Cognitive Aspects of Parallel Programming

并行计算包括通过将计算问题分解成几个部分并同时计算这些部分来在更短的时间内解决计算问题。并行处理硬件所需的硬件继续快速发展，使得每秒的理论运算次数不断增加。然而，众所周知，并行编程系统没有跟上硬件的进步：虽然已经提出了数百种并行语言，但除了语言开发人员本身之外，很少有语言占据主导地位。这样做的一个原因是可用性--即开发和维护既正确又达到所需性能级别的并行程序的容易程度。目前关于并行编程可用性的知识体系非常有限，无论是在贡献的数量上还是在研究的方法上。现有研究的一个主要局限性是它们没有建立在一个理论框架上。这意味着，尽管他们可能能够在特定的实验环境中确定一种语言是否比另一种语言更好，但很难将结果应用到更广泛的领域。编写程序涉及的不仅仅是句法和语义知识。它还包括图式知识，它由一般性的解决方案组成，正如对程序设计的认知过程的研究所显示的那样。编写程序也与理解程序密切相关，因为模式可以通过阅读程序来学习，而实际编程通常涉及修改现有程序。理解程序需要读者构建一个心理模型。这些程序设计和理解理论已被应用于过程性和面向对象程序设计，但不适用于并行程序设计。该建议的长期研究目标是发现并行程序员使用的认知模型和过程，以及这些模型和过程如何依赖于编程语言。这项工作将扩展目前关于编程心理学的知识体系，这些知识只应用于顺序编程。这项工作将在评估并行程序设计语言的可用性、设计支持工具和并行程序设计的教学中具有重要的应用。这项拟议的研究建立在我的研究生教科书《并行计算的元素》的基础上，这本书由查普曼和霍尔/CRC出版社于2017年在他们的计算科学系列中出版。本书介绍了实际上与并行编程相关的算法和实现结构。这种结构经历了并行硬件和软件的许多变化。这项拟议的工作将调查这些结构如何形成并行程序员在编写和理解程序时使用的心理模型的一部分。

Parallel computing involves solving a computing problem in less time by breaking it down in parts and computing those parts simultaneously. The hardware required for parallel processing hardware continues to progress rapidly, enabling ever-increasing theoretical operation counts per second. However, it is well known that parallel programming systems haven't kept up with the advances in hardware: while hundreds of parallel languages have been proposed, very few have taken hold beyond the language developers themselves. One reason for this is usability - that is the degree of ease with which one can develop and maintain parallel programs that are both correct and reach the desired level of performance. The current body of knowledge of parallel programming usability is very limited, both in the number of contributions and the methodology of the studies. A major limitation of existing studies is that they are not based on a theoretical framework. This means that while they may be able to determine if one language is better than another in a particular experimental context, it is difficult to apply the results more generally. Writing a program involves more than just syntactic and semantic knowledge. It also involves schematic knowledge, which consists of generic solutions, as shown in studies of the cognitive processes of program design. Writing programs is also intimately tied to understanding programs, since schemas can be learned by reading programs, and practical programming often involves modifying existing programs. Program understanding requires the reader to construct a mental model. These theories of program design and understanding have been applied to procedural and object oriented programming but not to parallel programming.The long-term research goal of this proposal is to discover the cognitive models and processes used by parallel programmers, and how these depend on the programming language. This work will extend the current body of knowledge on the psychology of programming, which has only been applied to sequential programming. This work will have important applications to assessing the usability of parallel programming languages, designing supporting tools, and the teaching of parallel programming. The proposed research builds on my graduate textbook, Elements of Parallel Computing, published in 2017 by Chapman and Hall/CRC Press, in their Computational Science series. This book presents the algorithmic and implementation structures that are practically relevant for parallel programming. Such structures have endured through many changes in parallel hardware and software. The proposed work will investigate how these structures form part of the mental models that parallel programmers use when writing and understanding programs.