Slashdot Mirror
Flash Mob Supercomputer?
dan of the north writes "The NY Times (free reg yyy bbb) is running an article on flash mob computing. More info on the first event in SF on April 3, 2004. The goal is to run Linpack and "build a home-brew computer powerful enough to be added to a list of the world's 500 fastest computers." Minimum requirements are 1.3 GHZ Pentium III/AMD equivalent or better with 256MB of RAM, a 100 Base-T network connection and a CD-ROM - laptops preferred. "After taking a shot at a speed record, the computer will be reorganized to serve as the host of a giant multiplayer video game tournament." Cool... a 2fer!"
so, I'm going, obviously, but my big question is, beyond benchmarking, are we going to actually COMPUTE anything?
-and occasionaly a giant moose.
Add a few thousand temporarily owned computers and the odds go up quite a bit towards getting on the top 500 unclassified supercomputers.
It will never work. Clusters are bad enough bandwidth wise. Wifi will get destroyed. Processing power is relatively cheap and easy these days, but to really get a top50 you need to have a bandwidth solution (such as myrinet, or host of other proprietary things). Gigabit ethernet isn't even fast enough for a lot of applications.
Singer: Let me see your nodes!
Crowd: *yells*
Singer (to promoter): What the hell is a node anyways?
Promoter: Don't worry about it, just go with it!
Singer (to crowd again): I said let me see those fuckin' nodes!
Crowd: *yells louder*
Singer: Fuck right. That's what I fuckin' like to hear. Now, for our next motherfuckin' song, I want to see the most massive, the most fuckin' atrocious motherfuckin' pit on this motherfuckin' planet.
Yeah, that scenario was implausible. Thus, I don't see supercomputers in concerts anytime soon.
Until Slashdot fixes the funny modifier, use insightful or interesting. The poster knows your intentions.
I think the comment proposes distributed computing, not clustering. This is pretty reasonable wrt network consumption. (seti@home is not known for slowing down your net connection!)
It could actually work.
Ceci n'est pas une signature
For notebooks, 100% cpu usage has a definite impact. Seriously reduces battery life.
I am one of many. My idea is not unique, nor do I expect my voice alone to sway you. I speak in a chorus of opinion.
Given how long this will take to set up, battery power isn't a viable option. Still, using a laptop is a pretty good idea. If you compare a laptop drawing 65 watts to a desktop drawing 300 at full CPU utilization, with a knoppix CD spinning at full speed, plus monitor power, you see that they gain nearly an order of magnitude in energy efficiency, though this is probably offset a bit by the lower clock speed on the laptop processors.
Now, let's generously assume that each laptop is drawing half an amp at 110 volts. At 1200 laptops, that's 600 amps. The circuit breakers in my house trip at 15 amps, but I'll generously assume this facility has 50 amp wall circuits. That would still require 12 entire circuits, plus a safety factor, nevermind all my generous back-of-envelope assumptions.
Okay, so assuming they've got a lot of extension cords, now we just have to deal with space. Let's assume, again, generously, that each person + computer + associated infrastructure needs only one square meter of floor space. This makes the space requirement equivalent to a 30m x 40m area, or about two World Cup soccer fields. I hope they've got one hell of a big gym.
Heat is, by comparison, a relatively minor issue. If the facility can handle a crowd that large, adding their low-power laptops is minor. People tend to dissipate about 100 watts anyway, so the laptops won't be the most significant source of difficulty.
It sounds like a very daunting task they have ahead of them. I hope they've already gotten these problems figured out, because this project sounds really cool.
WARNING: there is a trojan on your
Yes, but you require a minimum of a 100 Base-T connection. You want to create one of the world's top 100 supercomputers using Ethernet? Good luck in beating that latency, guys....next time, see if you can get a flash mob of infiniband vendors to come along for the ride.
I bet we'll be seeing quite a few of those at flashmob supercomputing events...
blink...
I'm not sure the SOFTWARE aspect of this "Super Computer" but, uh... maybe the programmers are coding this in a Microsoft language...
Mentioned before is SETI@home (software distributed amongst many clients) and another (I'm too lazy to scroll back up and read). This "Psuedo-Supercomputer" aspect is mearly a distributed software application which shares data over a network. All hosts must run a program to "Link" them together in such a manner that they distribute the computing. Therefore, the platform of which this client is programmed for would be the limiting factor in the sucessfull implementation of this "SuperComputer". (eventually they may write clients for other platforms... but it sounds like they have settled on the PC for now).
Plus, they are Macs... heh. (what I mean by this is that I have an underlying contempt for something I can't rip open and reassemble the guts of, and Macs are much too "User Friendly" to properly "fix" in this manner.)
Hopefully this is coming off as humerous and not offensive, both of which I am capable of in a somewhat unpredictable manner, but in all actuality, it's probably just me rambling...
I started out with nothing, and still have most of it left.
...except, the promoters normally don't want you walking out with a high quality digital recording of the event. If you did, you would be less inclined to buy music from them.
If I seem short sighted, it is because I stand on the shoulders of midgets
Well not only that, but they will be booting off a provided Linux CD, which will enable them to join the cluster.
:)
If you are so worried just remove the HDD from the machine you take, and you won't have any problems?
"Hey! Unless this is a nude love-in, get the hell off my property!!"
Sure it's a cool stunt, but aside from running the Linpack benchmark, what will this pile 'o' pcs do? It will, of course, do nothing, for it will only exist about a day.
I've seen this sort of thing happen before: people devoting energy and money to what amounts to a technical fetish. The end state is a world where people like Robert G. Brown build themselves home beowulf clusters with no discernible purpose. (RGB: you're a nice guy and all, but I find it hard to believe that you need all that horsepower for personal use).
I'd rather see an article about broadband users organising themselves into a GLOBUS grid. For that matter, I'd like to see a comprehensive system for bug tracking MPICH (I've seen some weird bugs there). There's lots of things I'd like to see written about or developed. Tomorrows 'infinity + 1' Supercluster ain't it.
...and thus forcing the New York Times to implement a "pay per view" system for each article.
Thank you! That's some great thinking on your part!
Whoever designed level 61 in Frozen Bubble is a sadistic bastard.
Having done a fair bit of linpack benchmarking on 'real' clusters I have to say I think their chances are very slim. The interconnect makes a huge difference particularly when you have so little RAM per node. I doubt they will have a nonblocking switch architecture which makes a significant difference in Linpack (even setting aside it not being gigabit.) Also, MPI applications (Linpack included) often run into bottlenecks with wait conditions, some of the slower nodes will probably end up choking the entire cluster. A few problem laptops with bad RAM modules, and they can spend more time than they have pulling their hair out troubleshooting.
:)
That being said, it doesn't seem like all that serious an enterprise. Good luck to them, and if they have fun, hey all the better.
From flashmobcomputing.org: "Today, supercomputing is controlled largely by governmental organizations, academic research institutions, animation studios, and recently human genome companies. This means that the problems that get solved by supercomputers are narrow in scope and tightly controlled. We want to change that."
This is so stupid.
Besides, I don't think with 100Mbps interconnect their cluster will scale beyond performance of 32 P4 nodes with Myrinet...
This is an old idea - it's called bootlegging, and with a few notable exceptions, almost all record labels oppose it. Think about it - this idea would be much cheaper and easier to implement if they just offered to send you a CD later and took $5 and a postal address at the T-shirt stall. Plus, they would get revenue from all the non-ipod owners as well, and could fix the parts where the vocalist sings out of tune.
So who's the idiot?
If I seem short sighted, it is because I stand on the shoulders of midgets
This sounds like a fun event, but I'd be very surprised if they can make this work. Sure, if you get enough computers together, the "theoretical peak" (basically the aggregate FLOP rate of all the computers) can be really high, but for a problem with dependencies like the LINPACK benchmark, communication latency is what is going to limit the runtime when you try to hook a thousand or so laptop computers together using only fast ethernet. I think that even Gigabit ethernet would scale poorly in such a setting (decent bandwidth, but startup time for a communication is slow compared to something like Myrinet or Infiniband).
Maybe it's possible, but I think it would require a pretty creative network topology and some pretty clever re-structuring of the linpack benchmark (which is allowed by the Top 500 list rules, BTW).
How is this a flash mob if it's advertised weeks ahead and isn't going to just show up and then disperse quickly? This is just a planned event with no RSVP list, that doesn't make it a flash mob, they're just jumping on that currently-hip (not for long let's hope) name to get attention and feel cool. I'm surprised they didn't actually say "it's a flash mob of metrosexuals!" to really get attention from lazy people who love idiotic hip slang.
sig:
See the "..for smart people" banners Wired runs here? Look elsewhere guys.
This time we only know about it 6 weeks in advance... oh wait. Why is that "flash"???
This sig intentionally left justified.