Autonomic management of Green data centers – Ctrl-Green

Green computing which refers to sustainable computing, i.e. practices for disposing of computers with minimal impact on the environment, has become a crucial issue for organizations which manage data center.

New generations of green data centers are designed to improve their efficiency in terms of Power Usage Efficiency (PUE), Carbon Usage Efficiency (CUE), Data Center Infrastructure Efficiency (DCIE) ...
The management of these infrastructures has to take into account these previous metrics, which implies that it has to be adapted according to its usage. Most of these adaptations cannot be performed manually and have to be automated in order to be reactive.

Autonomic management systems have been proposed as a solution for the management of distributed infrastructures to automate tasks and reduce administration burden. An autonomic manager is usually built as a control loop that monitors the environment and reacts to events such as failures or overloads and reconfigures applications and/or infrastructures accordingly and autonomously. Such autonomic management systems are very promising and several experiments have shown that they can be used to efficiently manage energy.

However, important issues are unsatisfactorily addressed:

- energy management may be implemented in different layers, be it at the hardware level, operating system level or middleware level. Therefore several control loops may be implemented in these different levels, and they have to take globally consistent decisions.
- green computing is not restricted to energy management. Previously enumerated metrics (PUE, CUE, DCIE ...) show that many factors (materialized by sensors and actuators) may be taken into account in order to be green efficient. Again, it means that many control loops will have to be implemented to manage these factors, which raises again a consistency problem.
- green management is not the only aspect which has to be handle by an autonomic management system. Such a system may also include management policies for enforcing availability or scalability. These different control loops could take contradictory decisions, so here again, we must ensure global consistency, i.e. manage the trade-off between availability performance and energy.

In summary, a green-aware autonomic management system has to allow the coexistence of several Autonomic Managers (AM) with different goals, implemented in different layers in the same environment. Therefore it intrinsically implies (i) an evolution of current autonomic system to integrate these new green metrics and control capabilities and (ii) a need for coordination of multiple autonomic managers that use these metrics. Our aim is to propose a high-level and tool-supported approach to design such systems, relying on autonomic management and model-based control techniques, in order to master the complexity of resource management in such environments.

Accordingly, this project precisely investigates this question, following 4 directions:
- Reactive control techniques for AM coordination. The objective is to rely on synchronous languages and discrete controller synthesis to program, verify and generate the controller required to enable the cooperation of multiple AMs.
- Controllable autonomic management system. The objective is here to provide an autonomic management platform which features the system support required by the above generated controller and which integrates new green data center capabilities.
- Application to green computing. We aim at implementing several scenarii which involve the coordination of multiple AMs for energy management in different layers (hardware, OS, middleware) as well as multiple AMs which target different management domain (performance, availability, energy …).
- Experimentation on a real green data center. The objective is to evaluate the above autonomic system and reactive control techniques in a real-life green data center running real life applications.