ITR: Distributed Hybrid Optimization Techniques with Applications to Proteomics and Genomics

许多实际感兴趣的计算问题可以在搜索方面提出。例子包括在科学、工程和商业的许多领域中自然出现的离散优化问题。强大的网络基础设施的发展加上价格合理的计算设备的出现，有助于将注意力集中在使用并行操作的多个处理器来提高搜索性能的技术上。我们的研究重点是可扩展的，容错的，分布式搜索技术及其应用程序的设计，开发，实施和评估，一般与其他优化技术，构建混合优化系统。这里描述的大部分工作都依赖于我们自己设计的一个独特的底层并行化范例，称为nagging。 与传统的问题划分技术不同，唠叨使一个问题的多个重构（或一个问题的多个部分）相互对抗，以提高性能。我们将改进和完善我们对唠叨的理解，包括研究结合唠叨和分区策略的自适应策略;改进和完善我们的分布式搜索基础设施，包括例如使用加密协议来支持跨管理边界的新应用程序的认证和分发;并继续我们的跨学科合作努力，将分布式搜索与其他优化技术一起应用于科学计算问题。

Many computational problems of practical interest can be posed in terms of search. Examples include discrete optimization problem that arise naturally in many areas of science, engineering and business. The development of a robust network infrastructure coupled with the advent of reasonably-priced computing equipment has helped focus attention on techniques that use multiple processors operating in parallel to improve search performance. Our research focuses on the design, development, implementation and evaluation of scalable, fault-tolerant, distributed search technique and their application, generally in concert with other optimization techniques, to construct hybrid optimization systems. Much of the work described here relies on a unique underlying parallelization paradigm of our own design called nagging. It contrast to more traditional problem partitioning techniques, nagging plays mutiple reformations of a problem --- or portions of a problem --- against each other to improve performance. We will improve and refine our understanding of nagging, including research on adaptive strategies for combining nagging and partitioning strategies; to improve and refine our distributed search infrastructure, including, for example, the use of cryptographic protocols to support authentication and distribution of new applications across management boundaries; and to continue our interdisciplinary collaborative efforts to apply distributed search concert with other optimization techniques to scientific computing problems.