Dominion: A Distributed System for Large Scale Decision Support Applications

经济压力和全球竞争正迫使美国制造商和政府以尽可能快的速度和成本效益做出决定。这些决策问题包括设施选址、人力和物资运输、作战战略分析、全球协调制造和分子设计。Advanced Process Combinatorics建议将分布式决策支持系统Dominion商业化，它将使用严格的优化原则而不是目前实践中基于规则或其他启发式方法进行决策，从而实现最具成本效益的决策。此外，尽管目前大型决策问题需要昂贵的超级计算机来利用严格的优化技术，但Dominion将大量为成本低得多的平台(桌面工作站)开发。这将允许没有较大规模计算机的组织转而利用其当前安装的工作站和网络。它进一步消除了大型时间关键型计算中的单点故障。Dominion将通过将可并行优化问题分布在一个异类的、全球分布的工作站网络中来运行。分发中的每个计算代理完全能够解决整体问题的一部分，并将其工作的一部分分派到其他联网代理和从其他联网代理接收。与当前流行的主从模型相比，该模型可以很容易地扩展到非常大的网络，并且对部分网络故障具有明显的健壮性。决策支持软件的销售额很容易达到每年数亿美元。Dominion将用新的功能增强当前的商业产品，并允许以低得多的成本优化更大的决策问题。

Economic pressures and global competition are forcing US manufacturers and government to make decisions as quickly and cost effectively as possible. Such decision problems include locating facilities, transporting manpower and materials , battle strategy analysis, globally coordinating manufacturing, and molecular design. Advanced Process Combinatorics proposes commercialization of the distributed decision support system DOMINION, which will make decisions using rigorous optimization principles rather than rule-based or other heuristic approaches currently in practice, allowing for the most cost effective decisions possible. Furthermore, although large decision problems currently require expensive supercomputers to exploit rigorous optimization technology, DOMINION will be developed for much lower cost platforms (desktop workstations) in large number. This will allow organizations without larger scale computers to utilize their currently installed workstations and networks instead. It further removes the single point of failure fault from large, time critical computations. DOMINION will operate by distributing a parallelizable optimization problem across a heterogeneous, globally distributed network of workstations. Each computational agent in the distribution is fully capable of solving portions of the overall problem and dispatching portions of its work to and receiving from other networked agents. This model, in contrast to the currently popular master-slave models, is easily scalable to very large networks, and is significantly more robust to partial network failure. Sales of decision support software easily range in the hundreds of millions of dollars per year. DOMINION will augment the current commerical offerings with new capabilities, and allow optimization of larger decision problems at a much lower cost.