Approach to calculation of characteristics of reliability of the cluster computing system from two interchangeable servers is offered. The purpose of article consists in the increase in accuracy of calculation of characteristics of reliability of cluster system reached by introduction of premises about not exponential distribution of duration of trouble-free operation and restoration of servers.
The semi-Markov model of such system in the form of two conditionally independent accidental processes of restoration is described. For accounting not of the exponential law of distributions of duration of trouble-free operation and restoration of servers of a cluster two-dimensional process until a double failure is considered. In case of any start state when one server works, and another – is not present, the moment of a double failure comes when both servers are in repair state. As an index of reliability of a computing cluster the average time of trouble-free operation of a cluster determined through probability to find cluster system in working order in arbitrary time point is used. Use of analytical expressions for calculation of average time of trouble-free operation of a computing cluster from two servers, one of which is in a hot reserve, is justified. At the same time important distinctive feature of the considered semi-Markov model of reliability of a computing cluster is that duration of restoration of the failed server can have arbitrary distribution law. For specification of characteristics of reliability of the cluster computing system the possibility of use of work benches of monitoring is discussed. Results of numerical calculations of characteristics of reliability of the cluster computing system are given. The main result – the analytical impact assessment of coefficient of a variation not of exponential distribution of duration of restoration of the server in the used reliability model on value of average time of trouble-free operation of a cluster showing as far as the accuracy of calculation of this index of reliability of the cluster computing system from two interchangeable servers increases. Feasibility of specification of characteristics of reliability of the cluster computing system by means of work benches of monitoring is justified.