Architecture for performance instrumentation in distributed systems.

摘要

The managing of applications in a distributed environment is a complex task and lacks performance measurement facilities for large-scale deployment.; The performance management tasks of application design, deployment, bottle-neck analysis, and capacity planning require the collection, modeling and analysis of workload data. Conventional techniques used to design and manage high performance, monolithic applications are inadequate for distributed systems in general. A systematic approach based on performance engineering is required, supplemented by stronger support of performance metrics in the reference model for open distributed processing (RM-ODP).; This work describes performance instrumentation and measurement interface specifications that support performance related tasks such as configuration planning, application tuning, bottleneck analysis, and capacity planning. These performance measurement capabilities are a necessary component of any commercially viable computer technology and are currently insufficient in the Open Software Foundation (OSF) Distributed Computing Environment (DCE).; The guiding philosophy is to define a set of standardized performance instrumentation that is consistently collected, reported, and interpreted in a heterogeneous environment. An architecture of an efficient, scalable Distributed Measurement System (DMS) is described. The DMS is a software-based measurement infrastructure that defines standard performance metrics, instrumentation and interfaces. DMS provides correlated performance metrics across objects and their channels, integrates disparate measurement interfaces from a node's nucleus object (operating system) and channels (networking), and efficiently transports collected data. The architecture is realized as a prototype based on the OSF DCE.; Other important aspects of this work include evaluating the DMS as a monitor for quality of service (QoS) and collecting model building parameters. The impact of DMS on processor, network utilization, workload responsiveness, and various other factors that control DMS affect its overhead have been studied closely. Lastly, the scalability of DMS is considered for large distributed environments. Preliminary results indicate that DMS is well suited for monitoring, QoS and supporting workload characterization for model building.