Organization of multi-level data storage
B. Ya. Sovetov, Tatiana M. Tatarnikova, E. D. Poymanova · Информационно-управляющие системы · 2019
Introduction: In the face of a constantly increasing data volume, almost all organizations need to form a well-developed storageinfrastructure. The infrastructure is implemented using multi-level storage technologies: designing systems of storage devicesheterogeneous in physical nature and access architecture. On the other hand, the files are also subject to the requirements about theguaranteed storage time, even at the legislative level. The lack of efficient storage organization algorithms with different storagerequirements actualizes new storage resource management models. Purpose: Developing new models and algorithms for managing datastorage resources, taking into account the layered principle of data storage. Results: A multi-level data storage algorithm is proposedwhich allows you to distribute files in a data storage system in accordance with the sequential use of vertical, horizontal and dynamicfile allocation. The algorithms for data placement and migration across the storage hierarchies differ in the use of metadata analysis,providing a way for proactive management of the storage resources based on guaranteed retention periods. A data storage model isproposed in the form of a matrix whose size corresponds to the number of vertical and horizontal levels in the system. It is suggested todistribute files to the cells of the matrix using the Kohonen neural network tools. The effect of using the Kohonen network is the abilityto analyze metadata and the frequency of accessing files in one step, avoiding sequential implementation of the allocation procedures.The results of the experiment have demonstrated the effectiveness of the Kohonen network tools in solving the problem of placing filesin a multi-level data storage system. Practical relevance: The proposed model and algorithms can be used in the design of data storagesystems, taking into account the multi-level principle of data storage.