Slashdot Mirror
The Uncertain Promise of Utility Computing
icke writes "A quick overview of where the Economist thinks we are with the The Next Big Thing, also known as Stuff that doesn't work yet. Quoting: 'It is increasingly painful to watch Carly Fiorina, the boss of Hewlett-Packard (HP), as she tries to explain to yet another conference audience what her new grand vision of "adaptive" information technology is about. It has something to do with "Darwinian reference architectures", she suggests, and also with "modularising" and "integrating", as well as with lots of "enabling" and "processes". IBM, HP's arch rival, is trying even harder, with a marketing splurge for what it calls "on-demand computing". Microsoft's Bill Gates talks of "seamless computing". Other vendors prefer "ubiquitous", "autonomous" or "utility" computing. Forrester Research, a consultancy, likes "organic". Gartner, a rival, opts for "real-time". Clearly, something monumental must be going on in the world of computing for these technology titans simultaneously to discover something that is so profound and yet so hard to name.'"
All that they're flogging is this: lots of intercommunicating little computers in everything. We're already about halfway there -- between the XBox, Tivo, and KISS Technology's (GPL-violating) DVD player, *normal* people are more likely than ever to have a computer connected to their television without even knowing it.
to me, they sound like all the different Microsoft execs (Ballmer, Gates, etc) trying to answer the question "What is .NET?" I know there was a business2.com article that sampled some responses, but I cant seem to find it at the moment. IIRC, one quote was along the lines "So much of our stuff has a '.NET' label attached to it, even we dont know what it is at times."
The One Rule Of Chess You'll Ever Need: Don't play someone who carries a kit in their bookbag.
I'm fairly certain that used to be the case when *everybody* was running mainframe environments (not that lots of folks still aren't), but the key to the new version of this is that it'll be done over the network.
Look at it from IBM's perspective. You can have 8 extra processors on-site for each client for those few times when they need the extra CPU, or you can have massive datacenters all over the world with a pool of extra CPU's to draw from. The latter will lead to unprecedented economies of scale as you can reassign computrons dynamically between clients to whomever needs them most, while still maintaining a comfortable cushion. Those economies of scale likely mean both lower prices for the customers as well as increased profit for IBM, because it drastically increases the efficiency of their services.
I would be surprised if IBM was *not* working on a way to make applications portable across architectures also, and the push towards Linux on everything would seem to support this endeavour, irrespective of all the other reasons.
Imagine buying systems capabilities instead of machines. Let's say you need gobs of CPU but not so much I/O bandwidth. Your jobs are allocated to a Power-based compute node. Let's say you need gobs of I/O bandwidth but not so much CPU. Your jobs are allocated to a zSeries machine. Now things get *really* interesting when your job first needs lots of I/O, then lots of CPU, then settles down for a bit. Your job could get reallocated across the grid based on its needs at any given moment.
The technical end of making transfers of processes and datasets seamless is where the difficulty lays, and all of the 800lb gorillas are chomping at the bit to get it working first. The first one to do it right stands to make a fortune.
Dan