Slashdot Mirror
Grid Computing: Conceptual Flyover For Developers
An anonymous reader writes "This article relates many Grid computing concepts to known quantities for developers, such as object-oriented programming, XML, and Web services. The author offers a reading list of white papers, articles, and books where you can find out more about Grid computing."
The linked article is written in May 2003 yet it's new now?
Hmm. It's not that hard, but I agree: the author seems to obfuscate a fairly simple to explain principle.
In a few words: grid computing is the use of many connected computers for one task.
Or, you might want to think about it as multithreading, but spread out over multiple machines.
The author is making a case for a standardization of how this should be handled.
Please login to access my lawn
I was merely highlighting how universal Grid Computing can be, and that you didn't actually need 'proper computers' to do it.
And yes, I understand the difference between a Cluster and a Grid.
Together, We Can Make Slashdot Better. I Do NOT Mod ACs. - Check Me Out
The grid discussed here seems only to be built on the OGSA and Globus Toolkit, and Globus has not really covered itself in glory with their poor UIs etc.
Grid seems to address occasional demand for "much more power" from your computing resource, but does not really provide a consistent flexible computing resource.
The academic world uses External Grids to pool resources but private Enterprise has little to gain from these External Grids in exchange for a HUGE security problem.
And Internal Grids? These are so immature as to beggar belief. Why risk investing in these configurations when bang per buck is so uninviting.
In a way it's a matter of taste, but I'd define it this way: "parallel" -> many CPU's, but quite close in 1 place, like in a SMP desktop. "distributed" -> with network in between, as in Beowulf cluster (possibly over the internet).
What would make "distributed" a Grid? The fact that it's 'everywhere', always working/available somewhere, like P2P networks. You can take your equipment off the network, but the network (ehh, grid) goes happily on doing its thing.
This becomes really useful when it's a easy to use and commonplace as the internet today. Send out some software, it grabs a piece of data here, grabs a program there, finds a server to do the computation, and reports back to you with the result. Got some cycles to spare? Put some in the Grid, earn money. Just wait and see, some day computing power will be supplied and consumed the way electric power is today.
Grid computing is the concept of using distributed resources as one big resource. For example, Boise State currently uses all of the computers in the Engineering labs as a super computer when classes are not in session. Micron Technology uses all of its desktop systems as one big super computer.
Todd Tannebaum just gave an exec lent keynote at Boise State's HPC Workshop. He explained that while computing power has increased on a system by system basis, the total available computing power to a single person has actually decreased. For example, if you wanted all the computing resources in the 70's you simply logged into the department computer. Now you can't do that. You get a fraction of the total power in the department. Grid computing is attempting to fix that by using the desktop systems together as a big super computer.
For more information on Grid Computing check out Condor. It is a super powerfull grid computing environment.
The idea seems to be to turn the whole network into a cluster.
This is answered in the FAQ at gridcomputing.com:
The key distinction between clusters and grids mainly lies in the way resources are managed. In case of clusters, the resource allocation is performed by a centralised resource manager and all nodes cooperatively work together as a single unified resource. In case of Grids, each node has its own resource manager and don't aim for providing a single system view.
Grid Computing - simple.
As far as the user is concerned, you have 1 Black Box system containing everything.
The physical implementation could be a single supercomputer, or a whole host of different systems spread all over teh place. But the the end user, it's just a single computer that handles all their stuff for them.
That's a grid. Check out http://wwws.sun.com/software/n1gridsystem/ for a good overview of how this can be implemented.
Remember, a true grid system is more than just raw CPUs, it's all about services, processor, disk resource, the whole thing.
People should not be afraid of their governments - Governments should be afraid of their people.
Think of grids as a distributed set of clusters. Instead of using distributed networks of single PCs you connect clusters.
The aim is get rid of the usual limitation to clusters (homogenious hardware/os) and make use of all the clusters in a virtual organization (think university network, distributed.net, seti@home protein folding and similar research institutions). You can then use a resource broker to specifiy you resource, say all big-endian machines with more than 1 gb of memory or all pentium machines.
Having played a little with Globus toolkit 2 the main limitation was that it required similar hardware/os to run as it was a C sdk, requiring a separate binary for all platforms. GT3 is written in Java to get rid of that limitation.
Basicly you write your program, compile it, place it on a resource visible to the grid and specifiy what kind of hardware you want to run the program on (number of cpus, memory requirements etc) and submit the job. Then wait for the results to pop out.
It has been standard practice over the past couple of years to overhype Grid, a practice which I suppose was intended to bootstrap interest but which instead just tends to leave people feeling confused and vaguely betrayed as they discover that what was presented as a production capability turns out to be a research project in its early stages. The article is typical of the approach.
There's a huge conceptual gap between "Grid as utility computing" and "Grid as a usable set of application services" which no amount of hype will close. Those of us who actually have to talk to upset users about this are left in a very embarassing position.
Yes, the basic Grid vision is right, and the Grid Services Architecture appears conceptually sound, even if implementations are not yet complete, let alone fully interoperable. No, in practice it's nothing remotely like plugging an appliance into a power outlet to toast your bread. Will it ever be? I'd say that at this point in history, we can't know.
Apart from the requirement of functional completeness over a very ambitious domain of capabilities, the Grid computational model must also achieve an ambitious degree of interoperability, that is, if our goal is truly to capture unused compute cycles rather than to justify the development of new computing infrastructure. Amd on the subject of interoperability, Gordon Bell has a few cautionary words:
Gordon Bell is being somewhat unkind here, because he knows better than to treat the hype at face value. Your perspective is more realistic. Supercomputing is a relatively natural platform for implementing Grid applications, but adoption is proving to be nontrivial even there. That's the place to watch for it. Meanwhile, my advice would be to ignore the hype, but by all means read Foster and Kessleman. It's interesting stuff.