Kuro5hin.org: technology and culture, from the trenches
create account | help/FAQ | contact | links | search | IRC | site news
[ Everything | Diaries | Technology | Science | Culture | Politics | Media | News | Internet | Op-Ed | Fiction | Meta | MLP ]
We need your support: buy an ad | premium membership

[P]
Distributed Computing and Cash

By dieman in News
Tue Jul 11, 2000 at 03:49:26 AM EST
Tags: Technology (all tags)
Technology

Why does everyone feel that distributed computing will gain you cash? A recent CNN article and discussion on The Other Site seems to think so. What have you seen out there that makes you think that distributed computing (as in the massive parallel computing projects with tons of distant machines) as a profitable venture? And, what do you think makes it unfeasable for the future or even today? But, are the possible public gains worth much more than what any one of us could make from a corp for running a distributed application?


Sponsors

Voxel dot net
o Managed Hosting
o VoxCAST Content Delivery
o Raw Infrastructure

Login

Related Links
o recent CNN article
o discussion on The Other Site
o Also by dieman


Display: Sort:
Distributed Computing and Cash | 16 comments (13 topical, 3 editorial, 0 hidden)
CPU cyles are cheaper than the administrative over (3.60 / 9) (#1)
by freakazoid on Mon Jul 10, 2000 at 06:09:54 PM EST

I just don't believe this will ever work. With CPUs getting cheaper every day, it's hard to believe that the overhead of actually trying to track contributions of clock cycles and disburse payments will ever be low enough that it would be cheaper for a company to pay you and then resell your CPU time than it would be for them to build their own Beowulf cluster and lease out time on that. People are messy things to deal with. On the other hand, I think it makes a lot of sense for large-scale volunteer projects like SETI@Home and the distributed.net stuff, since they're not trying to make money on it or resell the CPU time.

Re: CPU cyles are cheaper than the administrative (4.00 / 1) (#8)
by joshv on Tue Jul 11, 2000 at 09:54:42 AM EST

So what if CPU cycles are getting cheaper every day, this will just expand the scope of what can be accomplished with a distributed network of computers.

As for pricing, how much do you have to be payed to give something that is essentially free to yourself (your already wasted spare CPU cycle) From the point of view of the CPU cycle re-seller, just enough to make you download their client over another's.

Millions of computers on the Internet will always be able to complete certain types of problems much faster and more cheaply than commercially available supercomputers. The problem is that these CPU cycles are not easily assessible to people who are perfectly willing to pay for them. The infrastructure just needs to be built to enable the tracking of task, cycles and payment - once that is done it can be a very economically viable model.

-josh

[ Parent ]
Re: CPU cyles are cheaper than the administrative (4.50 / 2) (#14)
by um... lucas on Tue Jul 11, 2000 at 02:35:19 PM EST

I think just the opposite:

With connection speeds getting faster and the average home/office computer getting faster, for some applications, it would make solid sense to buy other people's spare cycles rather than buy a machine to perform your computations for you. Last time I looked, Origins, Crays, RS6000's and E10000's were pretty expensive machines. They also need adminstrators, electrictiy, space, cooling, security, etc.

If you went with a distributed approach, not only would you not need to buy a multi-million dollar machine, you wouldn't need to pay for any of the other overhead. If people want to participate, they'll need to take care of all of those other costs themselves.

As for figuring out how much to pay... I don't think it'll ever need to be by CYCLE. But rather by block, a la distributed.net/SETI. Determine a base amount of time a block should take on a machine and determine how much that is worth. (say you've got a P600 and it takes 24 hours to complete a block. Blocks are worth $2/a piece, so for every P600, you'll expect to pay $48/month. Alpha 600's on the other hand complete blocks every 6 hours, so at that rate for each alpha 600 you'll expect to pay around $192 a month). You then have to factor in Moore's law and slice the amount each block is worth to you on a regular basis.

The system for tracking work blocks is already in place with Dist.net and SETI. Adding an accounting app shouldn't be much effort, so you come in MUCH cheaper of an initial investment than buying a supercomputer of your own, and though you have probably slightly higher operating expenses month to month, you don't need to worry about hardware upgrades, system failures (because blocks can be redistributed across nodes if any go offline for an extended period), etc...

It's just that someone needs to figure out other uses for the tech other than the 2 previous examples... What's worth money to people?

[ Parent ]
Buckazoids rule! (2.50 / 2) (#5)
by HiQ on Tue Jul 11, 2000 at 05:32:47 AM EST

First the world, and now the internet starts falling for the big buckazoid$; everything was, is and ever will be dominated by money. Quite sickening IMHO.

However, one of the not mentioned problems with distributed computing is the planning aspect. If you are running a volunteer type of job, than end dates are not too importents. This is however not the case with the commercial guys and their inevitable sponsors. They want to make sure they get worth for their money and often insist on a fixed delivery date. If you hand over the execution of the computing to people you don't know and who are not 100% reliable, how can you set a fixed end date for your projects??
How to make a sig
without having an idea
just made a HiQ

Re: End dates (3.00 / 1) (#6)
by guppie on Tue Jul 11, 2000 at 08:42:17 AM EST

I can't see how the planning would be any more difficult than with a "normal" IT project. I mean, at my workplace, I don't know half of the people, I don't trust them 100%, and I'm still expected to estimate end dates for projects.
Our golden rule is to calculate how long it would take with normal progress, and then multiply it by Pi...

What? The land of the free? Whoever told you that is your enemy.
-Zack de la Rocha
[ Parent ]
Re: End dates (3.00 / 1) (#7)
by HiQ on Tue Jul 11, 2000 at 09:10:26 AM EST

I don't agree with that: in a 'normal' IT project (if there ever is one), the projectmanagers ought to know the workers involved, check on progress and act when things turn sour. In a distributed computing environment there's no obligation for the participants to 'deliver'. For instance, my computer at work is calculating Mersenne primes for PrimeNet, whenever I'm not working hard enough (boy, do I make a contribution or what ;).
However, when I'm on holiday, or leave the office for a meeting in another town, I turn of my computer, resulting in a delay in the delivery of the endresult of the prime I'm currently calculating. Now in the Mersenne project, there is no end goal, they just search for new primes. In other (commercial) calculations I can imagine that there are end goals, which have to be reached at a certain date!
How to make a sig
without having an idea
just made a HiQ
[ Parent ]
Re: End dates (5.00 / 1) (#10)
by joshv on Tue Jul 11, 2000 at 10:02:53 AM EST

Come on, this is just not that hard.

The 'CPU cycle reseller' will know how many clients he has running out there, and the average number of CPU cycles each client provides each day.

The CPU-reseller will have to have a great deal of control over which tasks or projects are running on your PC, and their relative priority, but if they have this, and have a general idea of the total CPU cycles required for the completion of a new project, they can give a very good estimate of completion dates.

-josh

[ Parent ]
Re: End dates (4.00 / 1) (#9)
by joshv on Tue Jul 11, 2000 at 09:57:23 AM EST

Our golden rule is to calculate how long it would take with normal progress, and then multiply it by Pi...

With a slightly more skilled staff you might be able to get away with 'e'.

-josh

[ Parent ]

Banner Ads (3.00 / 1) (#11)
by Neuromancer on Tue Jul 11, 2000 at 11:39:00 AM EST

It's kind of like banner ads. They're only paying a little, but it add's up. For them it is obviously profitable, since they will surely take the lions share of profit.

The silicon is not the only cost of CPU cycles. (4.00 / 1) (#12)
by mahlen on Tue Jul 11, 2000 at 11:56:01 AM EST

I don't know if the scheduling and privacy issues will be worked out anytime soon. But i can see some solid fiscal reasons why a company would want to use distributed cycles, rather than buy the computers for themselves.

CPU cycles have costs associated with them that a company with large jobs may not want to deal with. Computers take up considerable space, need copious air conditioning, and require repairs, and expensive staff to look after them (I know one portal company that is replacing it's U2's with quad boxes and better primarily because they don't have room for U2's anymore). They also become obsolete and depreciate. They also are static; these costs don't go away when you don't need the cycles.

So, total up these costs and you have an amount of money (say, $x) that the cycle user doesn't have to pay if the cycles are distributed. Split that $x between the owner of the distributed machines, the cycle user, and the organization in the middle, and everyone wins. The owners of the distributed machines would be doing all the repairs, housing, and so forth anyway, so they win. The cycle user can now rent or buy much smaller quarters and doesn't have to come up with the cash for the cycles up front; it can buy the cycles as they use them. Seems like a neat solution to me.

My main question is what kinds of money-producing computations can be done in this distributed fashion and not need results in real time. But i suspect that where there are accesible cycles, someone will find a use for them.

<PLUG>And if you want to sign up for ProcessTree, I wouldn't mind if you used this link.</PLUG>

mahlen

The truth is rarely pure, and never simple.

--Oscar Wilde

Who really wins... (3.00 / 1) (#13)
by commiskey on Tue Jul 11, 2000 at 12:20:28 PM EST

Since I first heard of this effort, I've wondered about what kind of pay scale they'll use. Is it enough to justify leaving my computer on (i.e., will it pay enough per cycle to cover the power bill and then some)? And what about the situation where the computer user incurs no costs by running this program? Let's pretend for a moment that you're a broke college student who happens to work in a computer lab. Let's say you've worked there for two years, and you've got keys to the place, etc. The lab's closed from midnight to 8 am. Their website suggests that a single system could be expected to generate roughly the cost of a dialup connection -- or about $20/month. Now, multiply that by 80 machines, each working nothing else for eight hours a day (assuming you'll disable it during the day to keep lab users from noticing). Even if each one only turns up $20, that's $1600 a month - $19200 a year. Not a bad job perk, if you ask me.

Re: Who really wins... (none / 0) (#16)
by Anonymous Hero on Thu Jul 13, 2000 at 10:12:56 AM EST

What about the univeristy computer services contract that states that campus computers are for coursework only, and not for money.

[ Parent ]
Security could be a problem (4.00 / 1) (#15)
by Anonymous Hero on Tue Jul 11, 2000 at 04:20:01 PM EST

Security is a major concern both ways. On the one hand, you need to make sure your distributed background process isn't snooping your hard drive or cracking your machine.

On the other hand, the user of a distributed network needs to worry about whether its members are spying on the process or sending back bad results -- faking them to get more credit, or just throwing a monkeywrench into the works.

I don't know if it'll ever be possible to make lots of money doing distributed computing, but at least it's a constructive effort; it provides a concrete service that they deserve to make money from. If it works, and they get competitors, an actual market in computing power could spring up...

Distributed Computing and Cash | 16 comments (13 topical, 3 editorial, 0 hidden)
Display: Sort:

kuro5hin.org

[XML]
All trademarks and copyrights on this page are owned by their respective companies. The Rest 2000 - Present Kuro5hin.org Inc.
See our legalese page for copyright policies. Please also read our Privacy Policy.
Kuro5hin.org is powered by Free Software, including Apache, Perl, and Linux, The Scoop Engine that runs this site is freely available, under the terms of the GPL.
Need some help? Email help@kuro5hin.org.
My heart's the long stairs.

Powered by Scoop create account | help/FAQ | mission | links | search | IRC | YOU choose the stories!