Experimental Software Systems: Software Foundations for Experiment Management

[9870684] miller, Barton P.Livny， miron威斯康星大学麦迪逊分校实验软件系统：实验管理的软件基础高性能并行和网络化系统的发展是一个渐进的过程。系统从模型和模拟开始，并通过一系列实现步骤进行。在每个步骤中，关键问题是性能如何变化以及变化了多少？当我们将实现与模型或模拟进行比较时，以及当我们更改实现以使用不同的算法、库、语言、处理器数量、网络类型、输入、工作负载或调度算法时，这个问题很重要。该项目正在开发一个实验管理设备，称为X-Man，以提供直接攻击这些问题的基础设施。系统开发人员以及开发模型和仿真环境的人员将立即使用该基础结构。它将为协作开发小组提供一个交换、比较和参考点，支持协作的、地理上分布的开发。X-Man的好处包括：(1)直接支持代码开发过程，为程序员提供对程序演变的具体理解。(2)将基准测试从一项特别的活动转变为一项具有定量的、定义良好的结构的活动。(3)提供基础设施和分析，以支持模型和仿真验证。(4)动态系统效果评价框架，分析自适应算法导致的行为差异。

9870684Miller, Barton P.Livny, MironUniversity of Wisconsin-MadisonExperimental Software Systems: Software Fondations for Experiment ManagementThe development of a high-performance parallel and networked system is an evolutionary process. Systems start with models and simulations and progress through a sequence of implementation steps. At each step, the key question is how and how much did the performance change? This question is important when we are comparing implementation to models or simulations, and when we change an implementation to run with a different algorithm, library, language, number of processors, type of network, input, work load, or scheduling algorithm. This project is developing an Experiment Management facility, called X-Man, to provide an infrastructure to directly attack these issues. This infrastructure will be of immediate use by system developers, and those developing models and simulation environments. It will provide a point of exchange, comparison, and reference for cooperating development groups, supporting collaborative, geographically distributed development. The benefits from X-Man include: (1) Directly support the code development process, providing programmers with concrete understanding of program evolution. (2) Change benchmarking from an ad- hoc activity to one with a quantitative, well-defined structure. (3) Provide both infrastructure and analysis to support model and simulation validation. (4) A framework to evaluate the effect of dynamic systems, analyzing behavioral differences caused by adaptive algorithms.