TaPS: A Performance Evaluation Suite for Task-based Execution Frameworks

任务执行框架，如并行编程库、计算工作流系统和函数即服务平台，使得将不同的任务组合成一个单一的应用程序，以达到计算目标。任务执行框架在任意硬件上抽象应用程序任务的并行执行。随着计算科学越来越需要利用并行计算和/或异构硬件，对这些任务执行器的研究也在加速。然而，缺乏评估标准使得比较和对比新系统与现有实现变得具有挑战性。在这里，我们介绍了TaPS，即任务性能套件，以支持并行任务执行框架的持续研究。TaPS提供了(1)一个统一的、模块化的接口，用于使用任意执行框架和数据管理系统编写和评估应用程序；(2)一个初始的参考合成和真实科学应用程序集。我们讨论了TaPS的设计如何支持可靠的框架评估，并通过使用提供的参考应用程序的基准测试调查来演示TaPS。

TaPS: A Task Performance Suite for Evaluating Parallel Task Executors

TaPS provides a unified, modular interface for evaluating task-based execution frameworks and an initial set of reference synthetic and real-world science applications.

TaPS supports the reliable evaluation of frameworks through a survey of benchmarks using the provided reference applications. The suite allows for the evaluation of arbitrary execution frameworks and data management systems. The paper also discusses the design of TaPS and its potential for future research.

Related work includes research into task-based execution frameworks and parallel computing, such as the MPI standard, OpenMP, and Apache Spark.