Parallel computing in heterogeneous machines based on the CPU donation approach

Abstract

Several years ago, Moore law became irrelevant, and increasing the computing power in a single machine became more and more complicated and not practically efficient. Nowadays, the research is focusing more on parallel computing in all of its types and forms. In this paper, we propose a new architecture of computing based on the HDCS (distributed heterogeneous computing system) or the worker approach, where the computing is done in several machines of different natures connected to the same network and which can be even extended to cover the Internet as well. These machines are supposed to be used for other purposes and they are exploited to do some computing only when they are idle. This approach might be used for computing types, where the handled task can be divided into several independent tasks. This approach offers lots of benefits, and it can almost be 100% free. In our proposed research work, we performed practical tests to exercise this computing method which was applied specifically to the preprocessing stage that helps to resolve any classification problem. The proposed algorithm was able to run on five separate machines, which are a raspberry PI embedded system, two phones and two laptops. The final decision was taken by one of the five machines and the obtained empirical results were motivating and very satisfactory. In addition, we demonstrated the ability of this scheme to be extended to any number of machines, so that we can build a very powerful machine for free, in the case of CPU donation.