Tech Area / Field. Brief Description of TechnologyInnovative fault-tolerant technology of distributed parallel data processing is offered

• INF-COM: Information and Communications / High Performance Computing and Networking

Brief Description of Technology
Innovative fault-tolerant technology of distributed parallel data processing is offered. It is based on an original conception of distributed programmable-structure computer systems (DCS).

Programmability of a DCS structure is understood as an automatic, changing possibility of virtual links between elementary machines (EM) for adapting DCS to structures and parameters of user problems solved under the condition of multiple failures of system resources.

Each EM consists of a processing element and distributed commutator section. The latter is connected with commutator sections of neighboring EM. Commutations between commutator sections form the DCS structure. The connection between commutator section poles may be changed by a program. It permits the user to adapt the DCS structure dynamically to the solved problem structure and to support the DCS acting under the condition of EM failures.

Organizing DCS functioning is based on the concept of a virtual subsystem (VS). This subsystem consists of virtual elementary machines (VEMs) and virtual links (VLs).

From the user's point of view, the VS is a unique resource. This resource is a functional equivalent of a system with a number of EM and a graph of intermachine connections that correspond to the user's task. Every real EM can be included into several virtual subsystems and can execute their functions in a multiprogram mode.

Hence, the described conception of functioning DCS systems provides:
(1) the formation of a virtual special-purpose DCS with a structure suitable for the user’s problem;
(2) mapping of the virtual DCS to the physical ubsystem as determined by the user;
(3) support of the special-purpose DCS structure under the conditions of changing resources distinguished for its realization.

The DCS structure programmability conception supposes the development of a full decentralized operating system (DOS) that provides the scalable DCS with a homogeneous structure and with robustness indices.

The DOS kernel is based on synchronization of primitives, which are the Deikstra semaphore generalizations. These generalizations let us extend the semaphore concept of structured data (list, queue, array), and of EM units (processor, memory, etc.). Minimizing the number of primitive types leads to the distributed operating system compactness.

DOS has the ability to work on any fault-free set of EM that are connected by fault-free links. DOS includes a distributed self-diagnosability component, which lets us determine the DCS technical state under the condition of multiple faultson the system level (the EM level).

Parallel languages were created by the introduction into the ordinary staff languages of the DOS kernel primitives implementing operations with generalized semaphores and processes.

The described means can serve as a basis for the development of the distributed means of recovery and the higher levels of DOS, including means for jobs and distributed files management.

Effectiveness indices were introduced to estimate the reliability and robustness of large-scale distributed computer systems. Analytic and computer-realized methods of reliability and robustness estimation were developed and tested.

Legal Aspects
DCS structure programmability properties were approved in some Russian multi-mini- and multi-microcomputer systems and networks (1970-1996), namely, in two generations of the MICROS family of multi-microprocessor systems (1982,1996). Now DOS components are under development for a robust multi-transputer DCS.

Special Facilities in Use and Their Specifications
None in this research.

Scientific Papers
Khoroshevsky V.G. Robustness of large-scale distributed computer systems // Proc. EUROSIM'96. HPCN Challenges in Telecomp and Telecom: Parallel Simulation of Complex Systems and Large-Scale Applications. Elsevier Science B.V., 1996, p.141-150.

Khoroshevsky V.G. Analysis of Functioning Complex Distributed Computer Systems // Systems Analysis: Modeling. Simulations. A journal of mathematical modeling and simulation in systems analysis. Gordon and Breach Science Publishers, 1995, v.18-19, p.673-675.

Khoroshevskii V.G., Dimitriev Yu.K., Zadorozhnyi A.F., and Tarkov M.S. Micros-2 Programmable-Structure Computing Systems // Engineering Simulation, Gordon and Breach Science Publishers SA 1995, v.13, p.85-104.

Khoroshevsky V.G. and Tarkov M.S. Computer systems with a programmable structure and parallel modeling. // Mathematical Modeling and Applied Mathematics. A.A. Samarsky and M.P. Spagovas, Eds. Elsevier Science Publ.B.V.(North-Holland), 1992, p.261-268.

Dimitriev Yu.K. Multiprocessor computer system self-diagnosis under local constraints // Optoelectronics, Instrumentation and Data Processing, 1996, no. 1, p.239-248.

Dimitriev Yu.K. A Decentralized Self-Diagnosis of Computer Systems with Programmable Structure // Proc. IFIP'93, p. 218-223.

Дата добавления: 2015-07-20