ITR: Feedback Control of Dynamic Computing Systems

软件系统已经变得越来越复杂，并被期望在广泛的运行范围内提供服务。为了实现服务目标，软件开发人员提供了许多调整参数，这些参数必须仔细设置，系统才能提供可接受的性能和功能。充分调整这些参数通常需要系统管理员进行大量的反复试验。随着工作负荷和运行条件的不断变化，参数必须无休止地重新调整才能提供良好的性能。本研究项目将在两个示例系统中应用反馈控制来自动化参数调整循环。第一种是广域复制的文件系统，通过乐观并发控制提供可变的一致性和性能。反馈意见将用于安排副本对账和数据迁移，以向最终用户提供可接受的性能和一致性。第二个系统是数据库引擎，它提供了大量需要考虑的调优参数。反馈控制将用于使数据库管理员能够管理其服务器系统日益增长的复杂性。通过这两个例子和其他工作，一个动态计算系统的反馈控制框架将开始出现。这个框架将使未来的计算系统能够提供可靠和健壮的操作和可预测的性能，而不需要昂贵的维护和持续的手动调整。

Software systems have become increasingly complex, and are expected toprovide service over wide operating ranges. To meet service goals,software developers provide many tuning parameters that must beset carefully for the system to provide acceptable performance andfunctionality. Tuning these parameters adequately often requires extensivetrial-and-error experimentation by system administrators. As workloads andoperating conditions continually change, the parameters must be endlesslyre-tuned to provide good performance.This research project will apply feedback control to automate thisparameter tuning loop in two example systems. The first, a wide-areareplicated file system, provides variable consistency and performancethrough optimistic concurrency control. Feedback will be used to schedulereplica reconciliation and data migration to provide acceptableperformance and consistency to end users. The second system, a databaseengine, offers a large variety of tuning parameters to be considered.Feedback control will be used to enable database administrators to managethe growing complexity of their server systems. From these two examples and other work, a framework for feedback control ofdynamic computing systems will begin to emerge. This framework will enablefuture computing systems to offer reliable and robust operation andpredictable performance, without costly maintenance and continual manualtuning.