Evaluation and Testing Framework for Cloud and Edge Computing Applications

当今最先进的应用程序通常部署在云基础设施中，利用成本优势、弹性可伸缩性和无限资源的假象。虽然这满足了许多应用领域的需求，比如企业web应用程序，但它可能不适合现代基于物联网（IoT）的应用程序。物联网应用，如自动驾驶或可穿戴健康监测应用，会发出大量数据流，需要以近乎实时的方式进行处理。由于网络延迟，将数据传输到云端进行处理给对延迟敏感的应用程序带来了挑战。为了解决这个问题，可以将处理转移到产生数据的网络边缘。然而，边缘设备通常具有有限的计算能力。因此，需要同时使用边缘和云基础设施来实现低延迟，同时访问可扩展和无限的资源。应用程序开发人员需要对他们的应用程序进行分区，并根据他们对底层基础设施的需求在边缘和云设备的频谱上分发它们。这就需要根据功能和性能评估许多不同的可能配置，并选择满足应用程序需求的配置。我的研究计划旨在提供一组工具和技术，用于评估不同分布策略的性能含义，以及测试分区应用程序的功能。我的长期目标是为分布式数据处理应用程序的可靠和安全部署设计解决方案，以便以具有成本效益和能源意识的方式利用计算资源进行实时决策。该计划的第一个主题是自动化集成测试，重点是将应用程序划分为较小的模块，并在给定的异构设备上测试分区应用程序的集成。这个主题的短期目标是1)设计有效的应用程序分区策略和2)自动生成集成测试脚本。本课程的第二个主题是边缘云计算应用程序的性能评估，将探讨分布在边缘和云设备上的应用程序的性能评估技术。本课题的短期目标是：3)预测不同应用模块卸载策略下应用程序的执行时间；4)模拟同址应用程序对边缘计算资源的干扰；5)设计一种智能卸载应用模块到共享计算资源的调度技术。该计划的成果将帮助应用运营商通过有效利用边缘和云资源来保持所需的性能和功能，并具有在各种应用领域使用的巨大潜力。

Today's state-of-the-art applications are typically deployed in cloud infrastructure, leveraging cost benefits, elastic scalability, and the illusion of infinite resources. While this satisfies the needs of many application domains such as enterprise web applications, it might not be suitable for modern Internet of Things (IoT)-based applications. IoT applications, such as autonomous driving or wearable health monitoring applications, emit large volume data streams that need to be processed in near real-time fashion. Transferring data to the cloud for processing imposes challenges for latency-sensitive applications due to network latency. To tackle this problem, the processing can be moved to the edge of the network where the data are produced. However, edge devices usually have limited computational capacities. Therefore, there is a need to use both edge and cloud infrastructure at the same time to achieve low latency while having access to scalable and infinite resources. Application developers need to partition their applications and distribute them over the spectrum of edge and cloud devices based on their demand from the underlying infrastructure. This motivates the need to evaluate many different possible configurations in terms of the functionality and performance and selecting the ones that meet the needs of the application. My research program aims to provide a set of tools and techniques for evaluating the performance implication of different distribution policies and for testing the functionalities of partitioned applications. My long-term objective is to devise solutions for reliable and safe deployment of distributed data processing applications for real-time decision-making that utilize computing resources in a cost-effective and energy-aware manner. The first theme of this program, automated integration testing, focuses on partitioning applications into smaller modules and testing the integration of the partitioned application on given heterogeneous devices. The short-term objectives of this theme are to 1) devise effective application partitioning strategies and 2) automated generation of integration test scripts. The second theme of this program, performance evaluation of edge-cloud computing applications, will explore performance evaluation techniques for applications distributed over edge and cloud devices. The short-term objectives of this theme are to 3) predict the execution time of applications under different application modules offloading policies, 4) model the interference of co-located applications on edge computing resources, and 5) devise a scheduling technique that intelligently offloads application modules to shared computing resources. Outcomes of this program will help application operators maintain desired performance and functionality with effective use of edge and cloud resources and have great potential to be used in various application areas.