Development of a Means and Algorithm for Balancing the Load of Processors in a Reconfigurable Computing System

Abstract

The article is devoted to the development of a means and algorithm for balancing the load of processors of a reconfigurable computing system. A theoretical overview of the subject area of the research is given. Architectural solutions of modern analogues often do not allow scaling the system, which provides increased productivity and responsiveness. Modern solutions do not contribute to improving performance due to the use of shared memory and bus topology for switching processors, which impose additional conditions on shared memory access and synchronization of processor caches. The proposed approach will partially avoid the existing disadvantages of known solutions by using a cluster architecture to build a reconfigurable computing system, for the efficient operation of which all processors must be balanced. The issue of balancing the system load is not fully resolved, therefore, relevant. Further, the choice of the architecture of the reconfigurable computing system is justified, for which the synthesized algorithm of planning and dispatching is suitable. A description of the task planning algorithm for reconfigurable computing systems is given. The principles of interaction of processors in a reconfigurable computing system are described, taking into account the chosen architecture. The efficiency of the synthesized algorithm is analyzed using the example of test data. Using the input data, the load balancing algorithm identified various options for the distribution of tasks, one of which was marked as optimal, which is characterized by a minimum load spread of all processors of the reconfigurable computing system. At the end of the article, the main conclusions on the work are formulated.