METHOD OF DECREASING NUMBER OF MUTUAL TESTING DURING SELF-TESTING OF MULTIPROCESSOR SYSTEMS

Abstract

The subject matter of this article is the testing processes in multiprocessor systems, first of all - fault-tolerant control systems, when the number m of allowed failures from n processors is known in advance and is limited. The topology of interprocessor communications can be represented as a directed graph of the circulant type, where the number of input and output channels between each one processor and the other processors in the system is not less than m. The goal is to minimize the number of mutual checks in the system when performing its self-testing. Tasks: develop an effective method for organizing the mutual testing of processors in multiprocessor systems and algorithm for its implementation, estimate the number of elementary checks and prove its rightness. The following results were obtained. Based on the known method of diagnosis of multiprocessor systems with a regular structure, and m≤4, with parallel and independent testing, the task of organizing of testing process for any m <(n ⁄ 2) was formulated. In this process, only one pair of processors participate in testing in one moment of time, thus it was called sequential. what special about this organization is that the choice of the next pair is based on the analysis of the history of the process. An algorithm for performing the method has been proposed. The Preparata-Metz-Chen model was chosen as the model of faults, as the closest one to reality. Conclusions. The scientific novelty of the results is as follows: The method for organizing of self-testing of multiprocessor systems with a connection topology described by a circulant graph (with at least m input and m output edges) has been proposed, and this method allows reducing the number of mutual tests in the system. It has been proved that the state (serviceable/faulty) of all processors of the system can be determined after performing no more than n + 2m checks. Practical value - the method allows to reduce the time spent by system on self-testing, and, therefore, to increase its performance, since the system performs this task constantly in the process of exploitation. The advantages of the selected topology: it is suitable for any integer values of n

Keywords

• многопроцессорная система
• самотестирование
• диагностический граф
• циркулянт