EN FR
EN FR


Section: Application Domains

Control of data centers

Data centers are another example of a large scale reconfigurable and distributed system: they are composed of thousands of servers on which Virtual Machines (VM) can be (de)activated, migrated, etc. depending on the requests of the customers, on the load of the servers and on the power consumption. Autonomic management functionalities already exist to deploy and configure applications in such a distributed environment. They can also monitor the environment and react to events such as failures or overloads and reconfigure applications and/or infrastructures accordingly and autonomously. To supervise these systems, Autonomic Managers (AM) can be deployed in order to apply administration policies of specific aspects to the different entities of a data center (servers, VM, web services, power supply, etc). These AMs may be implemented in different layers: the hardware level, the operating system level or the middleware level. Therefore several control loops may coexists, and they have to take globally consistent decisions to manage the trade-off between availability, performance, scalability, security and energy consumption. This leads to multi-criteria optimization and control problems in order to automatically derive controllers in charge of the coordination of the different AMs. We are relatively new on this topic, that will require more technical investment. But we are driven to it by both the convergence of IT and networking, by virtualization techniques that reach networks (see the growing research effort about network operating systems), and by the call for more automation in the management of clouds. We believe our experience in network management can help. Some members of the SUMO team are already involved in the ANR Ctrl-Green, which addresses the controller coordination problem. We are also in contact with the Myriads team, which research interests moved from OS for grids/clouds to autonomic methods. This is supported as well by the activities of b<>com, the local IRT, where some projects in cloud management and in networking may start joint activities.