Buying Computing by the Computon

theodp writes "Seeking to emulate the pricing models utilities use to charge customers for kilowatt-hours of electricity based on the ebb and flow of power demand, HP Researchers have come up with a new unit-of-computing metric, the Computon, which is not to be confused with the 'Power Unit' and 'Service Unit' pricing metrics from Sun and IBM. California, here we come!"

Comments

I like my servers like I like my music

[the_2nd_coming]

I like to own them!!!

so basicly what HP is saying is that depending on how hard I work the servers will effect some monthly payment I make to them.

so, does this lower the cost of service contracts becasue companies that push their servers harder require more service than those who have low or moderate useage?

Re:I like my servers like I like my music

[EndlessNameless]

I like to own them!!!

RTFA. This applies to customers who outsource their IT to HP. If you actually own the hardware then this article doesn't directly affect you.

so, does this lower the cost of service contracts becasue companies that push their servers harder require more service than those who have low or moderate useage?

In theory, yes. However, this wouldn't be the first time that a company used obsfucation as a sly means of increasing its revenues.

i wonder

[iosmart]

i wonder how many computons it'll take to actually determine the number of computons all of your different users have consumed. :-D it sounds like a complicated process and it sounds as if it is geared towards systems that are used by large numbers of people...

FLOP?

[Vagary]

What's wrong with the good old flop? Or even simply instructions (of BogoMIP fame) or cycles? When you're dealing with the volume that grids (which is what this story is really about) will produce, you don't need a precise metric. And thanks to the Halting Problem, you'll be forced to buy the "computons" in even lots or risk losing computation time while transactions wait to clear.

The important thing here is that HP is putting forward the idea of computation as a commodity. I just wish some researchers could have published a journal article instead of letting the marketing dept. get their greasy paws all over it.

Re:Is that Something like MIPS

[jat850]

I don't think so, but I'm not positive. While MIPS is a (one of many) metric of processor speed, I think the computon is more a measurement of the resources used by a corporation for pricing guidelines.

Re:Compare with computron

[Vagary]

IANAP, but I wouldn't call the proposed correspondence between negentropy and information (as defined by Shannon) "pseudo-science". It's precisely this sort of cross-disciplinary metaphor which is so highly valued within Mathematics.

Liquid things do have higher levels of entropy than solid things. And computers do get hot because they're determining where electrons are and then forgetting that information (to use Norrestranders' (in The User Illusion) way of putting things).* If you could constrain the molecules in an object, it would be colder. The factory comment, though, is part of a strawman argument.

* I seem to recall reading something on /. years ago about computing that recycles the contents of registers to lower waste heat. Am I on drugs?

Re:Is that Something like MIPS

[plover]

MIPS == "Meaningless Indication of Processor Speed."

Unless, of course, you wish to believe the old folks who might otherwise tell you it stands for Millions of Instructions Per Second. Back in the good-old-days, before the current abundant crop of benchmarks, people tended to measure CPUs more simply. You used to hear arguments of "my RISC chip performs more cycles per second than your CISC chip" or "my CISC chip performs more work per cycle than your RISC chip." (Anyone else notice the passing of CISC and RISC from the lexicon?)

Personally, I think we need a unit to measure the accelerating number of benchmarks created every year. How about the Shtonestone?

And to stay on topic (not that it matters much on Slashdot), the name has been "computron" for years and years.

Re:Is that Something like MIPS

[Black Copter Control]

At the University Of Alberta (back in the 70's ~ 80's they charged based on things like VM/CPU integral ( 1Min CPU time * 4 Meg of Virtual memory) and page-months of disk storage.

This sounds like a good bit more complicated, and could lead to rather wierd results like you end up paying more for CPU time because you had to wait longer to get the computation done (it was a high-load period).

The cheapest CPU time on MTS were 'deferred priority' batch jobs. They generally only got ran on weekends and after midnight (when nobody else was using the system). If you were working at those times it was much worth your time to do expensive compiles etc. as deferred batch jobs (if the queue was empty, they'd run pretty quick, and you'd get charged about 1/10 as much as doing it from your terminal).

Re:How novel an idea

[LinuxHam]

Currently IBM gives you a server that can handle up to Z work but only charges you for X, where X is the amount you actually use and Z is large enough that you can handle significant spikes.

I am currently working on one of IBM's biggest On Demand deals. You seem to know enough about IBM's On Demand initiative to know that it's not the same as HP's angle.. at all. The article specifically mentions HP hosting the equipment, as opposed to the IBM way, which is to basically perform a server consolidation onto IBM hardware hosted at the customer's site bugged to report back to IBM utilization numbers.

There are a couple more huge deals about to go public that I can't mention here, but suffice it to say, HP hasn't won, nor can they. IBM is way out front in this arena. We just kicked HP out of a place they "owned" for many years, and replaced the equipment with IOD gear on 1/10th of the number of source servers. The p690 AIX and x440 Intel servers completely rock.

Re:Compare with computron

[William Tanksley]

I seem to recall reading something on /. years ago about computing that recycles the contents of registers to lower waste heat.

Almost. Reversible computing builds all of its primitives to prevent losing information -- evidently, this directly causes it to produce less heat. See Baker's papers for more information.

Am I on drugs?

Um... Hard to say. Perhaps this would make a good Ask Slashdot question? :-)

-Billy

Only me?

[pla]

Seeking to emulate the pricing models utilities use to charge customers for kilowatt-hours of electricity based on the ebb and flow of power demand

A question for my fellow /.'ers...

When I read things like this, I feel very, very unhappy. I have a PC that does what I want, when I want it, and I don't pay any additional fees to use its capabilities. I don't pay more if I actually fill my HDD vs leaving it nearly empty. I don't pay more if I leave my CPU 99.9% idle compared to running three distributed clients just to keep every single cycle busy doing "real" work.

I feel similarly about the software I use. I have an OS and a few apps, and I don't pay more when I actually use them compared with leaving them sitting uselessly on the disk. I don't pay for each image I Photoshop, I don't pay for each program invoked by the OS, I don't even pay every time I decide to surf the web.

Even media files, I don't pay-per-view. If I queue up a bunch of Vorbis files, I don't pay every time I listen. Nor do I pay for watching a DVD I own.

Yet, companies keep trying to move their business models to "buy once, pay forever". I can see the obvious benefits to the company, but it has NO benefit to the consumers.

So to get to my actual question... Does anyone see even the faintest bit of logic behind these companies moving toward pay-per-use schemes? Not logic like "we'll make more money if we get enough suckers", obviously, but some real sensible reason why people might prefer to abandon any concept of "owning" the things they use daily, rather than paying continuously for "access" to them?

Re:Destined for failure

[dprice]

I agree that in its current form, this on-demand computing model will probably fail. Utility computing is mostly a vision put together by a bunch of executives that believe they can build a revenue stream analogous to power companies. They push the vision through the marketing department who make the vision all glossy and ready to sell to other companies. Research and Development teams are then instructed to come up with an implementation of the utility computing vision. From that directive you get the research guys writing papers about "computons". The rest of what gets developed looks pretty much like the bad-old-days of timeshare computer systems (charged by time and storage usage) from over 20 years ago. So far I have not seen any revolutionary technologies for utility computing, just rehashes of old ideas on modern hardware. It is still users connecting through a network to a bunch of servers with some software to account for usage. In the end, utility computing is just a name for bunch of people in suits selling IT services to a bunch of people in suits.

Re:Coding Revolution

[Jerf]

What on earth are you talking about? "Clean code" and "optimized code" are opposing forces!

"Premature optimization is the root of all evil."

The cleanest code will also be some of the most inefficient. (Note it does not work the other way around, of course, so read that sentence carefully before criticizing it!) The most efficient code will be pug-fugly, incredibly difficult to read without intimate knowlege of the whole system code, and will be very difficult to correctly change to boot.

Done much programming when efficiency was a concern? Doesn't sound like it.

The freedom from the need to worry so much about efficiency is one of the best things that has ever happened to any engineering discipline. It can be taken too far but the answer is better balance, not to fly off in the opposite direction.