Collaborative Research: PPoSS: Planning: Fixpoint: an operating system and architecture for data-centric computing

计算机操作系统运行读取输入数据和产生输出数据的程序。然而，今天的操作系统通常不会跟踪足够的信息来确保生成给定数据的过程可以重现，特别是当生成数据的程序可能已经考虑了来自网络、来自磁盘上的可变文件或来自诸如线程调度之类的不确定现象的输入时。这个项目将设计一种新的操作系统，名为Fixpoint，它明确地表示和命名对数据的计算：根据程序对数据的最小依赖，每次调用都以可复制的内容寻址方式进行。如果成功，该项目将对计算机系统的日常使用方式产生重大影响。通过使所有计算在默认情况下都是可重现的，并且可以很容易地共享，Fixpoint可以提高科学重现性，从而增加公众对科学结果的信心。这个项目的假设是，通过改变软件的表示方式，可以在可伸缩性、安全性、效率、性能和重复性方面释放出实质性的好处。Fixpoint的操作系统可见数据流将使其能够利用大规模瞬时并行性，这意味着目前有效地需要批处理的并行任务--向服务器提交作业和等待几个小时--将变得接近实时。这将改变人们期望在计算机上交互进行的各种操作。在数据中发现的错误，即使是在事实发生后，也可以在依赖它们的计算中得到支持，类似于今天重新计算电子表格。这一裁决反映了NSF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。

Computer operating systems run programs that read input data and produce output data. However, today's operating systems generally don't keep track of enough information to make sure that the process of generating a given piece of data can be reproduced, especially when the program that produced it may have considered input that came over the network, from a mutable file on disk, or from a non-deterministic phenomenon such as thread scheduling. This project will design a new kind of operating system, called Fixpoint, that explicitly represents and names computations on data: each invocation of a program, in terms of its minimal data-dependencies, in a reproducible content-addressed manner. If successful, the project will have a significant impact on how computer systems are used every day. By making all computation reproducible by default and trivially shareable, Fixpoint can improve scientific reproducibility and as a consequence increase the public's confidence in scientific results. The hypothesis of this project is that by changing the way software is represented, substantial benefits can be unlocked in the areas of scalability, security, efficiency, performance, and reproducibility. Fixpoint's operating-system-visible dataflow will give it an ability to take advantage of massive transient parallelism, which means that parallel tasks that effectively require batch-processing today -- submitting jobs to servers and waiting hours -- will become near real-time. This will change the kinds of operations that people expect to do interactively at a computer. Errors discovered in data, even after the fact, can be backed out of computations that depend on them, similar to recalculating a spreadsheet today.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.