AFS (was Re: Mainframes & Unix)
Stefan Monnier
le...@cs.unc.edu (Jon Leech) writes:
> Yes, exactly why we ended up with it. It is of significant benefit to
> administrators, not to users, and sysadmins, like anyone else, naturally
> tend to serve their own interests first when in a position to do so.
As a user, I must say that (even though the cron-job, mail-filtering and other
such non-interactive uses were a problem) I loved AFS (don't have access to it
any more). I especially loved its ACLs, the ability to give specific access to
someone who doesn't have an account in your cell (but does in another) and the
ability to access files from "any place" (which is theoretically possible with
NFS, but practically isn't for reasons of efficiency and security).
Also the consistency guarantees of AFS may not be the same as UFS', but they
are much better than NFS'. Have you already tried a distributed make with NFS ?
AFS is not the panacea, but it's a step in the right direction.
And about the security issue, I'm wondering how this problem is supposed to be
solved. (I can't find any security-related group that seem related to this
issue). How can the file-server be sure that it's YOU who has decided to run
script X at time Y if you're not present ? Thgat is: how should your identity
be kept around for later use ?
Stefan
Bryan O'Sullivan
s> I especially loved its ACLs, the ability to give specific access to
s> someone who doesn't have an account in your cell (but does in
s> another) and the ability to access files from "any place" (which is
s> theoretically possible with NFS, but practically isn't for reasons
s> of efficiency and security).
It's worth noting that Solaris 2.5 supports ACLs over NFS, and also
allows you to access files from more or less "any place" (modulo the
usual security concerns).
s> Also the consistency guarantees of AFS may not be the same as UFS',
s> but they are much better than NFS'. Have you already tried a
s> distributed make with NFS ?
As a practical matter, I have never been bitten by consistency
problems when performing tasks like distributed builds. Such problems
do bite upon occasion, though, it is true.
s> AFS is not the panacea, but it's a step in the right direction.
NFS is catching up rapidly in many respects (admittedly after quite a
long time), and it is a lot cheaper to the customer than are AFS and
DFS.
<b
--
Let us pray:
What a Great System. b...@eng.sun.com
Please Do Not Crash. b...@serpentine.com
^G^IP@P6 http://www.serpentine.com/~bos
Pradip Srimani
For a Postscript version, Please check the URL
http://www.cs.colostate.edu/~srimani/call-ijcs.ps.Z
CALL FOR PAPERS
Special Issue of
INTERNATIONAL JOURNAL OF COMPUTER SIMULATION
on
Modeling and Simulation of Routing
Protocols in Computer Networks
Due to fundamental physical limitations on processing speeds
of sequential computers, the future-generation high performance
computing environment will eventually rely entirely on exploiting
the inherent parallelism in problems and implementing their
solutions on parallel machines. Networks and their performance
become an integral part of high performance computing in
general. Although most of the real life parallel machines use
either mesh or hypercube connections, in the recent years a
number of other networks have been proposed that seem to be
attractive alternatives. On the other hand, although most of the
machines use circuit switching to route messages, recent
advancement in routing protocols (wormhole routing, deflection,
randomized and hybrid etc) need be properly investigated and
studied to make the right choice for a network. Simulation is the
way to study various aspects of all these different protocols when
implemented on different networks -- congestion, fault
tolerance, latency, communication bandwidth to name a very few.
The purpose of this special issue is to focus on the design,
simulation and analysis of all aspects of all kinds of routing
protocols to be used in network architectures.
Papers should emphasize new original contributions of either
theoretical or experimental nature. Papers must be
written in IJCS style. Please submit five copies of your
manuscript to the guest editor by December 15, 1996: Pradip
K. Srimani, Department of Computer Science, Colorado State
University, Ft. Collins, CO 80523, Tel: (970) 491-7097,
Fax: (970) 491-2466, Email: sri...@CS.ColoState.Edu.
Instructions for submitting papers: Papers should
not exceed 30 double spaced pages including figures, tables
etc. Papers should not have been previously published,
nor currently submitted elsewhere for publication. Papers
should include a title page containing
title, authors' names and affiliations, postal and email
addresses, telephone numbers and Fax numbers. Papers should
include a 300-word abstract. [Note: If you are willing
to review papers for this special issue, please send a note
with research interest to the guest editor.]
Publishers: Ablex Publishing Corporation, 355 Chestnut
Street, Norwood, New Jersey 07648.
peter honeyman
Jon Leech writes:
> Yes, exactly why we ended up with it. It is of significant benefit to
> administrators, not to users, and sysadmins, like anyone else, naturally
> tend to serve their own interests first when in a position to do so.
what a bizarre way of looking at it. afs' substantial sys admin
advantages translate into lower cost and better service for users.
peter