Method of Distributed Two-Level Storage System Management in a Data Center

Abstract

Modern applications in cloud computing, internet of things and machine learning are I/O intensive. They use data storage systems as the main resource of a data center. The advent of new storage technologies makes it possible to increase the performance of I/O operations by integrating devices with different performance within the data storage system by utilizing the storage-tiering approach. To prevent data loss and service downtime, the data storage systems must ensure fault tolerance using data replication management. As modern hybrid IT infrastructures are based on hyperconverged systems, the development of new methods and models for storage management in order to ensure high performance of I/O operations, high availability and fault tolerance becomes an urgent need. The authors propose the management method based on the model of a distributed two-level data storage system. The proposed method uses the algorithms for data migration between fast and slow levels of the data storage system and the algorithms for replication of data between the nodes of distributed data storage. The simulation results indicate that the proposed management method allows increasing performance of I/O operations with files and evenly placing replicas of data blocks on the data center nodes.