CSR: Small: Collaborative Research: Software Defined Energy Adaptation in Large Scale Data Centers

University Of Minnesota-Twin Cities, Minneapolis MN

Investigators

Abstract

Today many compute and data intensive applications are running on the computers in data centers. Therefore, data centers consume large amounts of energy. Within a data center, many compute, storage and networking components are coordinated together to support these applications. This project studies how to efficiently allocate resources in data centers to satisfy the requirements of these applications with guaranteed quality of service (QoS) and at the same time taking advantage of the available renewable (green) energy to reduce the total energy to be consumed. As the data centers increase in size and complexity, it is becoming clear that the traditional distributed control of their resources poses daunting problems in ensuring the desired agility and QoS support demanded by the enterprise applications. This problem is further compounded by the need for aggressive energy management in order to minimize energy cost, cope with power and thermal related constraints, and adapt to variability in the energy produced from local "green" sources. A distributed management of energy is particularly difficult since energy by its nature is a central resource that must be properly divided up among various infrastructure components in order to yield acceptable application performance. This project explores a flexible, policy driven, software defined mechanism to manage energy at all levels of the data center and for all major resources including networks, storage systems, and compute servers. A crucial aspect in this research is the coordination of energy management decisions at various levels which is essential to achieve optimal performance under given energy/thermal constraints. The project brings in the emerging concept of software defined resource management to seamlessly tackle the energy sustainability and corresponding QoS issues in data centers. The project also forms a pilot realization of the software defined data center, since energy adaptation requires an adept management of all major data center resources. A set of new algorithms in resource allocation, resource monitoring, QoS enforcement, energy allocation and distribution will be designed and developed. These algorithms will be implemented into a specific framework to demonstrate and validate the usage and benefits of the research. The broader impacts of the project include mechanisms to enhance energy sustainability of data centers and a comprehensive mechanism for software control of large computing systems. The project also augments the research being done in the NSF Industry/University Cooperative Research Center (IUCRC) on intelligent storage and will use it as a conduit for industry adoption of this research.