Very Computer

->Telnet to fix things is good, but the setup should be stable enough and
->secure enough that the administrator doesn't have to fix things.
->Administration should only involve software ipgrades and additions.  A
->central server for applications will allow quick upgrades to applications
->with minimal intervention.  One thing I would like to kno is whether there is
->a way an administrator can set off a script on his admin machine and have it
->apply a security patch to all machines?  Say maybe a script to apply the
->patch in a shared directory, and then another that started on the admin
->machine logs into the others and runs the patch script.

Here's an easy way to do it:

 o Choose a distribution which will be the standard distribution for
   all of your machines.  In this example, let's say that the
   distribution is redhat.

 o Have a public, anonymous ftp server for your distributions with
   some sort of directory structure like this:

   /linux/distribution_name/version/architecture/[dist files go here]
   /linux/distribution_name/version/architecture/updates/[updates go here]

   So, in the redhat/intel case we could have:

   /linux/redhat/5.2/i386/updates

 o Setup a text file which has each machine's name or ip address and
   their distribution type and version.

 o Be sure that each machine has some sort of remote shell installed.
   You can use rsh if security is not important, or you can use ssh to
   guarantee that the correct host is always used and that the
   sessions are encypted.  They have the same syntax, so it shouldn't
   be difficult to go from one to the other.

 o Use a simple script to go through your text file, sending the
   appropriate command to each client.  Here's a perl example:

     [snip]

   $URL = "ftp://$FTP_SERVER/linux/$ASKED_FOR_TYPE/";
   while ($line = <TEXT_FILE>) {
      ($machine, $type, $arch) = split(/\s/, chop($line));
      if ($type eq $ASKED_FOR_TYPE) {

                    $URL . "/$arch/updates/$PACKAGE";
         print LOG "$MACHINE - $PACKAGE:" . `$command` . "\n";
      }
   }

 o You could then simply update all machines by typing:

   update_machines redhat sysklogd-1.3-26.rpm

   and it would update all redhat machines to that level of that
   package.  It wouldn't matter if they were intel, sparc or
   whatever.

Another option is to use rdist, and you would probably need to do that
if you had multiple distributions in use.

--
The wheel is turning but the hamster is dead.

Quote:

> I think Linux is ready for the corporate desktop.  Maybe not the
> consumer desktop, but the business desktop.  So, here is the challenge.
> We have a hypothetical business.  Say a reseller or something that uses
> a database to keep track of orders.  This business has about 30
> employees with desktops, a couple IT people, and some servers as
> needed.  We will assume the database application used to keep track of
> orders and sales is very portable (unrealistic I know, but this is for
> fun).  All the users need on their desktops are wordproccessor,
> spreadsheet, presentation, contact and scheduling software, GUI access
> to the database application, and Internet access.

> The challenge...
> Describe how to do this with Linux such that:
> 1) The users can get to their apps and do their work w/o the command
> line.

Quote:> 2) The administration is as easy as possible in the long run. Including
> installation and setup. Extra time initially to save time later is good
> though. (hint scripts hint) :-)

Quote:> 3) The user can't*up things such that the adminstrator has to fix
> problems like:
>      a) The word processor isn't in my menu anymore.

Quote:>      b) I get this a black screen with a blinking dot and can't do
> anything.

Quote:>      c) Any others I can't think of.
> 4) Let's try to get specific, as in what apps, how many servers,
> minimize # of IT people, costs, can we upgrade everyone remotely with
> scripts, etc.

Can't think of anything else. Using diskless terminals and keeping all
the servers together in one room should keep the number of IT people
down. If you need to connect with a fat client, there's always ssh or
telnet.

mawa
--

My Geek Code is no longer in my .signature. It's available on e-mail
request. It's sad to live in a world where knowing how to programme
your VCR actually lowers your social status...

Quote:>  Right!  that is what I am trying to make the point.  The next step is to
>compile the ideas together and see if I can come up with a couple of examples
>that will work in the real world.  So far it looks like two general configs.

>1) Setup workstations as X-terminals, and then run everything on a central
>server.
>    Pros:  Easy to administer.
>    Cons:  Loads up a the central server.

>2) Setup workstations with XDM, mount everything remotely.
>    Pros: Each workstation does its own work.
>    Cons: A little extra administration is required.

I therefore opt for minimising the admin workload, because either (a) it
will be outsourced, or (b) there'll be one guy who wears another hat and
has a lot of other things to do at the same time.

The X-term/workstation split is a *e to call. If you've only got
three or four client boxes, it may be the easiest route; if you've got
thirty, you're probably hammering that server quite hard, and may need
a couple of 'em.

Maybe for larger systems, independent workstations (KDE/StarOffice
installed locally, running KDM for logins, using NIS for passwords,
mounting /home via NFS so all user files are stored on a server) are a
better bet. But this ups the administration workload -- thirty sets of
logfiles to check, thirty machines to keep track of security loopholes
on, and so forth. No problem if you've got technically literate users
(snort!), not a major problem if you've got a full-strength IT support
department, but probably not a good solution in the "real world" (where
companies have neither).

Remember, this is the problem the NC was intended to address. So far,
it looks like Linux is a better answer than Windows (in any form), due
to Windows' long-term problem with maintaining user context; but it
still ain't perfect.

-- Charlie

-- Major Linux distribution
-- Two or three bad-ass servers
-- Thirty diskless zero-admin X terminals at <$1000 each

I have an issue with the X terminals. It just seems that in today's inexpensive
high performance computing environment, that this setup would be better served
by having the apps local to each machine. When it's possible to put together
a 333 Mhz ($95), 128 MB RAM ($150), 4.3 Gig HD ($120), 4 MB video card ($30),
100 Mbit Ethernet ($30), and a 17 Inch Monitor ($200) for less that $650 with
each desktop capable of running all the apps the user would need, why go with
X terminals. The scalability of the pure client/server metaphor sucks.

And with Linux the administation isn't that difficult either. Simply install
all the applications on a global file server which each of the clients mount
remotely. So the same with the home directories. While I'm personally not a
fan of YP (I used to have stability issues) it it perfectly capable of
distributing user information via the network. The local disk is used for
booting, the core system, and swap. All the apps and user directories are
mounted remotely.

You only need one high performance file server which should be able to easily
handle the load even if you add another 30 clients. I'm pretty sure the same
could not be said of the X terminals.

Just a thought.

BAJ

OK, performance of your proposed workstations will be less than half
that of the same boxes running NT or Win9X workstations. Linux has
pitiful desktop performance. This becomes painfully apparent, when
running the huge apps that you mention (Star Office 5.0 under KDE). If
any of the responders in this thread were actually running this
configuration on both platforms, they'd know that for themselves. The
identical Windows software or SO 5.0, outperforms the same software on
linux by well over 2:1. Add to that, X's poor font support (yes even
with xttf) still looks horrid, next to windows anti aliasing fonts and
speedier 16/ 24 bpp rendering.

IMO here's a better solution. Linux makes a great server but the
desktop performance and sophistication is ~3-5 years behind the
competition. I don't know why someone has not suggested REAL thin
clients, or running remote X software on cheap OEM wintel (233/64meg)
boxes and make one to three fast linux servers do the grunt work. You
can run just about any WM and StarOffice remotely with good
performance. Using a base100 network,  not base10 will maximize the
Xserver and file access performance. This is actually a cheaper and
IMO better solution. Capitalize on the strengths of both platforms.

BTW, SO 5.0 or applixware may be free as a personal edition but your
going to have to pay for it in the commercial setting, you've
proposed. The OEM windows boxes should not incur any significant costs
beyond  initial hardware expenses and maintenance, as you'll be using
linux/vnc (or other remote X software) & samba for your  servers and
not NT.

Once Vnc has initialized on the server, it's very fast in opening &
retrieving files, plus considering that multiple remotes running the
same software are being used. It should be more a server cached memory
issue. So performance can actually be better than you're proposed and
more complex linux stand alone server/client, office complex. You
COULD run remote linux clients as well, but I believe both the
performance and appearance will be better and won't cost anymore
(wintel software is OEM on hardware), by mixing the server client
base, as I've suggested.

In fact, I just tried this with two clients on my home base 100base
system and once past the initial opening of the 'first' StarOffice,
either vnc clients can be closed, then reopened, almost instantly
(within 2 seconds). My linux hardware is only K6-233 and 64 megs of
ram. So as you can see below, considerable but not excessive swapping
is involved. With Samba, the user files can be stored locally on the
wintel boxes. A real server, might consist of linux running dual P6's
400's and 256Megs of ram.

 10:42:55 up 4 days, 19:44,  2 users,  load average: 0.00, 0.09, 0.10
79 processes: 78 sleeping, 1 running, 0 zombie, 0 stopped
CPU states:  0.1% user,  1.3% system,  0.0% nice, 98.5% idle
Mem:   63136K av,  61912K used,   1224K free,  45648K shrd,    124K
buff
Swap: 104416K av,  37484K used,  66932K free                 23676K
cached

Just my 2bits more.

http://members.home.net/bigrex/

Quote:>I have an issue with the X terminals. It just seems that in today's
>inexpensive high performance computing environment, that this setup would be
>better served by having the apps local to each machine. When it's possible
>to put together a 333 Mhz ($95), 128 MB RAM ($150), 4.3 Gig HD ($120), 4 MB
>video card ($30), 100 Mbit Ethernet ($30), and a 17 Inch Monitor ($200) for
>less that $650 with each desktop capable of running all the apps the user
>would need, why go with X terminals. The scalability of the pure
>client/server metaphor sucks.

Quote:>And with Linux the administation isn't that difficult either. Simply install
>all the applications on a global file server which each of the clients mount
>remotely. So the same with the home directories. While I'm personally not a
>fan of YP (I used to have stability issues) it it perfectly capable of
>distributing user information via the network. The local disk is used for
>booting, the core system, and swap. All the apps and user directories are
>mounted remotely.

>You only need one high performance file server which should be able to easily
>handle the load even if you add another 30 clients. I'm pretty sure the same
>could not be said of the X terminals.

Mind you, if 10 people are running the same program (say, Applix) on the
server, that may be a better idea as it means lots of memory image gets
shared between users, thus reducing memory consumption.  Rather than needing
12MB of RAM on each workstation to hold Applix binaries, you need just 12MB
on the server.  Ten simultaneous users results in a savings of 108MB of RAM.
(I'm merely guesstimating about that 12MB figure...  It could be 16MB, or
6MB...  The principle still holds true...)

There are no completely unambiguous answers to this...

--
Those who do not understand Unix are condemned to reinvent it, poorly.  
-- Henry Spencer          <http://www.hex.net/~cbbrowne/lsf.html>

While I'll admit they may not be good monitors, simply click on the monitors
link at http://www.pricewatch.com. The refurbished 17 inchers actually start
at $150. But for good measure I picked a price point that reflected a somewhat
decent new 17 inch monitor.

- aside from that,
-your point is good.  An X-Term, sans monitor, should be able to cost on the
-order of $400, as you drop off the disk drive, cut down on RAM, and can go
-with a cheaper CPU.  Alternatively, run these sorts of systems as diskless
-Linux boxes, which obviously wind up with lots of cacheing, and you still
-wind up with cheap boxes on the desktop.

But I don't think price was necessarily the issue. The original poster of the
X terminal suggestion was looking at a $1000 price point.

My point is that it's better to distribute the computation load by bringing
that computation closer to the user. It certainly won't cost more to do and
the scalability is phenominal.

-
->And with Linux the administation isn't that difficult either. Simply install
->all the applications on a global file server which each of the clients mount
->remotely. So the same with the home directories. While I'm personally not a
->fan of YP (I used to have stability issues) it it perfectly capable of
->distributing user information via the network. The local disk is used for
->booting, the core system, and swap. All the apps and user directories are
->mounted remotely.
->
->You only need one high performance file server which should be able to easily
->handle the load even if you add another 30 clients. I'm pretty sure the same
->could not be said of the X terminals.
-
-It is indeed quite nice if you can push some of the processing load to the
-desktop; that cuts down on both the need for RAM and CPU resources on the
-server.

And more importantly the necessity of adding RAM and or CPU to the server
when you add clients, which you're going to have to have anyway.

-
-Mind you, if 10 people are running the same program (say, Applix) on the
-server, that may be a better idea as it means lots of memory image gets
-shared between users, thus reducing memory consumption.  Rather than needing
-12MB of RAM on each workstation to hold Applix binaries, you need just 12MB
-on the server.  Ten simultaneous users results in a savings of 108MB of RAM.
-(I'm merely guesstimating about that 12MB figure...  It could be 16MB, or
-6MB...  The principle still holds true...)

True on the memory. But the counter is how much load will it be on the server
to serve those 10 users? What's the variability of service when you have 2
users using the system as opposed to 20? By pushing the application out to
the client, even at the cost of extra RAM, you can create a system that
operates consistently no matter how many users are using it.

-
-There are no completely unambiguous answers to this...

It's a matter of tradeoffs certainly. But my first guess is that users will
happier with a system that's consistently fast on their desk instead of
one where it's slow one minute, and faster the next. Just a guess though...

BAJ

Don

<snipped excellent points.>

Quote:> -- Charlie

Well, my goal was all Linux.  Not a mixture.   Although, there was suggested diskless X terminals
running XDM.

Quote:

> IMO here's a better solution. Linux makes a great server but the
> desktop performance and sophistication is ~3-5 years behind the
> competition. I don't know why someone has not suggested REAL thin
> clients, or running remote X software on cheap OEM wintel (233/64meg)
> boxes and make one to three fast linux servers do the grunt work. You
> can run just about any WM and StarOffice remotely with good
> performance. Using a base100 network,  not base10 will maximize the
> Xserver and file access performance. This is actually a cheaper and
> IMO better solution. Capitalize on the strengths of both platforms.

> BTW, SO 5.0 or applixware may be free as a personal edition but your
> going to have to pay for it in the commercial setting, you've
> proposed. The OEM windows boxes should not incur any significant costs
> beyond  initial hardware expenses and maintenance, as you'll be using
> linux/vnc (or other remote X software) & samba for your  servers and
> not NT.