Authors: Danilo Oliveira; Jamilson Dantas; Nelson Rosa; Paulo Maciel; Rubens Matos; André Brinkmann
Addresses: Informatics Center, Federal University of Pernambuco, Recife, Brazil ' Informatics Center, Federal University of Pernambuco, Recife, Brazil ' Informatics Center, Federal University of Pernambuco, Recife, Brazil ' Informatics Center, Federal University of Pernambuco, Recife, Brazil ' Coordination of Informatics, Federal Institute of Education, Science, and Technology of Sergipe, Lagarto, Brazil ' Data Processing Center (ZDV), Johannes Guttenberg University, Mainz, Germany
Abstract: Creating a cost-effective and highly available private cloud infrastructure from scratch is challenging. In order to improve availability, system administrators commonly employ redundant hardware, which incurs increased acquisition cost. There are plenty of decisions involved in this process. It is necessary to specify how many servers will be employed, and each of them will have a role in the infrastructure either as a worker node or as a management node. One question that arises is: among all possible ways to arrange the servers, and considering a set of requirements (e.g., the minimum number of available workers, minimum availability), which configurations can sustain a minimal availability level (steady state and capacity oriented)? This work proposes an optimisation model for private clouds planning supported by analytical modelling for computing the cost of solutions. By combining the optimisation model with the bisection method, it is possible to minimise the acquisition cost while maintaining the high availability of the cloud. The proposed method can also be applied to other parallel or distributed systems with the same properties as the private clouds used in this study.
Keywords: cloud computing; steady-state availability; capacity-oriented availability; combinatorial optimisation.
International Journal of Web and Grid Services, 2019 Vol.15 No.4, pp.367 - 393
Accepted: 04 Jun 2019
Published online: 21 Oct 2019 *