Utilising cycle-accurate architectural simulation to aid in the design space exploration of future HPC hardware.

利用周期精确的架构仿真来帮助探索未来 HPC 硬件的设计空间。

• 批准号: 2445444
• 金额: --
• 依托单位: University of Bristol
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2445444
• 关键词: Utilising; cycle; accurate; architectural; simulation

The domain of high performance computing (HPC) aims to increase the feasibility of solving large, computationally-intense problems. Within this domain, high performance hardware is aggregated in such a way as to extract high levels of computational power unattainable by simple desktop computers; it is these approaches that enable HPC's aim. Both the techniques used to deliver high performance from computer architecture and, undoubtedly, the hardware itself need to be developed to maintain the performance advancement required by applications in HPC. To aid the development of high performance architectures, designers must be able to conduct studies to explore the design space and determine the impact of changes made. However, with the increasing complexity in HPC trends this space is expanding, which in turn, reduces the feasibility of these studies. There is a need to mitigate this complexity as to re-introduce the feasibility of potential gains.There is a wide range of applications seen to utilise the gains of HPC. Simulation and modelling are such applications and are of substantial importance to both modern science and engineering, they shift experimentation from the physical to the digital domain allowing for faster and more numerous investigation. ASiMoV, a EPSRC prosperity partnership, is a project that involves one such application with the focus on achieving "the world's first high fidelity simulation of a complete gas-turbine engine during operation''. As accurately stated by the project, "This level of simulation will require breakthroughs at all levels'' one of which being "Exascale HPC hardware''.The University of Bristol's HPC group is one of the contributors towards the ASiMoV project providing its academic expertise in advanced computer architecture to help solve the problems stated in ASiMoV. Amongst other contributions, they are currently developing an architectural simulator called the Simulation Engine (SimEng) framework to enable processor design space exploration.I aim to begin my research by aiding the University of Bristol's HPC group's contribution to the ASiMoV project through the continued development of SimEng whilst maintaining the focus on its primary goals, to be fast, accurate and easy to modify. I shall also be assessing whether the simulation framework could combat the complexity in HPC trends during exploration and whether it could identify architectural changes to support the advancement to Exascale Computing.SimEng is not the only architectural simulator. Many exist and in theory any of these could be used to address the aforementioned key issues in design space exploration. However, as stated in the ModSim 2019 presentation given by Simon McIntosh-Smith and with the focus on the ASiMoV project, the combination of speed and accuracy provided by these simulators aren't at the level needed for addressing exascale hardware design analysis. Thus, SimEng was devised to investigate these research questions.SimEng's involvement in the ASiMoV project offers the opportunity to work on motivated science use-cases namely. Beyond this, the challenges tackled within the project are repeated across many disciplines that are in need of simulation and modelling techniques to a similar fidelity, thus, the developments made have beneficial impacts outside the specific context of ASiMoV. Due to the University of Bristol's HPC group's strong industrial links, there are partnerships with ARM, Fujitsu and Marvell to explore future computer architecture design space aiding both the development of the simulation framework and the understanding of its impact within the industrial domain.The evolution of this simulation framework would help validate the feasible gains of possible architectural alterations whilst also enabling efficient and rigorous exploration with fast execution and flexible architectural modification respectively.This project falls within the EPSRC Information and communi

高性能计算（HPC）领域旨在提高解决大型计算密集型问题的可行性。在这个领域中，高性能硬件以这样一种方式聚合，以提取简单台式计算机无法达到的高水平计算能力;正是这些方法实现了HPC的目标。无论是用于从计算机体系结构提供高性能的技术，还是硬件本身，都需要进行开发，以保持HPC中应用程序所需的性能提升。为了帮助开发高性能架构，设计人员必须能够进行研究，以探索设计空间并确定所做更改的影响。然而，随着HPC趋势的日益复杂，这一空间正在扩大，这反过来又降低了这些研究的可行性。有必要减轻这种复杂性，以重新引入潜在收益的可行性。有广泛的应用程序可以利用HPC的收益。仿真和建模就是这样的应用，对现代科学和工程都非常重要，它们将实验从物理领域转移到数字领域，从而可以进行更快和更多的研究。ASiMoV是EPSRC的一个繁荣伙伴关系，是一个涉及这样一个应用程序的项目，重点是实现“世界上第一个完整的燃气涡轮发动机在运行过程中的高保真模拟”。正如该项目所准确指出的那样，“这种级别的模拟将需要在所有级别上取得突破”，其中之一是“Exascale HPC硬件”。布里斯托大学的HPC小组是ASiMoV项目的贡献者之一，提供其在先进计算机体系结构方面的学术专业知识，以帮助解决ASiMoV中提出的问题。除其他贡献外，他们目前正在开发一个架构模拟器称为模拟引擎（SimEng）框架，使处理器设计空间exploration.I的目标是开始我的研究，通过帮助大学的布里斯托的HPC组的贡献ASiMoV项目通过继续开发SimEng，同时保持专注于其主要目标，是快速，准确和易于修改。我还将评估模拟框架是否可以在探索过程中对抗HPC趋势的复杂性，以及它是否可以识别架构更改以支持向Exascale计算的推进。SimEng不是唯一的架构模拟器。理论上，其中任何一个都可以用来解决设计空间探索中的上述关键问题。然而，正如Simon McIntosh-Smith在ModSim 2019演示文稿中所述，这些模拟器提供的速度和准确性的组合并没有达到解决亿亿级硬件设计分析所需的水平。因此，SimEng被设计来调查这些研究问题。SimEng参与ASiMoV项目提供了一个机会，可以在有动机的科学用例上工作，即。除此之外，该项目中所解决的挑战在许多需要模拟和建模技术以达到类似保真度的学科中重复出现，因此，所做的开发在ASiMoV的特定背景之外产生了有益的影响。由于布里斯托大学的高性能计算小组与工业界有着密切的联系，因此与ARM建立了合作伙伴关系，富士通和马尔维尔将探索未来的计算机架构设计空间，帮助开发仿真框架，并了解其在工业领域的影响。该仿真框架的发展将有助于验证可能的架构更改的可行性，同时还可以实现高效和严格的本项目福尔斯属于EPSRC信息和通信项目，