CSR: Small: Towards Programming Datacenters

该项目旨在重新定义开发人员如何利用现代数据中心的计算能力。如今，开发人员必须指定如何在数据中心的多台计算机上分配每个计算作业。该项目将允许程序员仅为单台机器编写程序；然后，这些程序(通过该项目中开发的技术)转换为一种可以利用数据中心提供的丰富并行性的形式。因此，在这种方法中，开发人员不仅要对单个计算机进行编程，还要对整个数据中心进行编程。该项目利用了无服务器计算的最新进展，并通过解决几个基本问题来扩展其通用性，例如：-如何为现有和新兴的无服务器应用程序提供容错和改进的输入/输出性能？-当运行可能访问相同数据的无服务器应用程序时，如何有效地确保一致性？-是否可以自动翻译要在单个服务器上执行的代码，以便它可以在无服务器环境中运行，在该环境中，每次执行只允许在有限的时间内运行(在此之后终止)？无服务器计算是一种日益流行的模式，因为它允许应用程序开发人员部署应用程序，而无需管理单独的虚拟机。然而，当前的方法是不完整的，这影响了它的通用性(通过限制可以执行的应用程序)和正确性(因为正确实现应用程序需要开发人员对运行时和资源访问时间进行推理)。该项目将消除无服务器计算的这些限制，从而使更多的应用程序集能够部署为无服务器应用程序。这将使利用数据中心的力量变得更容易，从而在具有重大社会和商业重要性的无数领域取得进展。在这个项目中开发的工作将被放置在公共GitHub存储库中，并带有适当的文档，以便所有感兴趣的各方都可以使用这些技术。该奖项反映了NSF的法定使命，并已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 seeks to redefine how developers can exploit the computational power of modern datacenters. Today, developers must specify how each compute job is split across the many machines in a datacenter. This project will allow programmers to merely write programs intended for a single machine; these programs are then transformed (by the technologies developed in this project) into a form that can take advantage of the abundant parallelism available in datacenters. Thus, in this approach, developers are not just programming a single computer, but the entire datacenter.This project leverages the recent advances in serverless computing, and extends its generality by addressing several fundamental questions, such as:- How can one provide fault tolerance and improved Input/Output performance for both existing and emerging serverless applications?- How can one efficiently ensure consistency when running serverless applications which might access the same data? - Can one automatically translate code written to be executed on a single server so it can run in a serverless environment where each execution is only allowed to run for bounded time (after which it is terminated)? Serverless computing is an increasingly popular paradigm, since it allows application developers to deploy applications without managing individual virtual machines. However, the current approach is incomplete which both impacts its generality (by limiting applications that can be executed) and correctness (since correctly implementing applications requires developers to reason about runtimes and resource access times). This project will eliminate these limitations of serverless computing, thus enabling a larger set of applications to be deployed as serverless applications. This will make it easier to harness the power of datacenters, thereby enabling progress in myriad areas of significant social and commercial importance. The work developed in this project will be placed on a public github repository, with the appropriate documentation so that the technologies are available to all interested parties.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.