Abstract. Resource consumption in virtualized computing environment is an important problem in modern cloud infrastructures. This paper introduces methods of detecting and solving problems of unbalanced load nodes in cloud cluster. We propose multi-objective optimization problem analysis as the bin packaging problem, using criteria convolution method to make it a single objective problem and neural networks with back-propagation to reduce balancing time. The considered approach is modeled in AnyLogic and the integration software for the system is implemented on OpenStack cloud platform.Keywords: Cloud Computing, neural networks, load balancing, multi objective optimization, AnyLogic, OpenStack.
IntroductionInformation and communications technologies are becoming an important part of the infrastructure used for the innovative development of the scientific, technical, social and educational activities. A key component of this infrastructure is data-processing systems. The complexity of creating computer systems with the desired characteristics of the operating system, hardware components and network topology strongly depends on the characteristics of problems solved by these systems. That is why the problem of developing technologies for automated reconfiguration of computer systems is important. The usage of virtualization technologies and solutions based on the paradigm of "cloud computing" allows you to enhance significantly the ability to control computing resources, telecommunication environment and heterogeneous resources. This is particularly important in organization of high-performance parallel computing. These computations require expensive hardware and complex software. For example, one of the most important practical applications of cloud technologies nowadays is to process distributed solutions for control problems. In this case "distributed" means that each computational element could be a single agent (e.g., robot) performing computational tasks that require a lot of resources. Group of computational primitives also could solve this task. Systems like this could also provide interaction between groups of agents that can be physically distant.