Geekly Articles each Day
📜 ⬆️ ⬇️

Your personal distributed cluster or GPU: a Global Processing Unit

You've probably already heard about grid computing (grid computing), especially if you follow the development of any high-performance systems - now it is considered the most interesting and promising technology, the commercial rails of which are laid by Oracle, bringing it to the business environment for the first time. Basically, such computing farms (or grids, this term went from the power grid and in fact means that the calculations are not only distributed in space, but also accessible from any point in the required amount, as is electric energy). used for scientific calculations, visualization of complex scenes and other high-capacity calculations. By the way, do not think that this is the lot of only top-secret laboratories and large corporate customers. By no means! Using only open-source software, you yourself can build your own small computing grid network and thus use the combined resources for your needs. For example, it can be useful in educational institutions to integrate a fleet of computers of various classes into one network for conducting scientific calculations (after all, education often does not have funds for supercomputers, and if there is, then all the same tasks for scientists always require an order of magnitude more computing power. than there is).


The GPU (Global Processing Unit) project is designed to help use the resources of various computers, integrating them with the help of P2P technology into a single computing system. For this, the GPU uses the Gnutella network and protocol, working in it as a regular client, only sharing the CPU. While a typical GPU system includes a network of 5 to 15 computers (for the current 0.9x branch, the limit is 60 nodes in the network), which can be used for distributed video rendering, an experimental search engine is also available, which uses computers as search nodes for indexing and search for information on the Internet.

The GPU itself is distributed in open source codes and is an open framework for working with your applications or as an independent program. Of course, these are not libraries like MPI, and in order to use it, you will have to deal with the architecture and optimize the application, and the implementation language (ObjectPascal for the Delphi environment) imposes its own limitations, but if you aim to use this technology for your projects, you can do quite successfully.

')

Source: https://habr.com/ru/post/25224/

More articles:


All Articles