mount tables to keep track of domains, in
Sprite they use prefix tables to dynamically maintain the domain structure.
The great advantages of prefix tables is that they are created dynamically
and updated automatically when the system configuration changes. For file
data management, they used client and server caches for higher performance
and considered file consistency carefully.

The paper was organized and written perfectly. It's really easy for
readers to grasp why they want Sprite, what the goals of Sprite, and how to
acheive these goals in Sprite. The paper was organized hierarchically and
it's easy to follow every subsection when we have the basic knowledge of
Sprite in mind.