Tim Bray flies over the grid computing landscape and sends back a clear report on its current topography. Noting that “nobody agrees what Grid means,” he offers a catholic definition: “any scenario where you want a good way to throw a whole bunch of computers in parallel at a problem, and you want some infrastructure there to help.” He then divides grids into two categories: batch grids, which perform discrete jobs requiring a lot of data-crunching muscle (jobs with clear beginnings and endings), and service-oriented grids, which perform large quantities of transactions continuously (a la Google’s search engine). He then reviews a number of prominent grid technologies within each category. Bray also provides a succinct assessment of when it makes sense to use a grid (and when it doesn’t). The key is to recognize that transporting data is relatively more expensive than storing it or processing it, so when using a grid “you’d like to maximize the ratio of the amount of computation to the amount of data traffic.”
I see my job as being a popularizer – opening up specialized topics for general readers. (It’s a job that’s often pooh-poohed by the specialists, who equate simplification with dumbing-down, but I happen to think it’s pretty valuable.) So it’s always helpful to have an expert like Bray offer concise definitions and distinctions regarding subjects that are often buried in technical (and marketing) jargon. I’ve found it a challenge to explain things like grid computing and virtualization to laypersons, even though I think they’re essential to understanding the present and future of computing and the computing industry (and a lot of other things by extension). Through a process of whittling down, I’ve come up with the following definitions as starting points:
Grid computing: having two or more computers act as one computer
Virtual computing: having one computer act as two or more computers
Virtual grid computing: having two or more computers act as one computer that acts as two or more computers
Now, that last one, I admit, seems kind of confusing at first, but I think it ends up making the whole subject clearer in the end. It underscores the difference between partitions defined by hardware and partitions defined by software, which might also be called physical and logical partitions. In a virtual grid, you need to first erase the physical boundaries between the machines so that you can then create, with software, new boundaries when and where you need them.
At a technical level, it makes a lot of sense to transcend the physical. It’s when you try to do the same thing at the human level – going from virtual grid computing to virtual grid culture – that I get nervous. See? Everything I talk about here is connected. Rough Type may not be quite as schizophrenic as you assumed. The bipolar thing is another matter.
Grid computing is different than cluster computing (multiple nodes that try to simulate one) in that communication between the nodes is limited by distance, smaller bandwidth and/or the disparate technology of the nodes. As a result, it becomes impractical for the nodes to communicate as one. Instead, each node or cluster of nodes “bites off” a chunk of processing that it can do by itself. When it gets done processing, it returns the results to the collective. A grid computing environment is more useful for evolutionary or NP-complete problems than for massive integrated algorithms.
The definition of Grid has changed over the years, and has been redefined by each new group of people who adopt the concepts.
This might be useful background for some people. I gave this talk containing a Grid introduction and history at the SF Bay ACM meeting in October 2004. They posted the slides, which were the first part of my half day tutorial at CMG04.
I see my job as being a popularizer – opening up specialized topics for general readers. (It’s a job that’s often pooh-poohed by the specialists, who equate simplification with dumbing-down, but I happen to think it’s pretty valuable.)
I see you as a hack, and a pretty good one. It’s a much-underrated role; ‘hack’ probably doesn’t even look like a compliment unless you’re a hack yourself. (It has something in common with ‘hacker‘ in this respect.)
partitions defined by hardware and partitions defined by software, which might also be called physical and logical partitions. In a virtual grid, you need to first erase the physical boundaries between the machines so that you can then create, with software, new boundaries when and where you need them.
Zigackly. This is precisely why I used to spend a lot of my time shilling for the AS/400 / iSeries / i5. (Well, the reason I did it was that I was a hack and it was a good gig, but that was the reason I enjoyed doing it.) PC technology has a lot of catching up to do.
Could you repost those links, Adrian?
Zephram,
I think I have fixed Adrian’s links.
May I whine for a second? Folks, when you put html tags into comments, please please please proofread everything carefully before posting.
That’s it. Thanks.
Nick
I’ll say this: I am definitely adopting your definitions of grid and virtual computing. And, for that matter virtual grid computing.
“Grid” has become such a broken term that it has long needed someone to haul it behind the woodshed and give it a good working over.
Quote:
Pixels tend to require a lot of memory. The more pixels and the more bits per pixel you use, the more memory you need to store them. A typical screen with 1,000 by 1,000 pixels, in full color, needs 24 million bits of memory. When I was a freshman at MIT in 1961, memory cost about a dollar per bit. Today, 24 million bits costs $60, which means we can more or less ignore the large appetite for memory of pixel-oriented computer graphics.
As recently as five years ago, this was not the case, and people economised by using fewer pixels per screen and by using far fewer bits per pixel. In fact, early raster scan displays tended to use only one bit per pixel, out of which we inherited a special problem: the jaggies.
Unacceptable Jaggies
Have you ever wondered why you computer screen has jagged lines? Why do images of pyramids look like ziggurats? Why do uppercase E, L and T look so good, yet S, W and O look like they’ve been drawn by someone with palsy?
The reason is that only one bit per pixel is being used to display the image, and the result is a staircase effect, or spatial aliasing, which is absolutely unneccessary if hardware and software manufacturers would just use more bits per pixel and throw a little computing power at the problem.
Endquote.
Taken from ‘Being Digital’, by Nicholas Negroponte, 1995.
Scott McNealy said, we would cry if we knew what our desktop computing experience could be like. McNealy and his company Sun Microsystems, envision a more economical way to deliver computing resources to the end user. Having invested a lot of my time doing research, I have found that an understanding of Moore’s Law, requires appreciation of the total context of the last 50 years in computing. Negroponte’s book, Being Digital and also John Markoff’s book, What the Dormouse Said, are good resources for anyone trying to understand Moore’s Law. Having read Negroponte’s quote above, you will understand how the current model for delivery of computing power is a result of restrictions that existed only in the past. Namely the cost of memory was so variable, and so very high. Dell’s business model took advantage of that, by not holding too much stock. But as the availability of memory ‘overshoots’ the customers ability to use it, you will see that Dell’s business model looks less and less sound. Utility computing is just a way to attack the Dell business model, and by definition, that of Microsoft also.
Negroponte, in his book, Being Digital, makes a point that television’s 3×4 aspect ratio is a legacy from a standard popular in movies during post war America. But the movie industry changed and adopted new stanards to differentiate itself from TV’s 3×4 standard. As always, market differeniation is important and the film industry knew it. I think a similar thing is happening in computing. Computing is stuck with a standard that related to a shortage of memory.
Utility computing is like having super trucks deliver produce via a highway, to a mega-market located someplace on the outskirts of a metropolis. The alternate business model is having small vans shuttle small deliveries, down narrow restricted streets in the city centre. To facilitate the ‘just-in-time’ delivery model used by most city centre boutiques and small shops. Dell corporation’s business model does exactly that for selling computing power. They use inexpensive cardboard boxes and delivery vans to place their Intel boxes on every desktop. They use a carpet bombing advertising campaign to pin point individual customers, and catch them in their net. The small business is satisfied at present to buy its computing power in that way.
Utlity computing is another way to explore the opportunities presented by Moore’s Law. The people selling utility computing products nowadays, will learn new things about their end user. Things you can’t learn shipping card board boxes. The user will look less like an individual, and more like a group or larger gathering of people. The current Dell computer sales model implies the breaking up of companies into tiny individual bits. Indeed, Dells sales model has implied the breaking up of companies full stop. The historical sequence of things: government owned computers, corporation owned mainframes, department owned mini-computers and individual owned PCs. Now it seems we need to travel in the opposite direction agani. I predict the opportunities in computing in the future will be in supplying virtual mini-computers to workgroups. Allowing project groups to accomplish tasks, which are impossible to do using a fleet of Dell PCs. Where the whole group has it’s intelligence augmented many times by the advent of virtual mini-computers. If you need your project to go online, and connect people in different geographical locations, you need to minimise your administration costs. Having people own individual systems, is not going to reduce your administration costs, and it is not going to improve communications. A standard needs to be adopted for each project, and a virtual minicomputer is the only way of achieving that. There are advantages and saving in having all the bits in one place at the one time, exposed for manipulation by a group of inquisitive minds. Yet that opportunity is unavailable to a bunch of guys using Dells.
Sun Microsystems is a unique company. It has the necessary technology, engineering capability and IP rights to implement a new alternate vision of computing for work groups. Sun is trying to re-shape the perception of the computer user. A perception, which I think is in need of re-shaping. A return to a McCarthy-like vision of computing is something I would welcome. It would reinforce the idea of the company, and breath new life into it. Eventually computing could empower the institution. At the moment, institutions are crippled by this Dell, loads of tiny little boxes’ sales model. In the future, Sun Microsystems will benefit from an advanced understanding of the customer, equal to work group or smart mob, rather than individual. Lets face it, there is no such thing as ‘my information’ anymore, or my anything – it is just an illusion attached to the out-dated ideals of the PC. I hope that Sun will benefit from an understanding that is based on more dimensions, than card board boxes with Dell brand logos on them.
Data mining of information from check-outs, in mid-western shopping centers revealed a relationship between beer and nappies on thursday evening. In the real world, no one could have connected these two together. It took data mining to reveal that men were being sent by their wifes to buy nappies on a Thursday evening, and men thinking in advance, used the opportunity to buy beer for the weekend. When you sell anything at this larger scale, it allows to to explore the product(s) you are selling, in more dimensions.
At the moment, the Dell business model of boxing everything cannot benefit from that information about the consumer. Dell have come up with the equivalent of the small high street store way to deliver computing power. You get a standard package, a standard allocation of memory, software, processing power and storage. This is only a rough approximation about the market on any given year. It does not take into account the realities of business. Namely that a company might need gob loads of computing power, due to a high pressure project this month. But next month, the group’s usage level may drop to that of basic work processing and administration task. Companies know this, and are prevented from buying a new expensive and powerful Dell computer, because they know it will just sit there idle for alternate months in the year. The Dell model is too restrictive, it only explores some of the dimensions of computing. The market is wide open for a new alternative way to distribute computing power, to end user (workgroups) more effectively.
The current model doesn’t allow you to lever more memory and processing power today. While needing more bandwidth and accelerated colour displays for certain tasks next month. Just like beer and nappies were an important combination on thursday nights in the American mid-west. With virtualisation, you can explore how the dimensions of computing interact. With Dell, everything is decided by what you can fit into a card board box. I spent a year once working for Dell, putting their machines into card board boxes, and shipping them out. It it left me with a feeling, there is much waste in this paradigm. That there must be another way.
It goes back to Nicholas G. Carr’s observation in ‘Does IT Matter’ about the nature of the firm. Do you need to the company to disperse, and everyone is separate and connected ‘just in time’ by wires. IT companies like Microsoft and Dell, would have you believe, in needs to be like this. In effect, because IT infrastructure is fragmented, the company needs to be fragmented too. Like how software, encourages the user to think like the software. IT encourages the company to think like IT. Supposedly, using these windows desktop boxes, you can simulate a company by having everyone connected by email. I am living through that paradigm everyday and it doesn’t work for every project, or every stage of every project. It works for basic administration things. Stages of the project when people are waiting around. But when you need to turn the gas up, I think workgroups would be willing to opt for a more powerful enabler like a virtual minicomputer. You do need people to physically come together. To allow this to happen, you need a virtual minicomputer. There are important reason for having large companies. It takes strong companies with vision, to stick to their goal and see it through. In the information economy, this conviction seems to be crucial to success.
Brian O’ Hanlon.
Quote:
The free-lance designers of games today must realise that their games are most likely to be best-sellers if built for a general-purpose platform, of which Intel alone plans to sell a hundred million a year. For this reason, the computer graphics of PCs will evolve rapidly toward what you see today in the most advanced arcade games. PC-based games will overtake dedicated game systems as we know them today. The only place where special-purpose hardware may play a near-term role is in virtual reality. Endquote.
From ‘Being Digital’ by Nicholas Negroponte, 1995.
If you all want a guilt-edged, shining example of computing for groups rather than individuals, you need only look at gaming consoles. What has happen to PC gaming since Negroponte’s 1995 prediction of doom for the gaming console – is that PC gaming is reserved for people who game alone in their bedrooms. While the gaming console has managed to capture the audience of gamers who prefer their experience to be communal. Namely, in the living room, with bags of pop corn and loud cheering and fun in general. The PC gaming experience is a different kind of gaming experience. Negroponte underestimated the resolve of console companies ten years ago. He also underestimated the value of a communal experience of computing. It does serve to underline, how blinkered the vision can become. Powerful PCs will fall into a similar niche in business. Their use being reserved for occasions where the task demands a user to withdraw from the group. In order to concentrate upon a problem in isolation. This will affect the way IT companies do their business. People will want to return to collective authorship. They will naturally choose a platform that facilitates that. That is why I encourage readers of this blog site, to check out John Markoff’s book, What the Dormouse Said. Because if nothing else, it gives you a sense of what communal workgroups were like. Both their advantages and their disadvantages.
Brian O’ Hanlon.
I am passionate on this subject, having worked with many architects in the past doing urban masterplans. It is an entirely frustrating experience trying to get a computer powerful enough to accelerate the project. By the time those nice new Dell boxes arrive through the door, most of the project has been completed. Completed, without the aid of a computer to understand the design problem in more dimensions. The human brain and a sketchpad is much faster than waiting for accounts to approve the purchase of a new computer. So the problem of designing architecture, has never been fully acclerated by computers. Microsoft Excel provides a powerful tool to business, to undertake ‘what-if’ analysis. It would be nice to do the same in designing buildings and urban spaces. You only get a few weeks on most projects, to make a computer really do work for you. It could be two years, before you need to use a computer again, on a hard design problem, in the same way. Which is why most companies, will not invest in expensive silicon. It is too costly to have a depreciating asset on their books. I wrote these two articles for CG Architect web site. They might prove interesting to readers here.
Brian O’ Hanlon.
http://www.cgarchitect.com/news/Reviews/Review044_1.asp
http://www.cgarchitect.com/news/Reviews/Review048_1.asp
Few corporates have any need for grid computing. They’re an alleged solution in search of a problem
The key thing about grid computing for me, is that it doesn’t approach the problem, as shipping atoms directly to customers, so the customer can extract computational power from the silicon. It works in automobiles to generate your power locally, because it reduces the pollution believe it or not. If we were to use batteries in cars, the power would still have to be generated at a central power station. I was watching a good Michael Douglas old movie last night, called the China Syndrome. It highlights some of the problems with central power plants. The movie is a very interesting counterpoint to today’s debate about oil and the middle east. Check it out, if you get a chance. Grid computing is useful to me, because it delivers you the bits via a re-usable conduit like fibre optic. Instead of shipping you electronic equipment, which needs to be dumped and adds eventually to the land filling problem. Michael Dell has built an empire around the mere fact, that all that electronic equipment becomes redundant quickly and needs to be dumped. Inadvertedly, you might say, Michael is really in the business of moving thrash.
Brian O’ Hanlon.
Jaron Lanier, doesn’t seem to go for the collectivism fetish, as he explains in this article:
http://www.edge.org/documents/archive/edge183.html
Lanier makes a lot of good points on weaknesses with collective authorship. Richard M. Stallman, I think is at the other side of the debate, and talks about the positive sides of collectivism.
Quote from Lanier:
What we are witnessing today is the alarming rise of the fallacy of the infallible collective. Numerous elite organizations have been swept off their feet by the idea. They are inspired by the rise of the Wikipedia, by the wealth of Google, and by the rush of entrepreneurs to be the most Meta. Government agencies, top corporate planning departments, and major universities have all gotten the bug.
As a consultant, I used to be asked to test an idea or propose a new one to solve a problem. In the last couple of years I’ve often been asked to work quite differently. You might find me and the other consultants filling out survey forms or tweaking edits to a collective essay. I’m saying and doing much less than I used to, even though I’m still being paid the same amount. Maybe I shouldn’t complain, but the actions of big institutions do matter, and it’s time to speak out against the collectivity fad that is upon us.
Another quote from Jaron Lanier, worthy to be posted here I feel.
“When a government bureaucrat sets a price, for instance, the result is often inferior to the answer that would come from a reasonably informed collective that is reasonably free of manipulation or runaway internal resonances. But when a collective designs a product, you get design by committee, which is a derogatory expression for a reason.”
A very good resource, to understand this point, is Frederick Brook’s literature. In particular, if you can watch it, his 2001 Siggraph lecture, where he directly responds to the ideas Eric S. Raymond put foward in his book, The Cathedral and the Bazaar. Basically, Fred noticed that Linux hackers know a lot about the things they are designing as a group. Namely, computer programs. But where you don’t understand enough about what you are designing, a group is not nearly as effective.