Using the Dist Simulator to Teach Parallel Computing Concepts
Using the Dist Simulator to Teach Parallel Computing Concepts
复制标题
使用 Dist 模拟器教授并行计算概念
DOI：
--
复制
发表时间：
--
期刊：
影响因子：
--
通讯作者：
A. Pears
中科院分区：
文献类型：
--
作者：
A. Pears
本文介绍了多机模拟器(DIST)的特点，并讨论了它作为教学和研究工具在灵活性和成本方面的优势。DIST通过提供复杂的编译器前端支持两种主要的编程模型(共享和分布式内存)。语言功能提供对硬件同步指令(TSET、TCLR)的用户级访问，并为并行执行的生成、重定位和协调提供高级设施[4]。DIST还允许学生和研究人员调查各种体系结构参数对软件性能的影响。选项包括高速缓存线大小、高速缓存大小、高速缓存一致性方案、网络拓扑、网络和CPU的相对时钟速度、无效目录大小以及共享或分布式内存。还提供了分布式共享内存体系结构模型。使用DIST进行教学使我们能够采用更加亲身实践的方式进行学习。这导致了课程方法的改变，将重点放在从实践经验中学习。讲课集中于提供技术，帮助解决学生在精心挑选的一系列实践练习中遇到的问题。结果是增加了学生的竞争和自我激励。在干扰、数据分区、负载平衡、竞争条件等问题上的直接经验有助于全班将注意力集中在课程中提出的解决方案上，并加强他们的同化。此外，由于他们的经验，学生们对他们解决并行编程问题的能力也有了更多的信任。与之前开设的理论课程相比，学生们对这门课程的态度也更加积极。
This paper presents the features of a multicomputer simulator (DiST))5, 9] and discusses its advantages in terms of exibility and cost as a teaching and research tool. DiST supports two major programming models (shared and distributed memory) by providing a sophisticated compiler front end. Language features provide user level access to hardware synchroni-sation instructions (TSET, TCLR), as well as providing high level facilities for the generation, relocation, and coordination of parallel executionn4]. DiST also allows students and researchers to investigate the software performance implications of a wide range of architectural parameters. Choices include, cache-line size, cache size, cache consistency scheme, network topology, relative clock speeds for network and cpu, invalidation directory size, and shared or distributed memory. A distributed shared memory architecture model is also availablee11]. Teaching using DiST allows us to adopt a much more hands-on approach to learning. This has resulted in a change of approach within the course to focus on learning through practical experience. Lectures concentrate on providing techniques which help to solve problems that students encounter in a carefully chosen series of practical exercises. 1 The result is increased student competition and self motivation. Direct experience with problems like interference, data partitioning, load balancing, race conditions etc helps to focus the class on the solutions presented in the lectures and reinforce their assimilation. In addition students have increased conndence in their ability to solve parallel programming problems as a result of their experiences. Students also feel more positive about the course, in comparison to the more theoretical course they were previously undertaking.