GreenIT – EnerGy-efficient REsourcE AllocatioN in AutonomIc Cloud CompuTing

Principal Investigator: Prof. Pascal Bouvry
Funding source(s): CORE/FNR

A simple concept that has emerged out of the conceptions of heterogeneous distributed computing, grid computing, utility computing, and autonomic computing is that of cloud computing (CC). In CC, end-users do not own or rent any part of the infrastructure. They (the end-users) simply use the services available through the CC paradigm and pay for the services used. The CC paradigm can offer any conceivable form of services, such as databases, computational resources, social networking, and telephonic services.

As such, CC can be perceived as an immediate extension of data centers that are huge facilities housing IT services on hundreds of high-end computing servers. One of their major concern is the huge amounts of electrical power consumption, and most of the energy was spent to cool the underutilized high-end computing servers. There is no general shift by the CC service providers on the issue of energy consumption; therefore, CC may reach a similar faith as data centers, if not made energy-efficient.

Green-IT aims to provide a holistic autonomic energy-efficient solution to manage, provision, and administer the various resources within the CC paradigm.

The main research challenges that will be tackled to achieve the holistic approach are noted below.

• Development of meta-models: CC is a complex system of numerous pervasive devices that request services over heterogeneous network infrastructures from a data center that is energy gobbler. Because each computing entity’s performance is defined uniquely, we must develop meta-models that can adequately define a unified performance metric of the system, and the system’s properties, constraints, and optimization criteria.
• Develop resource management methodologies: With several possible objectives and constraints, the meta-models must result in multi-objective multi-constraint problems (MOC). Green-IT will develop, refine, and evolve solutions for MOC that will primarily be motivated based on, goal programming, adaptive weighted sums approach, homotopic functions, boundary intersections, multi-level programming, Stackelberg games, and multi-objective genetic algorithm solvers.
• Develop autonomic resource management: The anytime anywhere slogan only will be effective when an autonomic management of resources can be achieved. The resource allocation methodologies developed must go further refinement such that the system at hand is self- healing, repairing, and optimizing. In particular, it is our intention to utilize multi-agent systems (MAS) that can learn to adapt (machine learning methodologies) and gracefully evolve to adapt (evolutionary game theoretical methodologies).

Because the CC paradigm is in its infancy, definitions, protocols, policies, implementations, are all undergoing scrutiny. Besides major IT companies, such as Google, IBM, Microsoft, and Amazon, there is no sole university-based research group that is conducting cutting-edge research in autonomic energy-efficient management of resources in CC. This will ensure UL’s lead in research on CC that will significantly impact our daily lives. The large set of data centers based in Luxembourg (public and private) are the obviously targets for Green-IT. Luxconnect, as confirmed by Prof. Engel, president of Luxconnect data center, will take an active part in this project and Dr. Koenig (advisor to UL rector) confirmed that Green-IT fits into UL sustainability plan for Esch-Belval.