Autonomic and adaptive load distribution strategies for reducing energy consumption under performance constraints in data centers
The virtualization strategies of IT resources have been evolving all possible fields of IT markets and industries. Nowadays, almost everything is or might be shifted to the cloud and proposed in the market for different customer sectors as services based on the model of cloud computing. However, this model has also introduced new challenges in addition to the normal system landscape engineering ones. Accordingly, many obstacles are spotted in dealing with that rapid growth of IT system landscapes due to the increase of their structural complexity. The engineering process of the system landscape itself is not anymore the central task to optimize but also crucial to efficiently utilize that system landscape. In other words, reducing the tremendous costs and investments in the IT infrastructure by the IT service providers is not anymore the only concern but rather reducing the associated operational costs of that infrastructure. Many studies stressed on the electricity consumption and its large proportion of the overall operational costs of IT services providers. Virtual machines live migration is a recent topic in addition to some others, in which the allocation of resources based on various load distribution strategy is investigated to accomplish an efficient energy consumption in data centers. More precisely, active virtual machines are migrated between available physical hosts to minimize the number of active servers. The major challenge in designing load management strategies lies in understanding the nature of the incoming workload patterns and their characteristics. Since the heterogeneity of the incoming workload patterns is considerably high, the presented solution approaches in the literature are either problem-specific or highly generic. Both types suffer major drawbacks in terms of applicability and the designed objective function. The aim of this research is to present an autonomic load distribution strategy, which adapts to the heterogenic nature of the incoming workload patterns in data centers with the minimum required human intervention to reduce operational costs under performance constraints.