XPS: EXPL: FP: Collaborative Research: SPANDAN: Scalable Parallel Algorithms for Network Dynamics Analysis

SPANDAN项目的目标是创建一个与体系结构无关的新型框架，用于设计高效、可移植和可扩展的并行算法，用于分析大规模动态网络。SPANDAN不仅将提供一种直观的方法来高效地将顺序算法转换为适用于动态网络的可扩展并行算法，而且还将提供分析评估机制，并作为应用程序和系统级调整之间的中介层。为了评估SPANDAN框架在现实世界应用中的有效性，PI将与社会科学家和生物学家合作。他们还将把研究成果整合到各种课程中，如网络分析、并行算法和生物信息学。他们将进一步与高中合作开发暑期课程，目标是鼓励女性和少数族裔学生从事与IT相关的职业。作为基本方法，SPANDAN框架将利用图稀疏技术将网络划分为稀疏子图(证书)，这些稀疏子图形成稀疏化树的叶子。这种创新的方法将导致设计和分析高效的并行算法来更新动态网络，并减少与并行化非结构化数据相关的内存延迟。具体来说，将设计并行算法来保持网络的拓扑特征，并更新有影响力的顶点和社区。为了展示可移植性和性能，开发的算法将在分布式内存集群、共享内存多核和大规模多线程Cray-XMT上实现。有关更多信息，请参阅项目网站：http://cs.mst.edu/labs/crewman/projects/SPANDAN/

The goal of SPANDAN project is to create a novel architecture-independent framework for designing efficient, portable and scalable parallel algorithms for analyzing large-scale dynamic networks. SPANDAN will not only provide an intuitive methodology for efficiently translating sequential algorithms into scalable parallel algorithms for dynamic networks, but also provide mechanisms for their analytical evaluation and serve as a mediatory layer between applications and system level tuning. To evaluate the effectiveness of SPANDAN framework in real-world applications, the PIs will collaborate with social scientists and biologists. They will also integrate research findings into various courses such as network analysis, parallel algorithms, and bioinformatics. They will further collaborate with high schools to develop summer courses with the goal of encouraging women and minority students to pursue IT-related careers. As the underlying methodology, the SPANDAN framework will exploit graph sparsification techniques to divide the network into sparse subgraphs (certificates) that form the leaves of a sparsification tree. This innovative approach will lead to the design and analysis of efficient parallel algorithms for updating dynamic networks, and reduction of memory latency associated with parallelizing unstructured data. Specifically parallel algorithms will be designed for maintaining network topological characteristics, and updating influential vertices and communities. To demonstrate portability and performance, the developed algorithms will be implemented on the distributed memory clusters, shared memory multicores, and massively multithreaded CRAY-XMT.For further information see the project web site at: http://cs.mst.edu/labs/crewman/projects/SPANDAN/