CSR: Small: Accelerated ParalleX (APX) for Enhanced Scaling AMR based Science

CSR：小型：Accelerated ParalleX (APX)，用于增强扩展基于 AMR 的科学

• 批准号: 1117470
• 负责人: Hartmut Kaiser
• 金额: $ 42.49万
• 依托单位: Louisiana State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2011
• 资助国家: 美国
• 起止时间: 2011-09-01 至 2015-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1117470&HistoricalAwards=false
• 关键词: CSR; Small; Accelerated; ParalleX; APX

Intellectual Merit:This project addresses scalability of applications on Exascale parallel computer architectures. Among the strategic challenges to scalability is identifying and exploiting new forms of parallelism as well as reducing overhead for effective fine grained parallelism execution. The project, "Accelerated ParalleX (APX) for Enhanced Scaling" project investigates combining a new model of computation with FPGA hardware support, in otherwise conventional multicore platforms, to realize significant gains in scalability. This approach is particularly applicable to the important class of adaptive mesh refinement based applications for colliding neutron stars and gamma ray bursts. Prior NSF funding supported the development of an experimental ParalleX prototype. That model is employed in this new research project. This framework extracts inherent parallelism implicit in structure meta-data, eliminates most global barriers, and releases adaptive control to overlap multiple phases of computation and intermediate communication for latency hiding and circumvention of contention hot spots. The research investigates the use of FPGA based hardware technology for accelerating system software to significantly reduce critical time path overhead in execution and directly enhance scalability. The experimental designs include synchronization atomics, thread scheduling and queues, and active message driven operations. In addition, for this class of science problems, higher precision floating-point arithmetic is becoming more important for such science questions as resolving the smallest possible black holes. FPGA technology will therefore accelerate multi-precision floating point arithmetic. This research, if successful, will advance the specific science domain of numerical relativity and, more broadly, those science and engineering disciplines relying on both AMR and strong scaling. It will advance near-term computer system science through an innovative application of available FPGA technology to general computational science and long-term future scalable system design. The performance model derived for this purpose may prove valuable for extended preliminary exploratory investigation for establishing bounds and sensitivities in a complex multi-faceted trade-off space. Broader Impact: The APX research results and resources will be applied to the distance-learning course distributed live to other national and international campuses to expand its content and extend its advanced topics section, in the short term, while motivating a new graduate level seminar course next year around its topic areas. Summer internships for under-represented undergraduate and high school students will be created at LSU and the Beowulf Bootcamp will be expanded for more high school students.

智力优势：该项目解决了Exascale并行计算机架构上应用程序的可扩展性。 可伸缩性的战略挑战之一是识别和利用新形式的并行性，以及减少有效的细粒度并行执行的开销。该项目名为“Accelerated ParalleX（APX）for Enhanced Scaling”，旨在研究在传统的多核平台中将新的计算模型与FPGA硬件支持相结合，以实现可扩展性的显著提高。这种方法特别适用于一类重要的自适应网格细化碰撞中子星和伽马射线爆发的应用程序。之前的NSF资金支持了实验性ParalleX原型的开发。这个新的研究项目采用了那个模型。该框架提取隐含在结构元数据中的固有并行性，消除大多数全局障碍，并释放自适应控制以重叠计算和中间通信的多个阶段，用于延迟隐藏和规避竞争热点。本研究探讨使用FPGA为基础的硬件技术加速系统软件，以显着减少关键时间路径开销的执行，并直接提高可扩展性。实验设计包括同步原子，线程调度和队列，以及主动消息驱动操作。此外，对于这类科学问题，更高精度的浮点运算对于解决尽可能小的黑洞等科学问题变得越来越重要。因此，FPGA技术将加速多精度浮点运算。这项研究如果成功，将推进数值相对论的特定科学领域，以及更广泛地说，那些依赖AMR和强尺度的科学和工程学科。它将通过将现有的FPGA技术创新应用于一般计算科学和长期的未来可扩展系统设计来推进近期的计算机系统科学。为此目的而得出的性能模型可能被证明是有价值的扩展初步探索性调查，建立在一个复杂的多方面的权衡空间的界限和敏感性。更广泛的影响：APX的研究成果和资源将应用于远程学习课程，并在短期内向其他国家和国际校园现场分发，以扩大其内容并扩展其高级主题部分，同时激励明年围绕其主题领域开设新的研究生级别研讨会课程。路易斯安那州立大学将为代表性不足的本科生和高中生提供暑期实习机会，贝奥武夫训练营将扩大到更多的高中生。