Dynamic resource allocation for database servers

拟议的项目将研究自主系统的适应，以减少对大规模互联网服务器系统的人工管理的需要。自主系统监控并动态调整其行为，以满足预设目标或行为要求。大规模集群系统经常出现在提供动态内容服务的组织中，例如电子商务、在线银行、股票交易和游戏。此类服务基于存储在数据库中的持久状态动态生成内容。传统上，数据库系统用于在给定时间在独立数据库引擎上运行单个应用程序。通常，在此配置中，专家管理员可以确定该应用程序的最佳数据库设置。大型数据中心的情况并非如此，为了降低成本，电子商务、拍卖和新闻等多个应用程序通常同时托管在相同的硬件上。可能共享相同基础设施的应用程序具有多样性，这使得配置和性能调优变得困难和耗时。此外，服务于动态内容的大规模集群系统的规模越来越大，这使得它们容易出现故障。从故障中恢复以及旨在减少故障概率的活动，如系统备份，占用了很大一部分人员时间，并可能出现人为操作错误。

The proposed project will investigate autonomic system adaptations to reduce the need for human-directed management of large-scale Internet server systems. Autonomic systems monitor and dynamically adapt their behavior to meet preset goals or behavioral requirements. Large-scale cluster systems are often present in organizations that provide dynamic content services such as e-commerce, on-line banking, stock trading and gaming. Such services generate content on-the-fly based on persistent state stored in databases. Database systems have traditionally been used to run a single application at a given time on a stand-alone database engine. Typically, in this configuration, an expert administrator could determine the optimal database settings for that application. This is not the case with large data centers, where multiple applications, such as, e-commerce, auctions and news, are commonly hosted concurrently on the same hardware, in order to reduce costs. The diverse nature of applications that may share the same infrastructure makes configuration and performance tuning difficult and time consuming. Furthermore, the increasing size of the large-scale cluster systems serving dynamic content makes them prone to failures. Recovering from failures, as well as activities aimed at reducing the probability of failures, such as, system backup take up a significant portion of personnel time and introduce the potential for human operator errors.