SI2-SSE: Collaborative Research: ADAPT: Next Generation Message Passing Interface (MPI) Library - Open MPI

高性能计算在许多领域重塑了科学和工业。然而，在过去几年中，硬件级别的快速发展与编程范式级别的相应变化是无法比拟的。根据几项主要研究的共识，预计大型系统的并行度将增加几个数量级。因此，作为事实上的标准消息传递范例的消息传递接口（MPI）缺乏处理当今体系结构的有效和可移植的方法。为了有效地处理这样的系统，MPI实现必须采用更多的异步和线程友好行为，以获得比现在更好的性能。s系统。MPI是高性能计算中使用最广泛的通信库之一，维护和进一步增强MPI将对科学界以外的领域产生深远的影响，并代表着所有科学和工程领域持续进步的关键组成部分。ADAPT项目在处理器架构和系统设计不断变革的背景下，增强、强化和现代化了Open MPI库。它为新一代Open MPI组件创建了一个可行的基础，支持快速探索新的物理功能，提供大大改进的性能可移植性，并致力于组件类之间的完全互操作性。更具体地说，ADAPT实现了可用于多核心系统的基本软件技术，以有效地执行基于mpi的应用程序，并容忍从当前的大型系统到即将到来的极端规模系统的故障停止过程故障。为了提高Open MPI的效率，ADAPT作为核心组件集成了有关硬件体系结构的知识，并允许软件堆栈的所有层完全访问这些信息。然后，进程放置、分布式拓扑、文件访问、点对点和集体通信可以适应这样的拓扑信息，从而提供更多的可移植性。ADAPT团队还在更新当前的集体通信层，以允许包含在组级别的基于任务的集体描述，从而调整到节点内和节点间拓扑结构。具有弹性功能的当前代码的计划扩展使Open MPI能够有效地应对硬错误和软错误类型的故障。这些功能可以用作MPI标准体中所有当前活动的容错建议的构建块。MPI已经是最相关的并行编程模型之一，是大多数并行应用程序中最重要的组成部分，也是大多数其他编程模型中最关键的通信部分之一。因此，研究团队的经验和新出现的能力可以使这些编程标准、工具和库的所有未来用户受益——不管规程是什么。主要MPI库（如Open MPI）的性能和功能的任何改进都有可能对应用程序社区产生直接而显著的影响。除了缩短应用程序到解决方案的时间外，它还有可能减少能源使用并最大限度地提高现有执行平台提供的性能。开放MPI库在政府研究机构（包括大学和国家实验室）以及私营部门的使用规模，是对所有科学和工程社区迅速产生影响的主要载体。

High-performance computing has reshaped science and industry in many areas. However, the rapid evolution at the hardware level over the last few years have been unmatched by corresponding changes at the programming paradigm level. According to the consensus of several major studies, the degree of parallelism on large systems is expected to increase by several orders of magnitude. As a result, the Message Passing Interface (MPI), which has been the de-facto standard message passing paradigm, lacks an efficient and portable way of handling today's architectures. To efficiently handle such systems, MPI implementations must adopt more asynchronous and thread-friendly behaviors to perform better than they do on today?s systems. Maintaining and further enhancing MPI, one of the most widely-used communication libraries in high-performance computing, will have a far-reaching impact beyond the scientific community, and represents a critical building block for continued advances in all areas of science and engineering.The ADAPT project enhances, hardens and modernizes the Open MPI library in the context of this ongoing revolution in processor architecture and system design. It creates a viable foundation for a new generation of Open MPI components, enabling the rapid exploration of new physical capabilities, providing greatly improved performance portability, and working toward full interoperability between classes of components. More specifically, ADAPT implements fundamental software techniques that can be used in many-core systems to efficiently execute MPI-based applications and to tolerate fail-stop process failures, at scales ranging from current large systems to the extreme scale systems that are coming soon. To improve the efficiency of Open MPI, ADAPT integrates, as a core component, knowledge about the hardware architecture, and allows all layers of the software stack full access to this information. Process placement, distributed topologies, file accesses, point-to-point and collective communications can then adapt to such topological information, providing more portability. The ADAPT team is also updating the current collective communication layer to allow for a task-based collective description contained at a group-level, which in turn adjusts to the intra and inter-node topology. Planned expansion of the current code with resilient capabilities allows Open MPI to efficiently survive hard and soft error types of failures. These capabilities can be used as building blocks for all currently active fault tolerance proposals in the MPI standard body.MPI is already one of the most relevant parallel programming models, the most important brick of most parallel applications, and one of the most critical communication pieces of most other programing models. Thus, the experience of the research team and emerging capabilities can benefit all future users of these programming standards, tools, and libraries--regardless of discipline. Any improvement in the performance and capabilities of a major MPI library such as Open MPI, has tremendous potential for an immediate and dramatic impact on the application communities. In addition to improving the time to solution for their applications, it has the potential to decrease the energy usage and maximize the performance delivered by the existing execution platforms. The scale at which the Open MPI library is used in government research institutions (including universities and national laboratories), as well as in the private sector, is a major vector for a quick impact on all scientific and engineering communities.