Re: [help] 1 cpu to rule them all

From: Juhan Leemet (juhan_at_logicognosis.com)
Date: 07/12/04 

Date: Mon, 12 Jul 2004 05:55:28 -0200

On Mon, 12 Jul 2004 00:31:20 +0200, David Wright wrote:
> Desi Cortez wrote:
>> Timberwoof wrote:
>>> Time was when that was a viable way to do things, but nowadays CPUs
>>> really are cheap enough that every user can have his own...
>>
>> I don't understand how this 4-user HP machine can be so 'cheap'...
>
> It depends on your definition of cheap... Maintaining 100 central
> application servers in an air-conditioned computer room and keeping the
> configuration and maintenance in one place is a lot more economical than
> having 1000 PC's dispersed around an office block...

How are the users going to access these application servers? They will
need a small/cheap thin client, i.e. a PC (or workstation?) which is the
cheapest commodity access point for network computing. You're not going
to run long cables for screens and/or keyboards. The only thing that makes
sense to snake around a building is ethernet (these days).

I'm in favor of putting shared services (home directories, shared
resources, special purpose computing devices) in a "glass house". However,
I cannot see any advantage in a multi-screen multi-user cluster machine.
We used to do that kind of thing 20 years ago, back in mainframe or
minicomputer days, when CPU was expensive and HAD to be shared. Nowadays
CPU is dirt cheap. PCs are too. Even workstations are not that expensive.

> If you cut the number of physical machines you have, you cut the number
> of support staff you need to maintain them as well...

You can make things easier by having lots of machines that are virtually
identical for "workstations". Put all the customized stuff (e.g. home
directories) on servers. Lock down the workstations. In an office, no one
needs "root" for their own workstation. Then they cannot screw things up.
They can still personalize their own setup, which is in their home dir. If
they screw that up, they get to buy the sysadmin lunch! 8^)

[cut a whole bunch or arguing... I don't really agree with it]
> With a site of 500 users with 6 mini computers, we had a helpdesk of 2
> people. With 500 PC's, we had 10-15 support staff running around the
> building. Plus the hardware maintenance costs go up exponentially with
> more PC's.

I think you could manage your PCs better. What O/S? Did users "tinker"?
The only reason someone should have to "run around" is to do a new
installation or replace a busted PC with another complete working PC. You
might have classes of users requiring bigger/smaller PCs, but you
shouldn't have 1000 totally different configuration PCs.

Ten years ago, I had a client who ran a banking type application on 500
OS/2 PCs with about 50 OS/2 servers scattered across Canada with a help
desk of 4 people in Toronto. Some additional application developers were
2nd tier support. They must have had a couple of "hardware guys" but no
one was "running around" (across Canada?!?). We did diagnostics and
software updates (including versions of O/S components, such as Comms
Mgr) "across the wire". It took some special scripting and a small test
lab to make absolutely sure that the update package was good and reliable.
Never lost a machine as "vegetable on a wire" while I was there (1+ years
of updates!). The next crowd got cocky/slack and "blew up" a server tho.

I believe one could do the same thing with Solaris or Linux. Dunno (and
don't care) about Windoze. I don't think we need bizarre H/W architectures. 

-- 
Juhan Leemet
Logicognosis, Inc.