Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!wuarchive!udel!rochester!pt.cs.cmu.edu!rain.andrew.cmu.edu!ddean
From: ddean@rain.andrew.cmu.edu (Drew Dean)
Newsgroups: comp.unix.admin
Subject: Flame Fest (Was  Re: Project Athena ( was Re: Non Destructive Version of rm))
Message-ID: <12975@pt.cs.cmu.edu>
Date: 9 May 91 17:57:53 GMT
References: <12049@mentor.cc.purdue.edu>
Distribution: na
Organization: Carnegie Mellon University
Lines: 35

In article <12049@mentor.cc.purdue.edu> asg@sage.cc.purdue.edu (The Grand Master) writes:
>We also have it here at GE where each person who has a workstation
>can still log into anny workstation and be able to access his disk without
>having to do mounting all over the place. If I want to get to a directory
>/tmp on the system a294 I do cd //a294/tmp - no problem.
Uh, guys --
	From this description, I'd be willing to bet that GE has a network
of Apollos.  Apollo has a network root concept, where each machine has its
name as an entry in the // directory, which is above the local root directory.
Several daemons (try) to keep this information consistent across the network.
This appears to be a legacy of Aegis, Apollo's proprietary OS (before they
merged it with Unix).  Aegis may or may not be a "good" OS; that's not at
issue here.  It was built for distributed workstations, though, from the
start.  I'd note that CMU RFS (not related at all to AT&T RFS) uses
/.. in the same way.  In these types of remote filesystems, the mounts are
implicit instead of explicit; again whether or not this is a good idea isn't
relevant.
	Conclusion: this has nothing to do with NFS and/or automounters.
	To add more flamage: the Sequent's I've heard about are either
NS32x32 or i386 based machines, ie. each processor is < 10 MIPS (meaningless
indicators of performance :-)).  They also seem to cost > $100,000 (and get a
lot higher than that with 40 CPU's).  Considering that HP will sell you a
50+ MIPS workstation (uniprocessor) for ~ $25K (with disks, etc), I really
doubt that an equal dollar amount of Sequents will buy more usuable
performance than workstations.  Tell me this:  If I'm running a large
application that hasn't been parallelized (so I can only use 1 processor),
say Mathematica, which is going to be faster ?  [I don't think Mathematica
actually runs on either platform, so substitute the package of your choice.]
Note that this shouldn't be construed as large multi-processor bashing,
because those machine have there place, too (big databases come to mind).
Just in terms of providing MIPS to people, cheap workstations are hard to beat.
-- 
Drew Dean
Drew_Dean@rain.andrew.cmu.edu
[CMU provides my net connection; they don't necessarily agree with me.]