Office-Cluster "Cyphoning"

Robert G. Brown rgb at phy.duke.edu
Fri Oct 19 06:30:22 PDT 2001


On Fri, 19 Oct 2001, Chad Owens wrote:

> 
> Re: the WIN/LINUX mixed cluster idea

There is another way to proceed that we are trying to advance at this
point.  One can install linux on all the office systems, install vmware
on the linux boxes, run Win9X under vmware.  

Advantages:

  a) Win9X then becomes just another linux task.  It can be "shut down"
when not in use and the system is totally available for background linux
tasks, regulated by any of the usual tools.

  b) WinXX actually becomes easier to install and maintain -- it sees a
"virtual machine" with absolutely identical hardware, and one can
(re)install the WinXX as a direct disk image at any time.  One mounts
user space for WinXX from a linux-side server.

  c) Viruses become irrelevant.  If a system gets one, you just copy in
a new image of the system.

  d) Since the office staff are running linux, they will quickly learn
that they can use netscape on the linux side to e.g. browse the web or
read mail.  They will learn that there are spreadsheets, mp3 players,
word processors, and so forth that all work pretty well on linux, and
may actually TRY them (with a bit of prodding) and come to rely less on
WinXX and Office, especially if they primarily use just the web and
mail.

Disadvantages:

  a) You have to buy a vmware license for each system.  Not horribly
expensive (cheaper than the Windows you run under it, and the human
maintenance costs it might reduce could actually save money overall),
but definitely annoying.

  b) vmware is not open source and is kernel specific, so you end up not
being able to track kernel updates as aggressively as you might wish to.
The office dual win/lin systems will be a bit behind the curve in
update-based functionality and security.

  c) Certain tools, like mosix (which also is kernel specific) may or
may not collide with the set up.  You may find that your cluster
configuration is a bit more constrained than a pure linux cluster.

All of them basically derive from the commercial closed source nature of
vmware.  Tres drag.  Netraverse's win4lin is a possible alternative to
vmware but also costs money (about the same amount of money) and we
haven't played with it thus far.

We are experimenting with this in the physics department here primarily
to try to reduce the maintenance required by the handful of office
systems running WinXX (almost the entire department runs linux -- more
than 100 desktops -- and requires LESS work by far than the four or five
office systems running WinXX to maintain).  However, we've thought hard
about doing it in the physics labs (which run WinNT) to recover their
wasted cycles at night.  The alternative is to arrange a scheduled
shutdown/reboot -- linux by night, NT by day -- which would also work
well enough except when a TA needed access to e.g. Excel and rebooted a
system back into WinNT in the middle of the night, possibly breaking a
computation.

> 
> A) build a special Linux kernel stripped of everything user interface
> related(GRAPHICS/XWIN LIBS/INTERFACES/ETC), and equip it to interface
> with an OS varient of choice (W9x/ME, 2k/XP, U*X, even Linux)
> 
> B) Position it in the bootup process similar in function as the NT HAL
> layer (essentially a triangular HAL - NETWORK / HARDWARE / InstalledOS),
> yet providing a 'resource adapted' enviornment to the OS of choice, which
> would adapt the availble resources based upon the track record of the
> machine usage of the user of that individual machine, and provide
> resources to be used by the cluster/server
> 
> C) Ideally, the drivers would be loaded to the linux machine (for the
> hardware)
> 
> D) Which would in turn respond to the OS in a similar as PnP BIOS devices
> currently do (perhaps even creating the effect that non PnP devices could
> be PnP capable to the installed OS.)
> 
> E) Then the Linux machine could insert process jobs in slack CPU time,
> which can be artificially created on machines that aren't hyper fast, or
> bloat normal waits on hyper fast machines, either yielding to the OS, or
> to the cluster process job, based on time to complete, "nice" level, etc.
> So if the server gets bombarded, and the users of the machines are the
> ones doing it, the server can claim some of their machine horsepower to
> #1 - throttle them back a little(like using an undersized carbureator),
> and #2 - use the shared CPU time to process the overload...
> 
> F) This also benifits the resident IT department, allowing them to have
> network access when the InstalledOS is dead or infected, allowing ease of
> re-ghosting, etc. Many Many options available on this front.
> 
> G) Also, if someone goes on vacation, and their computer isn't used, well
> more resources for the cluster...
> 
> ****************************************
> 
> Just a rough idea that has many issues involved, but it's something
> that's been roling around in my head for many months now...just wanted to
> see if anyone else would think it a good idea...
> 
> Sincerely,
> 
> Ruddykins
> 
> 
> ________________________________________________________________________________
> Get your FREE download of MSN Explorer at http://explorer.msn.com
> _______________________________________________ Beowulf mailing list,
> Beowulf at beowulf.org To change your subscription (digest mode or
> unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
> 

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu







More information about the Beowulf mailing list