A new generation of green data centers has been designed to improve their effectiveness in terms of Power Usage Efficiency (PUE) and Carbon Use Efficiency (CUE). The administration of these facilities should consider these metrics, which implies that they must be adapted for this purpose. Such adjustments cannot be made manually and must be automated to be reactive. Autonomous systems have been proposed to automate administration tasks of such facilities. Such a autonomic management is very promising and various experiments have shown interest in energy management. However, critical issues have not been addressed satisfactorily:
- Energy management can be considered at different levels (hardware, system, middleware). Multiple control loops can be implemented at every level and must take globally consistent decisions.
- Green computing is not the only aspect to be managed. Policies for scalability and availability have to taken into account regarding user needs. It is therefore necessary to manage the tradeoffs between performance, availability, energy, while incorporating new metrics and capabilities of green data centers.
In summary, autonomous systems for Green Computing should enable the coexistence of many autonomic managers (with different objectives, implemented in different layers) in the same environment.
Our approach is based on the development of autonomous systems to integrate these new metrics greens and control capabilities and coordination among autonomous managers using these metrics. We propose an approach to high-level features and tools for designing such systems based on the autonomous management and control techniques based on behavioral patterns, in order to control the complexity of the Green Computing. This project therefore explores this question in 4 directions:
- Reactive control techniques for coordination. The objective is to use synchronous languages and discrete controlle synthesis techniques to program, verify, and generate controllers required for the cooperation of autonomic managers.
- A controllable autonomous system: the aim is to provide a platform independent system that provides the support required for the implementation and integration of controllers generated above, and integrates new capabilities of data center green.
- The application to green computing: we have implemented several scenarios that involve coordination managers for Green Computing in different layers (hardware, system, middleware) while targeting different areas (performance, availability, energy, ...)
- Experiments on a real data center industry: the objective is to evaluate the autonomic systems and control techniques above in a green data center on which real applications execute.
Ctrl-Green (ANR-11-INFR 012 11) is a research project financed by the ANR (Agence Nationale de la Recherche) with the support of MINALOGIC. This project aims to study the hardware and software for optimizing energy consumption in data centers.