Collaborative Research: EAGER: Speeding-up large-scale simulations of atmospheric composition

该项目旨在提高GEOS-Chem 3D大气模拟软件的运行速度，并将华盛顿大学的计算机科学家以及麻省理工学院和华盛顿大学的大气科学家聚集在一起。GEOS-Chem模拟了地球大气化学成分的演变，是了解气候变化及其影响因素的重要工具。模拟是复杂的，加快模拟速度一直是密集努力的主题。特别是，为了利用“云”近乎无限的计算资源，可以将模拟分解成可以分配给不同处理器的独立的小任务。然后，挑战是决定如何最好地在处理器之间分配这些任务，以便尽可能快地完成模拟。这是这一提议的重点。GEOS-Chem使用地球的“立方体-球体”网格来模拟地球大气的演变，每个网格元素捕捉一个大气层柱，在那里复杂的化学模型模拟化学成分的演变。模拟以离散时间步长运行，相邻列在每个时间步长之后交换信息。柱子的计算成本各不相同，由于温差的原因，在日出和日落时达到峰值。当跨处理器分配列时，需要考虑这些变化，以确保负载尽可能均匀，因为不均匀的负载会显著降低模拟速度。设计实现这一目标的任务/列作业解决方案是本项目的主要智力重点。这映射到一个增加了复杂性的矢量调度问题，计算和通信都会影响结果。开发在整个计算环境中考虑这些因素的解决方案是该提案所针对的核心技术挑战。应对气候变化并了解其驱动因素是一个具有广泛社会影响的巨大挑战。快速模拟地球大气对其化学成分变化的反应的能力是解决这一问题的重要工具。该项目的预期贡献可以显著提高我们探索气候变化是如何产生的并制定缓解气候变化的解决方案的能力，这两者都可以产生重大的经济和社会影响。此外，该项目内在的跨学科性质将为该项目涉及的博士后提供独特的培训机会，并应使他们在内在的跨学科研究环境中取得成功。该项目的发现最初将在项目的GitHub网站上提供，网址为https://github.com/rochguerin/GEOS-Chem.然而，我们预计，随着投稿的成熟，特别是那些旨在增强地球观测系统-化学软件的投稿，许多投稿将迁移到位于https://geoschem.github.io/.This的地球观测-化学主要网站，该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project targets improving how fast the GEOS-Chem 3D atmospheric simulation software runs and brings together computer scientists from Washington University and atmospheric scientists from MIT and Washington University. GEOS-Chem models the evolution of the chemical composition of the earth atmosphere, and is an essential tool in understanding climate change and what affects it. The simulation is complex and speeding it up has been the subject of intensive efforts. In particular, to leverage the near limitless computational resources of the “cloud”, the simulation can be broken-up into small independent tasks that can be assigned to different processors. The challenge is then to decide how to best distribute those tasks across processors so that the simulation completes as fast as possible. This is the focus of this proposal.GEOS-Chem models the evolution of the earth’s atmosphere using a “cubed-sphere” grid of the globe, with each grid element capturing an atmospheric column where complex chemical models simulate the evolution of chemical composition. The simulation operates in discrete time steps with neighboring columns exchanging information after each time step. Columns’ computational costs vary, and peak at sunrise and sunset because of temperature differentials. These variations need to be accounted for when distributing columns across processors to ensure loads as uniform as possible, as uneven loads significantly slow-down the simulation. Designing task/column assignments solutions that realize this goal is the primary intellectual focus of this project. This maps to a vector scheduling problem with the added complexity that computations and communications both affect the outcome. Developing solutions that account for those factors across compute environments is a core technical challenge the proposal targets.Tackling climate change and understanding what drives it is an enormous challenge with broad societal implications. The ability to rapidly simulate how the earth atmosphere responds to changes in its chemical composition represents an essential tool in addressing it. The expected contributions of the project can significantly boost our ability to explore how climate change arises and develop solutions to mitigate it, both of which can have significant economic and societal impact. In addition, the inherently interdisciplinary nature of the effort will provide a unique training opportunity to the postdoc the project will involve, and should position them well to succeed in an inherently interdisciplinary research landscape.Findings from the project will initially be made available on the project’s GitHub site at https://github.com/rochguerin/GEOS-Chem. However, we anticipate that as, contributions mature, especially those targeting enhancements to the GEOS-Chem software, many will migrate to the main GEOS-Chem website at https://geoschem.github.io/.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.