Mikhaylova T. Rise of the efficiency of the cluster computer systems on the basis of analytical models.

An object is the cluster systems of a different topology. Purpose of investigates is rise of clusters system efficiencies on the basis of analytical models results taking into account accordance with decided tasks class for the chosen topology of cluster. Researches methods are theory of accidental processes, calculable mathematics and graph's theories. Scientific results: contribution to the theory and practice of construction of the cluster systems and design of the difficult systems in different subject domains. The value of practical results consists in the use of the developed models for the analysis and synthesis of the highly productive computer systems with different topology descriptions of largeness. Scientific novelty of the got results: the discrete model of cluster with the divided disks for estimation of its works efficiency is designed; it is possible to get basic descriptions of calculable environment: loads of knots, quantity of requests in knots, times of implementation in each ofknots; the parallel algorithm of the discrete Markov's model of the homogeneous cluster on the mesh of processors is offered, characteristics of parallelism (speedup and efficiency) allowing realizing this algorithm on parallel structures are got; the comparative analysis of discrete and continuous Markov's model of the homogeneous cluster is conducted; the range of applicability for these two models providing necessary authenticity of the results is defined; the continuous Markov's model of the different cluster structures are developed, for the estimations of the efficiency of their work during realization on them for different classes of tasks; the modified algorithm of the decision of the task of the synthesis of the cluster structures for the closed systems with the using of the method of the middle, which allows to optimize calculable structures which network models have the quantity of the states, exceeding almost on two orders the quantity of the states of the models explored by classic methods is offered. The developed methods and models are used in the educational processes of the Donetsk national technical university, the Donetsk institute of artificial intelligence, in NVP "AMI" for computation of composition and structure of the cluster systems. Sphere of the applications: analysis and synthesis of the cluster and multiprocessor systems.