Performance evaluation and management of enterprise application systems

组织已经开始严重依赖企业应用程序来支持他们的客户。这些应用程序通常用于提供面向业务的服务，如在线零售和客户关系管理。它们通常部署在复杂的多层平台上，这些平台集成了各种组件，如Web服务器、应用程序服务器和数据库服务器。构建此类系统的一个重要考虑因素是服务质量（QoS），例如，用户体验到的快速请求响应时间。性能差可能会使最终用户感到沮丧，并可能给运营这些服务的组织造成经济损失。拟议的研究计划旨在深入了解此类应用程序所经历的工作负载以及用于托管它们的现代多核服务器和虚拟化平台的属性如何影响性能。该计划的首要目标是利用这些见解来设计用于优化和管理企业应用程序系统性能的新技术。尽管多核服务器和虚拟化的快速发展，但是在定义新的方法来帮助组织优化企业应用程序在这些平台上的部署方面还没有足够的工作。现有的绝大多数研究都集中在批量和科学应用上。相比之下，对企业应用程序的关注较少，企业应用程序通常涉及更多的终端用户交互，因此受到更严格的性能要求的约束。关注这些应用程序的少数研究没有详细考虑可能影响其性能的重要因素的含义，例如突发性、现代服务器硬件的体系结构特性和与虚拟化相关的软件瓶颈。缺乏指导性的方法会导致糟糕的部署，从而导致企业遭受严重的经济损失并失去战略优势。该项目将通过性能调优、建模和运行时性能管理方面的新工作来弥合这一差距。

Organizations have come to rely heavily on enterprise applications to support their customers. These applications are typically used to provide business-oriented services such as online retailing and customer relationship management. They are often deployed on complex, multi-tiered platforms, which integrate various components such as Web servers, application servers, and database servers. An important consideration in building such systems is the Quality of Service (QoS), e.g., fast request response times, that the users experience. Poor performance can frustrate end users and can cause economic losses to organizations operating these services. The proposed research program seeks to gain an in-depth understanding of how the workloads experienced by such applications as well as properties of the modern multi-core server and virtualization platforms used to host them influence performance. The overarching objective of the program is to leverage such insights to devise novel techniques for optimizing and managing the performance of enterprise application systems.In spite of the rapid proliferation of multi-core servers and virtualization, not enough work exists on defining new methodologies that can help organizations optimize the deployment of enterprise applications on such platforms. A vast majority of existing research has focused on batch and scientific applications. In contrast, lesser attention has been devoted to enterprise applications, which typically involve more end user interaction and are hence governed by more stringent performance requirements. The few studies that have focused on these applications have not considered in detail the implications of important factors such as burstiness, architectural features of modern server hardware, and virtualization-related software bottlenecks, which can influence their performance. The lack of guiding methodologies can result in poor deployments, which cause enterprises to incur severe financial penalties and a loss of strategic edge. This program will bridge this gap through new efforts in the areas of performance tuning, modeling, and runtime performance management.