There is a need for faster and more powerful computers in some topics, like climate research, as researchers are under time pressure. They would like, for example, to present a working climate model before the reality takes over and there is no chance any more to react on the results of predictions from the model. In this article Karl Sarnow, Xplora portal manager, says us more about Grid Computing.
To deliver computational power, two ways have been proven to work:
- use the best hardware components, design concepts and most innovative technology to build the fastest computer possible. This approach is obvious and we can follow the news about new supercomputers every day. The disadvantage of this approach is the need for special software, running on these super machines, which in addition has to be redesigned with every new hardware concept;
- use ordinary computers, which will follow the edge of development with a delay but they are available in great amounts and then bundle the low computing power by summing them up. This approach is simple and uses well known resources. The computing power arises using the divide and conquer principle, which divides a big problem into small several problems and give them to many computers. After finding the solution of the small problems, they are given back to a central machine to rebuild the integral solution from the parts. The
idea behind this concept is called “parallel processing”.
Under normal conditions, parallel computers are built in one location, having many computers at disposal and reaching supercomputer units of hundreds of CPUs.
One step ahead is a concept, in which all computers in the world are contributing to the main solution. They receive problems from a central machine, which is finally building up the computational results from the individual solutions of the small PC's everywhere in the world. The distribution of problems to many PC's around the world as well as the collection of results, is managed via the Internet. To get a large number of PC owners volunteering in the computational power contribution, the integration into the massive parallel computing process is managed by a special screen saver program, which activates the CPU for contribution to the world wide computational process only, when the user is not active at the keyboard. This variant of parallel computing is called GRID computing.
What Grid Computing can be used for
Problems for GRID computing are those, that can be divided in many sub tasks. Examples are climate prediction problems, where a great number of parameters influence the prediction of a climate model. Every parameters influence can then be given to a CPU somewhere in a world, which would normally consume energy and not do anything but driving a screen saver.
In a GRID computing environment, this computer would be used for the calculation of a partial result, which has to be sent back to the GRID supervising computer for the assembly of the final result.
Apart from climate calculations, highly parallel computing solutions apply also to medical research, where a vast amount of possible combinations of molecular groups might be combined to achieve a therapeutic effect. In case of an epidemic event, the fast solution for an efficient medicament might save the life of millions of people.
How it is working
The main problem, the GRID computer is working on, is controlled by the dispatcher program, which gives away the problems and receives the solutions. This is an ordinary computer, which also should have some computational power, as it has to feed the millions of unused computers on the web in a reasonable time. Otherwise the several “crunching hungry machines” will do that what the GRID concept tries to avoid: eat energy without producing results. It is within the responsibility of the GRID hosting company to dispatch the right problems at the right time. For the users it is important to have a safe and trustful environment, avoiding that spy software, viruses and other varmints were smuggled in the grid environment.
The ways to participation
One of the most known GRID systems is hosted by IBM and is used for climate prediction as well as pharmaceutical research. Researchers might apply for inserting their problem into the GRID computing environment and get the ready assembled results back from the GRID computer itself. In order to prevent misuse of the GRID environment, researchers have to negotiate with the GRID dispatching society, IBM in this case. Other GRID environments are expected to develop, distributing their own GRID screen savers, asking the Internet community for their helping hands. It is important for both sides, the researchers asking for computing power and the clients, offering computing power, to be sure that no harmful, unwanted or illegal operation is supported.
Clients, who would like to offer their non-active time to the GRID community can register at IBM's GRID computing website. After registration, which makes sure that there is no misuse from the client side, a computer owner can download a screen saver program, which communicates with the dispatching computer. The screen saver programs are available for PC's running under Windows, Linux and Macs. After a successful installation, the program starts to contact the IBM's dispatching computer. A log file reports about the communication process itself and the progress of calculation.
In order to promote the idea of GRID computing environment, IBM has launched a website with a simulation of a GRID dispatching process. This website allows a user to work as a dispatcher, while random events are fed into the simulations news system. According to these events, the player has to arrange his dispatching priorities. The simulation gives a feedback, taking into account the chosen priorities. Following this approach, users can appreciate both the work of GRID computing and the importance of decision finding in the dispatching process.