[Beowulf] Application Deployment
james.p.lux at jpl.nasa.gov
Sun Oct 10 10:26:50 PDT 2004
----- Original Message -----
From: "Mark Hahn" <hahn at physics.mcmaster.ca>
To: <beowulf at beowulf.org>
Sent: Sunday, October 10, 2004 8:55 AM
Subject: Re: [Beowulf] Application Deployment
> > require training, expertise, etc. Sure, for Linux, the actual software
> > free, but that's a small fraction of the $100K/yr you're paying to folks
> > use the software.
> very interesting. one structural disadvantage that the windows ecosystem
> does labor under is that it must stick to the
> model. that is, msft is not quite ready to go to a ephemeral-client
> where desktops just PXE-boot and mount everything of consequence across
> the net. (not just thin-client, where clients are all hard-installed, but
> use only a thin app like a browser for whatever the user needs.)
I'm not sure, but MS is certainly heading towards the ephemeral client (with
local cacheing) model, since it enables such things as revenue based on a
per use of a component basis. Say you're a small software developer and
you've developed a really nifty piechart algorithm for Excel. MS wants to
give you a way that you could generate revenue from each use of this
component, and that sort of implies that the component is fetched from some
repository on the fly.
Same for "notepad" or you name it. I think they're desperately trying to
get away from the "transfer of tangible property" for software, because
sooner or later, shrink wrap licenses are going to get hammered in court (if
it looks, walks, and talks like a sale, then it IS a sale, and you should be
able to resell, etc., freely). On the other hand, if each and every time
you use the component (be it "MS Word", that clever Excel chart, etc.) you
are separately engaging in a revenue transaction, then you don't get into
those sticky areas.
> with lan-ipmi and pxe, it's almost reasonable to claim that support
> doesn't scale with increasing nodes. there are still costs that scale
> with number of users, number of apps. hardware maintenance always
> scales with number of moving parts, but for ephemeral clients, it's
> far easier to have spares. the infrastructure to support 1K clients
> all booting monday morning would be nontrivial, but very tractable.
> no doubt the lack of nazi DRM (uncontrolled and dangerous network!)
> is why the msft community hasn't taken this approach.
MS, and the legions of developers who develop for the Windows environment,
generally want some mechanism to be paid for their work. Per use revenue is
a nice way of getting around the "bootleg copy" problem. Who cares if you
copy it, if every time it runs, you have to hit a license server and pay
your little micropayment. In fact, bootlegs are great... they cost the
originator of the software nothing. It's the whole compatibility,
configuration managment thing that is a big problem (all those components
have to be compatible with all the other components, etc.).
MS, for all of their faults, doesn't have stupid people working for it. If
they could find a better way to "sell" software (or, more properly, the
added value provided by the software/content/what-have-you) that doesn't
rely on copyright (which everyone admits is poorly suited to such things),
they'd love it.
More information about the Beowulf