MRI: Development of Heterogeneous Cluster for Cyber-Physical System Hybrid Analytics

拟议的文书为新的研究活动提供了基础，加强了现有的研究工作，并推动了机构内外的新合作。它包括在该州和该地区唯一有能力的高性能计算资源。 异构集群将被用来产生对地震事件的上级建模和理解，提高神经图像分析能力，增强网络安全工具，并激发新的车辆碰撞重建方法。教育活动增加了该机构的产品，并提供了一个模板的课程开发，以准备个人在异构计算领域的职业生涯。此外，通过使用集群及其分析工具，可以强化理论和应用课程，使学生能够更大规模地接近和研究现实世界的问题。本项目以异构集群（CPU/MIC协处理器/FPGA）的形式构建了一种新型计算资源，作为跨越科学和工程领域多条研究线的研究和创新平台。具体来说，这个实验性的混合集群旨在服务和推进具有不同计算需求的新兴网络物理系统的研究。这项工作解决了当前的计算问题，这些问题在规模和范围上不断增长，并在执行时间和成本方面迅速耗尽了传统同构架构的能力。虽然存在各种计算平台，包括传统的CPU、现场可编程门阵列（FPGA）、许多集成核心（MIC）协处理器和图形处理单元（GPU），但这些问题通常具有基于算法需求和底层硬件能力的工作分配的特点。因此，该开发将传统CPU与FPGA、MIC和GPU相结合，形成了一种能够在计算性能方面实现下一个进步的异构架构。该仪器允许调查人员通过计算机模拟和分析解决更大，更复杂的问题，涵盖网络物理系统安全，地震建模，神经信息学，碰撞重建和化学反应建模等领域。CPU擅长于具有大量控制（分支）指令的复杂多线程应用; MIC擅长于具有高空间局部性和符合SIMD模型的问题; FPGA提供了一个空白画布，可以进行配置以最好地满足特定问题的需求。目前的困难问题需要这些类型的设备的优势，这种仪器提供了一个平台来解决这些问题，并研究如何最好地使用所提出的异构平台。这些结果将推进这些过程的理论理解。异构集群将扩大在未来的异构集群和互连架构的设计领域的理解，在统一的编程平台和语言，并在不同的计算元素之间的工作负载分工。最后，异构集群将支持Honeynets科学的进步，用于安全分析，网络安全监控工具的实施，以及高级安全漏洞发现。

The proposed instrument provides a foundation for new research activities enhancing existing research efforts, and fuels new collaboration within and the beyond institution. It comprise a uniquely capable high performance computing resource in the state and in the region. The heterogeneous cluster will be leveraged to yield superior modeling and understanding of seismic events, improve neuro-image analysis capabilities, enhance cyber-security tools, and inspire new vehicle crash-reconstruction methodologies. The educational activities augment the offerings of the institution and provide a template for development of a curriculum to prepare individuals for careers in the field of heterogeneous computing. Further, access to the cluster and the analytic tools developed for it fortify theoretic and applied courses, allowing students to approach and study real-world problems of a larger scale.This project builds a novel computational resource in the form of a heterogeneous cluster (CPU/MIC co-processor /FPGA), as a platform for research and innovation spanning several lines of inquiry across multitude fields in science and engineering. Specifically, this experimental hybrid cluster aims to service and advance research in emerging cyber-physical systems that have diverse computational needs. This work addresses current computing problems that continue to grow in size and scope, and quickly exhaust the capabilities of traditional homogeneous architectures in terms of execution time and cost. While various computing platforms exist, including the traditional CPU, field programmable gate arrays (FPGAs), many integrated core (MIC) co-processors, and graphic processing units (GPUs), these problems often possess traits that would benefit from a distribution of work based on the algorithmic needs and the capabilities of the underlying hardware. Thus, this development combines traditional CPUs with FPGAs, MICs, and GPUs forming a heterogeneous architecture capable of delivering the next advance in computational performance. The instrument allows investigators to address larger and more complex problems through computer simulation and analysis spanning the areas of cyber-physical system security, seismic modeling, neuroinformatics, crash reconstruction, and chemical reaction modeling. CPUs excel at complex multi-threaded applications with significant numbers of control (branch) instructions; MICs excel at problems with high spatial locality and that fit the SIMD mold; and FPGAs provide a blank canvas that can be configured to best fit the needs of a particular problem. Current hard problems require the strengths of each of these types of devices and this instrument provides a platform to solve such problems and to research how best to use the proposed heterogeneous platform. These results will advance the theoretical understanding of these processes. The heterogeneous cluster will expand understanding in the areas of the design of future heterogeneous clusters and interconnect architecture, in unified programming platforms and languages, and in workload division among diverse computational elements. Finally, the heterogeneous cluster will support advances in the science of Honeynets for security analysis, the implementation of bump-in-the-wire cyber-security monitoring tools, and advanced security vulnerability discovery.