Slashdot Mirror
Sun Grid Compute Utility
jbltgz writes "The Register is reporting that the long awaited Sun Grid Compute Utility has been opened to the public. Now you can run your CPU intensive jobs on a grid of AMD Opteron-based Sun Hardware for $1 per CPU per hour for a fraction of cost, in a fraction of the time."
How long will it be until botnet operators start up a similar service? Or am I out of date and they have already done this? Anyway kudos to Sun for offering this service.
ZzzzSleep
Emerald Astrology
I wonder how long it would take for someone to port the POVRay engine to Sun's grid? At $1 per CPU/hour, this could be a boon for amatuer 3D graphics designers and the Internet Ray Tracing competitors. Use low res renders during testing, then pay Sun $25 to get your high quality result back in 20 minutes rather than the next day. Could be a lot of fun. :-)
Can anyone think of other good uses for the average (or not so average) home user? Perhaps new image compression formats that rely on Sun's Grid to get the best compression/quality tradeoffs through brute-force power?
Javascript + Nintendo DSi = DSiCade
We're not having you modelling your nukes on our servers thankyouverymuch.
The ability to but powerful computing time is a cool idea that has been featured in several sci-fi novels. However the article fails to mention exactly how powerful these Sun CPUs are. How much bang do you get for your buck? They also fail to mention how hard it will be to write code for this platform. Can I simply send them some standard C source, or will I have to code using some special extensions that will make my code totally unportable and thus lock me into buying more and more time from them?
Philosophy.
I use grid computing for simulations. If I were charged for CPU-hours, you can bet I would be more careful about debugging. I've wasted thousands of CPU hours because of bugs, or sloppy configuration, in my simulator generating incorrect results. One bug was an infinite loop that resulted in 100 CPUs spinning for a week before I noticed!
Is it like getting an account on someone's server and then being able to do whatever the hell compute-intensive work you want? I can't seem to find the relavent details, or my Parkinson disease is kicking in.
Imagine a beowulf cluster of... oh, forget it
My blog. Good stuff (when I remember to update it). Read it.
Do you (the customer) supply the software to run on these distributed boxen?
Cause if that's the case, I can see a business model that involves lophtcrack or John the Ripper.
[Fuck Beta]
o0t!
I can outsource physics rendering at LAN parties and show my friends who's 1337 ;)
Casual Sun observers will be scratching their heads right about now, believing that Sun had already announced such a service a long time ago. That's correct.
rtfa kthnx
perpetually dwelling in the -1 pits
Click here to kick off a job on Sun's Compute Grid consisting of AMD Opteron-based Sun Hardware.
An Indian-American Hindu committed to non-violent thought/speech/action alarmed by the global explosion of radical Islam
The CPU is cheap, but you'll be paying an arm and a leg for "extras" like disk storage and memory.
Want a printout of your results? That's $100 per page . . .
Just think.
Upload your app, pay 1$ a CPU hour, charge to a credit card with a stolen identity...
Boom! -- Enter the biggest mail bomb credit card theft combo in history...
I wonder if I can play games on Sun's system. Perhaps a nice game of chess? Or maybe Global Thermonuclear War?
From their FAQ:
Q:
What are the components of the Sun Grid Compute Utility?
A:
The Sun Grid Compute Utility service consists of the following parts:
* Sun Fire dual processor Opteron-based servers with 4GB/RAM per CPU
* Solaris 10 (x64)
* Solaris 10 OS;
* Sun N1 Grid Engine 6 software;
* Grid Network Infrastructure of 1Gb switched Data Network and 100 Mb dedicated management network;
* Web-based access portal; and
* Internet-only access to upload data and applications (no physical access to location);
* Storage allocation of up to 10 GB per user account.
http://www.sun.com/service/sungrid/faq.xml
Wasn't their previous attempt to rent CPUs a failure?
I remember an article in slashdot about how the Sun grid was completely unused.
Wow, $1/CPU/hr. Same price as an MP3 off of iTunes, so it must be worthwhile, right?
OK, we are only about 3.5 months into the year of 2006, and lets look at some real data:
I run a few small to medium sized HPC clusters, and on one of them, here are the CPU hours used during 2006 -- 163,000+ this is on less than $500k of hardware that is years old. That would cost $163k just in computing time, not to include time to port applications, debug, etc.
Sun needs to be run by engineers and visionaries again, not by marketers. $1/CPU/hr is not going to do much better on those falling stock prices than selling $200 Linux PCs in Wal-Mart.
I mean, hell, here's 10TB of data that I'm currently backing up to tape.
Do you want me to package it as a J2EE WAR file? Fine!
Give a man a fish and you have fed him for today. Teach a man to fish, and he'll say "WHERE'S MY FISH, YOU IDIOT?"
if it boasted a 64bit Java VM. A mate of mine does some very interesting research in number theory, and a few of his applications would need massive amounts of fast addressable memory. 64bit of address space would conveniently suffice, i suppose. Any suggestions on what else (cheap, or at least affordable) to consider using, anyone?
:%s/Open Source/Free Software/g
YTARY!
where do I submit my deck of fortran punch-cards and where do I pickup the printout?
try { do() || do_not(); } catch (JediException err) { yoda(err); }
Sun has been offering this service for almost three years now, and to my knowledge they still have had zero customers. IBM seems to be the only utility computing supplier with any customers using their BlueGene solution in Yorktown. Virginia Tech even offered cycles to outside parties for $0.40 CPU/hour and had no takers. It's gonna be a while before something like this gets any traction. People are way too protective of their data to allow it off-site.
The time it'd take to send/upload the content would probably be longer than if you just plunked down and did it yourself on any relatively new computer.
The only way it'd make sense is if you have a lot of video to compress and even then, over the long term it might still make sense to buy a cheapo computer or two & DIY.
[Fuck Beta]
o0t!
I thought that Sun already had their grid available and that no one wanted to use it because they would have to agree to be in a marketing campaign. Is this still the case? The terms of service on the network.com site redirects to an error page.
Slashdot: Failed Car Analogies. Amateur Lawyering. Anecdote Battles.
This could be used to interactively zoom in and out of interesting fractal patterns with a high refresh rate.
Does this mean that we can now all get gentoo up and running in less then 50 hours?
Fire the current CEO and marketdroids. Apart from saving the salaries of a higly paid and underperforming execs, companies might actually want to do business with a company who's chief isn't soley concentrated on funneling as much money out of you as they possibly can. Thats if there are still some talented engineers who haven't deflected yet.
Scott McNealy they need you over at SCO.
If I could compile mencode/mplayer for Solaris I could upload my dvd isos and get sun to encode these for me in H264 for my HTPC.
I anticipate that each film would cost me ~$2. Not bad. Is that a safe bet? ANybody know what disk space they give for "personal files".
Now to explain to my ISP that I am not participating in illegal file sharing with +100GB per month of traffic is not going to be easy..
More seriously, I could use this to run some of my Monte-Carlo simulators..
Artificial intelligence is no match for natural stupidity
$1 per CPU per hour
On Slashdot the price should be published in WoW gold per SETI units elaborated.
That's 2.2 GHz per processor, times about a thousand processors or so. That's how modern supercomputers work. The processing nodes themselves are somewhat unimpressive, but they're built so that they scale really well, and deal with problems that are designed so as to be broken up into lots of little parts and solved simultaneously. So if you used all the processors on the machine for an hour, your bill (theoretically) would be $1,000.
... except that it has something like 12,000 of them.
The most powerful computer in the world right now, ASC Purple (it does nuclear weapons simulations for the USG), has 1.5 GHz RISC processors. Not exactly impressive, by today's standards
It's the infrastructure to get that many processors (and their associated dangly bits) talking to each other and working on the same problem efficiently that's expensive and nontrivial.
"Ladies and gentlemen, my killbot features Lotus Notes and a machine gun. It is the finest available."
In Soviet Russia, 'imagine' spells itself correctly (sorry, couldn't resist. Damn lack of impulse control...)
My blog. Good stuff (when I remember to update it). Read it.
I think they should just add a service where people could deliver or snail mail dual layer dvds or tapes with data, with a set of new dvds or tapes for results. One can overnight it with Fedex. Then, they run the program on the grid, load up the result on another set of dvds or tapes and overnight them back to you.
I see, so the point is it is for people who rarely need to do really large jobs, but do them very quickly.
So what sort of company/costumer needs to do really large jobs really quickly but doenst need to do them very often?
ASC Purple 63.39 TFlop/s 10240 processors
Blue Gene W 91.29 TFlop/s 40960 Processors
Blue Gene/L 280.6 TFlop/s 131072 Processors
09 F9 11 02 9D 74 E3 5B D8 41 56 C5 63 56 88 C0 is the magic number.
Like seti or folding@home except instead of donating your spare cpu cycles for one particular task you'd be making them available for anyone to rent?
The price per hour per cpu could be based on demand and could be distributed to all the contributers. Imagine all the processing power out there not being used. Especially the gpus on people's video cards while they're not playing games.
bite my glorious golden ass.
What he's saying is still correct though.
WTF? $1 to run a CPU for 1 hour
It being 100 CPUs at $100 an hour makes no difference.
Now if it was $1 to run on 100 CPU's for 1 hour, well, now we're talking.
As is it costs me a couple pennies a day to run an app on my 2.2ghz athlon64.
Don't know something? Look it up. Still don't know? Then ask.
So could someone explain to me what types of applications and organization would most benefit from this? Most larger scale operations already have an internal cluster setup, so that would eliminate them from contention. Since there is not web based API at this point, and not web interaction I can't imagine it being used for any time of spamming of any type. So the question remains, where is the business need?
Peter: I got an idea, an idea so smart my head would explode if I even began to know what I was talking about.
Oh please. We have cpu time for 10c an hour over at Andrea Arcangeli's CPUShare website:
...
http://www.cpushare.com/
Still experimental for now but soon
I did some math and I will build a server farm if I can get a steady 10c/hr/processor.
PenGun
Do What Now
Why wouldn't a mid- to large-sized organization use something like Condor? Just install it on everyone's (linux or win2k/xp) server/workstation, maybe set some prioritization scripts so that it would use more resources after-hours (when most people are out of the office, but have their systems on anyway), & save themselves $$ instead of paying to have their data on someone else's remote system?
anyone have any idea how much it would cost me to buy the # one spot on boinc?
p hp?pr=bo&st=0&to=100
Take down NEZ for one day-- that would be sweet
http://www.boincstats.com/stats/boinc_user_stats.
every day http://en.wikipedia.org/wiki/Special:Random
If only something like this were free.
Oh, wait Xgrid.
This is interesting. We have a 1600 CPU Linux cluster and we recently did an envelope calculation in regard to how much a single CPU hour would cost (based on initial investment, admin salaries, number of CPUs, projected number of processed jobs, projected end-of-life duration, etc) and we came up with $0.09/hour.
Anybody read Digital Fortress?
Any idea how powerful a cpu I get one hour of for $1?
Are there any competitors to the Sun Grid?
For example, is there a similar service for LINUX machines? I could certainly use a lot of extra machines to run simulations, but I mostly use LINUX. Having to switch everything over to Solaris makes this service much less useful.
Thanks.
One of the target markets for the Gifu Grid (a prefectural initiative here in Japan which we're supporting at my place of employment) is architectural firms doing earthquake simulations. I'm not privvy to the specifics, but apparently their big use case is spend a lot of man hours making changes to the building plan, get to Friday, hit "test for structural soundness" and go out drinking. When you're on a n-week construction cycle, spending two weeks waiting for the results of your compile (oops, sorry, at a 6.2 quake with certain assumptions this fails... BACK TO THE DRAWING BOARD!) costs you both money and your completition-on-time goal. But during the vast majority of the project cycle you *don't* need a server farm -- you've got X desktops running your favorite CAD software, which is all well and good, but if you had a supercomputer or beowulf cluster in the back room it would be powered down 98% of the time. So, theoretically, we could sell them time on our network -- we'll have your simulation done by Monday, you don't worry your architectural heads about setting up a Beowulf cluster.
Help poke pirates in the eyepatch, arr.
Here's an example: when your $1 billion (thats billion with a B) commercial highrise in Japan is undergoing earthquake testing, "Screw the grid alternative, I've got an Alienware at home we can run this on, should have the results back by 2127 or so..." and paying for, oh, 100,000 CPU hours (pulling number of of thin air) looks like a very attractive proposition (consider what you'd have to spend as an architectural firm to afford your own supercomputer or grid to make that happen, plus the maintenance and employee costs to keep it running, and the fact that you get no use of it for weeks or months at a time in between earthquake tests but the tests need to be done right and done fast so you can get on with construction).
Help poke pirates in the eyepatch, arr.
I imagine for most residential (and even some commercial) broadband users the latency of transferring the 4GB+ of VOB files outweighs the encoding time even on a modest desktop CPU
THIS THING CAN TURN ON A DIME, MACROSSZERO STYLE ALSO FUCK BETA, ~NYORON
This is all good with todays horsepower but what kind of company is going to upgrade a system like this in the famous 18 month period.
If Mores law holds true Suns revenue will half evrey 18 months, so the question is, is this permanent? Will they find new things to burn clocks? Or will they try to corner the market and still try to charge $1 according to a 2GHz machine.
The only way I could see this working is if they charge by clock rather than time and charge very little, this way you pay the same amount in 3 years but you get the result in 1/4 the time.
A loop, by its nature, continues. If that didn't make sense, start reading this sentence again.
Oh comon, they are just trying to cover-up getting slashdotted!
New.com article "Sun Grid hit by network attack" : http://news.com.com/2100-7349_3-6052968.html
Horns are really just a broken halo.
Skyne- no way. Nevermind. It'll never happen.
Unless I install this neat little proggy I wrote...
The 'Net is a waste of time, and that's exactly what's right about it. - William Gibson
You can run unmodified Linux binaries on Solaris 10 thanks to Janus.
That's nothing. Where's my Google Grid?
I'm going to go create my own technology news site, with blackjack and hookers. You know what? Forget the news site.
According to http://www.top500.org/, ASC Purple is number 3. Number 1 is BlueGene/L and number 2 is BlueGene W.
IBM was pushing computing-as-a-utility a few years ago. Their premise was that just as customers don't have their own power generators—they just buy electricity at market rates exactly as required—folks should be able to acquire computing time only as needed and let a company of professionals deal with all the maintenance issues. But no one bought into on-demand computing and IBM has since backed away from this position.
The reason this flopped for IBM was that everyone compared it to the "good old days" of dealing with shared computing facilities and batch loaders. It's the same reason people buy small clusters instead of paying for space with a national lab. The comparison to punch cards is apt, as are the discussions of "what if I spend all this time and money only to discover that I bug?"
I don't see how Sun is going to succeed in the market. Smaller customers will probably be better off buying a "personal supercomputer" rather than rent time. As for the really large customers, they will be running jobs 24/7 for years on lots of processors; it may in fact be cheaper for them to buy a BlueGene than rent time from Sun.
I stand corrected -- you're right, it's now number 3.
:)
There was a time when it was number 1, but damned if the folks at IBM haven't been busy in the meantime.
Computers...don't pay attention for a while, and suddenly everything you know is wrong.
"Ladies and gentlemen, my killbot features Lotus Notes and a machine gun. It is the finest available."
7054 frames in the rendering queue.
7036 frames in the rendering queue.
7001 frames in the rendering queue....
Biggest issue with this service: Data. Fast data crunching needs data. Lots of data. How does one get it there? And the result back. It's expensive and slow.
Tell your friends about xenu.net
From TFA:
"Sun has spent the last 18 months trying to make this happen. First, it battled the basic logistics behind creating a massive publicly accessible cluster, then it battled the government. Regulators didn't care for the idea that rogue nations could log on to the Sun cluster and design nuclear weapons or run dirty bomb simulations in their spare time."
"Sun has managed to assuage the government's fears by making the initial Sun grid a US thang only."
"Sun does plan to open the grid up to international customers one day."
"The grid that US customers find will be less ambitious than the one originally promised by Sun. Instead of myriad centers with 5,000 dual-core CPUs each, Sun will open one center with "less than" 5,000 dual-core CPUs."
So they were planning for it to be international with 5000 CPUs... and now it's US only and it's less than 5000 CPUs? Wow now that's what I call a suprise!
Time is an illusion. Lunchtime doubly so. - Douglas Adams
The grid itslef is comprised ov V20Z servers. Depending on which datacenter you are talking about they are either dual 2.0Ghz or dual 2.2Ghz, each with 4G of RAM. They have no disks and are booted via PXE.
The racks themselves have 32 systems, 4 racks make a pod (there was some discussion to call 2 rakcs a podlet). Depending on which datacenter you are talking about there are up to 3 pods.
There are also some management servers that are really not worth mentioning. They serve as the infrastructure to boot the pods etc.
One thing that is not put out on Sun's site is the pods will (and do) run Linux. You do not *have* to use Sol10 it is just that there are no Linux Infiniband drivers for the Infiniband (and yes thats about 2 tons of infinband cable connecting them, well ok in one data center).