Real-time Optimization of Requirements Models

Early life cycle risk models can represent the requirements that a development group would want to achieve, the risks that could prevent these requirements from being met, and mitigations that could alleviate those risks. Our task is the selection of the least expensive set of mitigations that achieve the highest attainment of requirements. As these risk models grow larger, the demand for faster optimization methods also increases, particularly when those models are used by a large room of debating experts as part of rapid interactive dialogues. Hence, there is a pressing need for “real-time requirements optimization”; i.e. requirements optimizers that can offer advice before an expert’s attention wanders to other issues. One candidate technology for real-time requirements optimization is the KEYS2 search engine. KEYS2 uses a very simple (hence, very fast) novel Bayesian technique that identifies both the useful succinct sets of mitigations as well as costattainment tradeoffs for partial solutions. This paper reports experiments demonstrating that KEYS2 runs four orders of magnitude faster than our previous implementations and outperforms standard search algorithms including a classic stochastic search (simulated annealing), a state-of-the art local search (MaxWalkSat), and a standard graph search (A*).