Practical Studies on Evaluation of Systems' Dependability Growth

摘要

Systems' Dependability has to be evaluated by unavailable time, which corresponds to service-stop time of users operation. Generally speaking, users want systems, which do not stop once they start the operation. We should design systems to automatically retry and correct errors to prevent service-stops. To meet such a high requirement, engineers should adopt redundancy and a multiplexing design philosophy to enhance systems' dependability. Once unmatched trouble has occurred on a multiplex systems, the systems must automatically make the safest choice (fail-safe), which is a harmless selection, and has been decided by the majority to the best, and it can avoid the service-stop from the results of fault. The results are evaluated by the decrease of total systems' unavailability. As for the product assurance operation, we adapted the graphical analysis with the time-serial plot on the log-log section paper; namely "dynamic evaluation" based on "logarithmic linear model". We can find the improvements and the reduction parameters from its slope graphically. This method is effective for improving operation dependability. This data is not only applicable to product assurance, but also useful for quotations in various fields on production planning and scheduling, for example man-hour planning, cost saving problem etc. From many observations, we know the "logarithmic linear model" fits, and is suited to analysis and predicts user satisfaction. When we discuss systems' dependability, systems' improvement and enhancement are very important. In this paper we will report several examples of the application to huge real-time running systems.