iPOEM: a GPS tool for integrated management in virtualized data centers.

A fundamental problem that confronts data center administrators in integrated management is to understand potential management options and evaluate corresponding space of the managed system's potential status. In this paper, we present iPOEM, a middleware with GPS-like UIs to support integrated power and performance management in virtualized data centers. iPOEM offers novel system positioning services to enable a declarative management methodology: administrators specify a target location in terms of system performance and power cost, and iPOEM returns the management configurations and operations that are required to drive the system to the target status. In the core of iPOEM lies an automated management configuration engine exposing two simple APIs: get_position() and put_position(). We study the relationships between system status and the management configurations in our problem domain, and design a logarithmic configuration searching algorithm for the engine. Several system positioning services are developed atop the engine, including an auto-piloting scheme leveraging sensitivity based optimization technology, and provide intuitive UIs to operation users. The iPOEM prototype is developed atop Usher, an open-source virtual machine management software. The evaluation driven by real data center workload traces shows that iPOEM renders both intuitive usage and efficient performance in the integrated management of virtualized data centers.