The dissertation solves the scientific and practical problem of effective management of the IT infrastructure of cloud data centers under conditions of uncertainty and variable workloads by creating a management methodology and on its basis the development and application of management information technology to provide cloud services with specified quality indicators to the end user. The proposed information technology is developed ensuring the efficient functioning of the IT infrastructure of the cloud service provider's data center by increasing the resource usage efficiency under conditions of variable workload, which unlike the known methods takes into account the developed methodology, essential characteristics of cloud computing, heterogeneity of IT infrastructure, its multilevel and hierarchy and uses adaptation to unpredictable and mixed workloads at the expense of forecasting, which allowed to ensure SLA requirements while reducing operating and capital costs.
The developed information technology is based on the concept of management of IT infrastructure of a cloud data center, which combines the following developed findings, approaches, and methods: the operator form of setting, analyzing and solving problems of IT infrastructure of the cloud data center; identifying and implementing dedicated resource and workload management strategies, as well as their implementation schemes; decomposition of cloud IT infrastructure at three levels (infrastructure, platforms, and applications); taking into account traditional, convergent and hyperconverged architectures of modern cloud data centers; application of the adaptive method of combined workload forecasting to determine the control influences on the IT infrastructure of cloud data centers; developing the Method of Integrated Resource Management for heterogeneous data centers based on SLA violations, power consumption and required power at the next management step; application of stochastic methods (beam search, simulated annealing, reinforcement learning) for implementation of particular strategies for IT infrastructure management of cloud data centers; application of a distributed two-level storage management method for hyperconverged systems; taking into account new metrics of the IT infrastructure state namely instantaneous and average viability coefficients of virtual machine, physical server imbalance indicator, the ratio of necessary resources to the average available resources, threshold of available resources) to determine the current management strategy; accounting of software-defined controllers for management of three primary data center resources namely computing, storage, network; the combination of centralized management using global data center manager and the decentralized management on the physical server level, depending on the chosen management strategy at the current management step.
To solve the problem of effective management of the IT infrastructure of the cloud data center and solving the correspondent set of tasks, the following were used: systems theory, methods of hierarchical systems theory, methods of mathematical programming, methods of operations research and decision theory, methods of mathematical and simulation modeling, methods of artificial intelligence theory, stochastic and heuristic search methods, forecasting methods, mathematical statistics methods, and cloud service models. The reliability and validity of the obtained results are conditioned by the correct use of the mathematical apparatus and are confirmed by the results of computational experiments.
The scientific novelty of the obtained results is determined by the following theoretical and practical results obtained by the author. For the first time the following has been developed: a methodology for managing the IT infrastructure of a cloud data center; a structural and functional model of a multilevel hierarchical software-defined system for managing the IT infrastructure of a cloud data center; an original information technology for managing the IT infrastructure of a cloud data center; an adaptive method of combined workload forecasting for cloud data center computing resources; a method for integrated management of IT infrastructure of cloud data center; a distributed datacenter cloud management method; a cloud datacenter power management method. The following has been further improved: the decomposition-compensation approach; the algorithms and methods of stochastic local search; an IT infrastructure management model with a coordinator.
Keywords: data center, cloud computing, management system, forecasting, stochastic search, optimization, two-level storage with replication, internet of things.
