Toward Energy‐Aware Scheduling Using Machine Learning

8.1 INTRODUCTION The cloud and the Web 2.0 have contributed to democratize the Internet, allowing everybody to share information, services, and IT resources around the network. With the arrival of digital social networks and the introduction of new IT infra-structures in the business world, the Internet population has grown enough to make the need for computing resources an important matter to be handled. While few years ago enterprises had all their IT infrastructures in privately owned data centers, nowadays the big IT corporations have started a data-center race, offering computing and storage resources at low prices, looking for outside companies to trust them for their data or IT needs. A single web application in the cloud can be easily used by people from around the world, so data and computation need to be available from everywhere, having in mind things such as the quality of service (QoS) and the service-level agreements (SLAs) between users and servers. Services offered by Google and YouTube, for example, must be replicated around the globe or just be efficient enough to move data, jobs, or applications among the data-center farms spread along the planet. Given the amount of applications running now on the cloud and the amount that will come, coordinating all its applications, resources, and services becomes by itself a hard optimization problem. Having powerful enough data-centers to server applications or computation time is not the only thing to keep in mind when building the Cloud. As energy-related costs have become a major cost factor for IT infrastructures and data centers, power consumption has become an important element to keep in mind when designing and managing them. This energetic cost is reflected in the electric consumption, which is sometimes nonlinear with the capacity of that data centers. It also has direct environmental impact and is conditioned by social pressure for efficiency. Companies dedicated to cloud-based services, and the research community are being challenged to find better and more efficient power-aware resource management strategies. Until now technological improvement sufficed to cover the increasing IT demand, bringing faster processors, bigger storage devices, and faster connections between resources. The energetic factor was not relevant enough to be focused on. Now the demand is growing faster than technological improvement, so each time we need bigger data centers to be cooled down in colder places, having enough power supply [1]. Reaching an optimal performance of cloud services and resource …