Abstract
Modern computing systems typically utilize a large number of computing nodes to perform coordinated computations in parallel or simultaneously. They can exhibit multiple performance states or levels due to statuses or failures of their consistent nodes. Performability analysis is concerned with assessing the probability that the computing system performs at a particular performance level. In the context of performability analysis, these computing systems can be modeled using k-to-l-out-of-n structures. This paper proposes new analytical methods based on binary decision diagrams (BDD) for the performability analysis of large computing systems with unrepairable computing nodes. A new and efficient BDD algorithm that makes full uses of the special k-to-l-out-of-n structure is first proposed for systems with computing node having identical computing powers. New simplification rules are further proposed to generate compact and canonical BDD models for systems with heterogeneous computing nodes characterized by different computing powers. Ordering heuristic is also explored to further reduce the size of BDD models. Examples are provided to illustrate the proposed BDD-based performability analysis methodology as well as its efficiency in analyzing large-scale computing systems.