NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/comp/unix/admin/6642

Text File | 1992-12-14 | 3.4 KB | 77 lines

Newsgroups: comp.unix.admin
Path: sparky!uunet!cis.ohio-state.edu!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!haven.umd.edu!decuac!pa.dec.com!pobox.cca.cr.rockwell.com!j_detter
From: [e-mail]
Message-ID: <[e-mail]>
Subject: diskless vs dataless vs datafull
Date: Mon, 14 Dec 92 16:44:21 -0600
X-Received: by usenet.pa.dec.com; id AA12560; Mon, 14 Dec 92 14:44:57 -0800
X-Received: by inet-gw-2.pa.dec.com; id AA14177; Mon, 14 Dec 92 14:44:44 -0800
X-Received: by bulls.cca.cr.rockwell.com (5.57/Ultrix3.0-C)
    id AA03400; Mon, 14 Dec 92 16:44:22 -0600
X-To: comp.unix.admin.usenet
X-Mts: smtp
Lines: 62

I manage a site of around 250 misc Unix boxes mostly Suns and
DECstations. We have designed our network so that the OS and
the users home directory reside on their local workstation and
their applications reside on a server.

The users at my site do primarily ECAD and MCAD. We run Unigraphics
for MCAD and Racal/Redac for ECAD. This means that we have a lot of
engineers and draftsman that require that the nodes be up almost
100 percent of the time.

Our method appears to be radically different than what everyone else
is doing and has our management worried.

They are now asking for data backing up why we are doing what we are
doing. The probability of failure in each approach, cost per seat in each
approach, cost of downtime in general (or at least a good way to
calculate it). etc ..

The main reason that we designed things the way we did was to reduce
network traffic and network dependency. In the evnet of a network
failure most users can continue to work on what they were working on.
It seems to me that with the other approaces users would be dead in
the water until the server came back.

The network that we reside on is heavily used and there is a good
possibility that the management portion of the network would be taken
out of Engineering where it currently is and moved to IS.

Has anyone else done any studies on the reliabilty/cost/manageability
of the three approaches.

The management of the / partition to me seems to be the same on all three.

We are currently planning to implement an SCCS/rdist based procedure for
maintaining the misc configuration files across the network.

I maintain that by keeping the OS installs indenitcal on all machines and
using a source control system for the configuration files that the datafull
approach would not cost anymore to administor. The only real cost would
be in the price of the disk that houses the /usr partition.

I maintain that the only real difference is in the time to load and
manage the /usr partition on each client.

I don't claim to be a Unix guru but this has worked fine for us so far.

We have had DEC try and force feed Athena to us, however I do not feel
that this is a good fit for our site either.


Any information on how other sites do things, how they would do things
if they could do it again and/or any real data that favors one methodolgy
or the other would greatly be appreciated.

 ______________________________________________________________________
| Jeff Detterman [phone]      Rockwell International          |
| Unix Network Administrator              Collins Commercail Avionics Div |
| [e-mail]    Cedar Rapids, Iowa 52498        |
 -----------------------------------------------------------------------