Methods for developing and validating survivability distributions

A previous report explored and discussed statistical methods and procedures that may be applied to validate the survivability of a complex system of systems that cannot be tested as an entity. It described a methodology where Monte Carlo simulation was used to develop the system survivability distribution from the component distributions using a system model that registers the logical interactions of the components to perform system functions. This paper discusses methods that can be used to develop the required survivability distributions based upon three sources of knowledge. These are (1) available test results; (2) little or no available test data, but a good understanding of the physical laws and phenomena which can be applied by computer simulation; and (3) neither test data nor adequate knowledge of the physics are known, in which case, one must rely upon, and quantify, the judgement of experts. This paper describes the relationship between the confidence bounds that can be placed on survivability and the number of tests conducted. It discusses the procedure for developing system level survivability distributions from the distributions for lower levels of integration. It demonstrates application of these techniques by defining a communications network for a Hypothetical System Architecture. A logic model for the performance of this communications network is developed, as well as the survivability distributions for the nodes and links based on two alternate data sets, reflecting the effects of increased testing of all elements. It then shows how this additional testing could be optimized by concentrating only on those elements contained in the low-order fault sets which the methodology identifies.