Building Sensors and Actuators for Adaptive Resource Management in Linux Systems

The foundation of adaptive resource management includes two basic elements: sensors and actuators. Sensors are monitors that collect management data, and actuators are control knobs that are available to change resource configuration, behavior, or allocation. Along with adaptive algorithms, the foundation forms one or more control loops that adapt to environment changes. In this paper we focus on building sensors and actuators in enterprise Linux systems for adaptive resource management. We start by reviewing the current status of industry open standards, including DMTF’s Common Information Model (CIM) for system management and Application Response Measurement (ARM), and build sensors for monitoring applications such as the Apache web server. The resulting ARM infrastructure includes a standalone ARM library for application instrumentation, an ARM Agent for collecting and distributing application statistics, a CIMOM ARM Provider, and a web server plug-in that captures URL-based statistics for web servers. For actuators we expose tunable system parameters by building a management tool to provide a consistent and user-friendly GUI, and create new actuators for per-process parameters and entries in configuration files.