Research on visual data storage methods

The article is dedicated to the study of the efficiency of various technologies and storage approaches for large volumes of visual data. Through the research, the modern data storage methods were analysed, along with the advantages and disadvantages of using relational and non-relational databases, available built-in technologies for working with larger files, such as MS SQL Server FILESTREAM, compared to storing this data in separate storage. Writing, reading and filtering benchmarks were performed with data of different sizes and volumes to determine the performance, as well as the efficiency of system resources usage. FILESTREAM was supposed to increase the performance of the system, but based on the research results, storing large-sized data increased write time by 110%, sequential read time by 118% and filtering time by 32%, compared to storing images in the database. Direct image storage in the database increased data fragmentation and reduced index efficiency, which ended up lowering system scalability and increasing experiment runtime under high load. Hybrid methods of data storage showed the best results. DBMS Microsoft SQL Server was 1.95 times on average faster than MongoDB, even though MongoDB showed less performance loss for data writing and filtering.