Methods and means of reliability increasing and information stability of the multiprocessor systems for objects of critical assignment

Reliability improvement of multiprocessor systems (MS) designed for controlling sophisticated objects, in particular those of critical purpose, is achieved in a variety of ways, which are discussed in this work, by introducing different types of redundancy, in particular temporal and hardware redundancies, and most often – by the fault-tolerant design of the system. The minimum number of additional processors required to increase the reliability (probability of operation without failure) of the system to the desired level is estimated. A method that uses internal reserves of MS, does not require additional processors and provides an ability to increase the reliability of the system more smoothly without the jumps is introduced. One of the necessary conditions for the functioning of fault-tolerant MS’s is the ability to self-test, methods to optimize this procedure in terms of reducing the time for the bus and matrix connection topologies were developed. A new structurally regular topology and modification of the method of testing 2-fault-tolerant MS’s, the goal of which is to minimize the temporal redundancy is proposed. The resulting number of test checks (which generally defines the testing time) is 2n, which is less than for other known methods of testing MS (n is the total number of processors). To improve the efficiency of testing and reconfiguration procedures on different criteria in general case, a method and algorithm for allocation of processors into subsets is proposed, in particular for the reduction of temporal redundancy, which can be used to increase reliability. In the course of a fairly long operation as the processors fail, system can go into a dangerous condition. To detect such a situation and assess it’s probability, a modification of the authors-developed model of MS behavior in a fault stream is proposed.

To enhance informational robust being connected to negative human influence during design stage of MS, the methods of many-version processor units’ design had next development. Namely, the method of version’ creation on basis of mathematic models’ automorfism had been proposed. For practical using of such method the method of forming of many-digital direct and inverse replacements with linear complexity of soft- or hardware had been modified. The number of versions to implement the group operations on base of group carry’ realization which have wide value of the ratio cost/speed had been designed. Obtained results allow the development of method to design the versions of computer units with using of an arithmetic in residues.

Evaluation of dependability of the values of increment of true processor units’ functioning probability due to existence of autocorrelation under probability variation of determined wraps of input signals was carried out. It is founded that calculating increment of processor units’ true functioning by means of autocorrecting facilities has increasing mode into the range of technical based enhance of input signals’ the number of digits. Also it is showed that this increment can be compatible (from engineer point of view) with probability of faults which are due to wraps of input signals. It is founded analytically an equality of the results of calculations of functioning’ trustworthiness under existence of determined warps of input data for selected Boolean functions both without taking into consideration of the functions’ superposition and with existence of such superposition.

PDF icon 2014_2616-p.pdf204 KB