posted
Interesting thought, but isn't the point of the internet the fact that it is INTERconnected NETworks, thus providing redundancy and theoretical immunity to overall electronic attack? If it got put as an application on a single computer, I'd imagine it'd be having complete crashes. . . .often. . . . .
Posts: 1323 | Registered: Aug 2001
| IP: Logged |
quote:Originally posted by Magson: If it got put as an application on a single computer, I'd imagine it'd be having complete crashes. . . .often. . . . .
Systems that huge generally have massive redundancy built in so a crash should be rare. It could still be blown up though which is a huge weakness.
Posts: 1327 | Registered: Aug 2007
| IP: Logged |
quote:The Kittyhawk project has created much discussion in the technology community. One of the main topics of discussion is the comparison of Kittyhawk to Skynet the fictional supercomputer, in the movie series Terminator, which attempts to destroy all human life and take over Earth.
Funny this coming around the same time as the release of the new Sarah Conner chronicles and the new terminator movie -- the first in a decade. Posts: 3295 | Registered: Jun 2004
| IP: Logged |
posted
The only way I could see this coming about would be as a temporary fix while they rebuild the Internet from the Hatrack up.
Posts: 636 | Registered: Apr 2002
| IP: Logged |
posted
The interesting thing to me is this: at no point in history has humanity been prepared for the consequences or possibilities represented in the next big invention.
No one thought there was an application for radio when they first heard it- it didn't catch on for decades, but our thinking and our prioritizing of ideas changed to fill the new creative space. This is partly why the recording industry developed.
I read an essay a few years ago about new inventions by Douglas Adams. He talked about what he called the "ages of sand," and how silicone has affected our lives over the years, our use of it morphing to fit what we need, but also our needs expanding to fill its possibilities. He argued essentially that people are mostly incapable of imagining what they are capable of, but tend to do things by sheer momentum, and because the possibilities are just there. We expand ourselves and what we do to fit the possibilities we see, but the ones we don't see are filled later.
With this computer, or rather with the coming age of computing, I think that intelligent machines will begin to take on a different role in our lives. Adams imagined that eventually computers or all electronic devices would be like guardian angels, or the natural world working for us everywhere, in places we couldn't see or understand individually. For some time we've applied computers to the technologies we had before computers: ie, typing is now word processing, film reals are now avi's, the printed world becomes the internet, in some ways, the internet is like the rest of the world in a limited form. But we begin to understand the world a different way and allow computers to do things for us that we had never done before- sound engineering is a good example, but even sound engineering was an idea and a practice throughout human history.
I think the interesting part comes when we discover our abilities to truly exceed our human accomplishments through the use of computers. It won't even feel like use, the way that language feels a part of us and not something that is a tool. I'm not afraid of this, but I know some people are. I think our essential nature will change as our technology allows it, but I think there is no other possible direction in which we can go and gain satisfaction. We advance, always, over time.
Stephen Hawking has also written some things about this, saying that the denial of technologies future impact on our lives is antithetical to our nature- that it isn't a matter of IF we become a race married in every way to advanced technology, but when- I think he points out that this is something that has already happened to us, and that we have evolved and survived by using tools, making advancement in technology a primary goal of or existence- seeking satisfaction by the accumulation of power and knowledge, and continuing our survival.
I welcome it, for one. I'm constantly excited by the idea of what I may see possible in this life.
Posts: 9912 | Registered: Nov 2005
| IP: Logged |
posted
As long as there are valid reasons to run your own server, people will not pay IBM to host sites for them.
Posts: 37449 | Registered: May 1999
| IP: Logged |
posted
Tom, they'll do it if the price is right. IBM's point in the white paper is a good one- internet providers are terribly inefficient at expanding their capacity, because they simply lack the capital to invest in more ambitious technology at the outset. As one article summed it up, internet providers look like power companies who try to supply power by hooking up fields of generators in big groups. No-one is building a big power plant with the systematic efficiency that is proposed here. The kind of processing bandwidth they're theorizing for this machine would actually handled more information than there is a known use for- and once you build one of these machines, it isn't the same leap to build another, better one.
Posts: 9912 | Registered: Nov 2005
| IP: Logged |
posted
What's the point? Privately funded data banks and server farms already exist. The GooglePlex alone is several times larger than BlueGene. And will probably have several times the capacity of Kittyhawk by the time Kittyhawk could be built. All privately funded. And nearly all built from off-the-shelf equipment.
posted
I'd be delighted if you would explain why that is. The three articles I read on this seemed to indicate there were significant benefits, such as faster and more efficient expansion, energy conservation, reliability, reduced maintenance, and superior performance.
Why would the whitepaper indicate that these advantages would exist? It is a business after all, they are looking for better ideas. Also, I should point out that no other company has access to the particular architecture that IBM proposes to use, and I don't imagine many, if any, would be capable of actually building this machine- including IBM. So I'm perplexed that you're explanation for it not existing is that there is no need, when historically speaking, we find new needs for new inventions, we don't not build them because we have workarounds for our current technological deficiencies.
Posts: 9912 | Registered: Nov 2005
| IP: Logged |
quote:The three articles I read on this seemed to indicate there were significant benefits, such as faster and more efficient expansion, energy conservation, reliability, reduced maintenance, and superior performance.
These are benefits only if you think that customers are demanding increased efficiency from "the Internet" -- especially at the cost of redundancy, flexibility, and privacy.
The advantage of the Internet as a medium is that it is in effect a giant distributed application.
For pure processing power, individual firms have always had the ability to build massive server farms on their back-end. While I can see a few reasons to build a single-machine architecture to duplicate this option, none of those reasons would include "hosting the Internet."
Posts: 37449 | Registered: May 1999
| IP: Logged |
posted
Reliability? Putting lots of things on one heavily interlinked computer is the least kind of reliability. While individual parts of the internet are low reliability right now, the thing as a whole is virtually indestructible. A single supercomputer, no matter how much in the way of redundant resources are available, has a much, much higher risk of bringing down substantial infrastructure.
Additionally, I doubt their ability to scale at anything like a similar rate to what the internet does, particularly and maintain reliability.
However, whatever the rhetoric of IBM over the potential capacity and suitability as a replacement for the internet, I do know that many, many people will buy capacity from them, and that there is a very good value to be had from that amount of computing capacity.
Commodity computing is taking off right now. Combine a slice of computing capacity from IBM with the amazing virtualized server management capabilities from 3tera and it will be possible to put together a more capable data center (in particular, with much faster internal communication, which opens up new architectural possibilities) for a fifth to half the price (and even less proportional hassle) of what it would take to run it internally, most likely.
Especially given how much buy in EC2 and other virutalization and commoditization efforts are having at their current primitive states, I find it safe to say that the value statement of having direct control over one's hardware is far less important for most than many think. Giving up some physical control usually allows for large decreases in expense and hassle and large increases in reliability and scalability.
aspectre: this will be bigger than all of what Google has or is likely to have in the next few years at their current rate of expansion, and not really the same sort of thing, anyways. Each individual google server farm is comparatively tiny, for one thing, and cannot stop being tiny with the architecture they currently have. Also, I'm not sure why you're harping about private funding -- I see no indication IBM is seeking gov't funding.
Orincoro: IBM trumpets every big project they do as more than it is. This is part of how they get great press, which leads to new consulting clients, which leads to new hardware sales. That is the IBM business model. In fact, just about every whitepaper you read from a business is distorted and hyped, having been filtered through layers of PR and executives after the facts departed the hands of the engineers (who are not immune to overhyping nifty things they've created, either).
Posts: 15770 | Registered: Dec 2001
| IP: Logged |
posted
GooglePlex alone is already much faster than IBM's BlueGeneL. If mainframing were a more efficient means of expansion, Google would have already done it.
Better energy conservation I'll give them, provided that IBM could actually build the Kittyhawk.
Better reliability, reduced maintenance, and superior performance are sales pitches. Might as well add that Kittyhawk will figure-skate better. It's easy to claim anything whatsoever about a product that doesn't exist.
That "faster and more efficient expansion" is so dubious that it's closer to a unicorn than a sales pitch. Multi-petaflop computing and tying together multi-petaflop computers isn't a simple scalability problem. ie It isn't even the same type of algorithmical/mathematical/physics problem as tying together thousands of gigaflop computers into multi-teraflop systems, or tying those multi-teraflop systems together into a multi-petaflop distributed system. As recently as a few years ago (within the time I've been Hatracking), developing algorithms to handle multi-petaflop computing had yet to be demonstrated as mathematically possible, let alone efficient&practical. And hardware engineers need working algorithms to design the circuits. Admittedly I haven't been paying attention to the field in the years since.....but then the types of science magazines I normally read would all have given such a mathematical proof or practical demonstration a screaming headline that a reader couldn't miss.
Kittyhawk certainly won't be more reliable. Not when one mishap can take down the whole Internet. eg Recently, some Iranian subs or some subs privately owned by wealthy Saudi fanatics probably severed four undersea Internet communication cables serving the MiddleEast. Taking out cables leading into a single facility on dry land would be a cakewalk.
Reduced maintenance isn't as important as ease of maintenance. If a component of a massively-parallel distributed-computing system goes out, ya just pull out the old computer and swap in a new one. Odds are extremely high that no one other than maintenance personnel would even notice an outage. Even with nearly all of the major cables into the MiddleEast cut, the sabotage reduced the speed at which people could access the Internet, but did not cause more than short-term outages for most users. Critical-need end-users such as financial institutions merely paid higher prices for satellite links. The closer Kittyhawk approaches maximum computational and energy efficiencies, the more closely bundled its subsytems will be. Which makes those subsystem bundles become ever more critical such that the failure of a critical subsystem is ever more likely to cause noticeable disruptions of service.
posted
There isn't a direct way to compare 'overall speed' (whatever that is) between Google's data centers and the Blue Gene/L, so I don't know where you're getting that one is faster than the other. Surely you have a citation before making such a bold assertion?
And your next statement underscores your lack of understanding of the differences: Google's value is not principally realized by many of the qualities IBM is going for in Blue Gene. For instance, Google's compute-farms (which are not even close to the level of integration that Blue Gene is; they are true clusters, where Blue Gene is a cluster/multiprocessor hybrid) would be horrible platforms for large-scale virtualization, while Blue Gene is ideal. Since google was not looking to do large scale virtualization with them, this is not a problem.
Posts: 15770 | Registered: Dec 2001
| IP: Logged |
posted
Btw, I am confident that the language I am using about size is accurate. Most estimates put Google at 'only' 450,000 machines, and that's in a variety of clusters scattered around. A single one of those clusters, the only thing comparable to one of the IBM computers, tops out at around 150,000 machines. A small Blue Gene/P will have 294,912 processors sometime this year, and the architecture scales much higher than that.
Posts: 15770 | Registered: Dec 2001
| IP: Logged |
quote:Originally posted by aspectre: ... As recently as a few years ago (within the time I've been Hatracking), developing algorithms to handle multi-petaflop computing had yet to be demonstrated as mathematically possible, let alone efficient&practical. And hardware engineers need working algorithms to design the circuits.
*raises eyebrow* Mathematically impossible?
Elaborate please, AFAIK, its not like simply using a massive computer would be NP-complete or anything. What *mathematical* problems would there be with massive computing on a scale such as this short of *implementation* problems such as the limited range of 32(64, 128, ...)-bit datatypes or so forth?
Posts: 7593 | Registered: Sep 2006
| IP: Logged |
posted
See, if we put all the Internet on one machine, or even one dozen machines, then countries like China, organizations like the CIA, and the Zurich Banking Council would be able to better manage your information--deleting what was inconvenient and squashing what was dangerous.
Of course, all of this in the name of National Security, Pedophilia Obliteration, and Capitalism/Property Rights.
Posts: 11895 | Registered: Apr 2002
| IP: Logged |
posted
Mucus: yeah; any trivially parallelizable algorithm will scale infinitely with computing power. That's what it means to be trivially parallelizable.
Posts: 15770 | Registered: Dec 2001
| IP: Logged |
Printer-friendly view of this topic
![[Big Grin]](biggrin.gif)
