Column storage. In the impression, the data is organized by line, that is, one of the records. There’s a column store?

“Big data is currently stored in two ways: row storage and column storage. There has been much debate over the two storage solutions, focusing on which one can handle massive amounts of data more efficiently while maintaining security, reliability, and integrity. From the current development situation, relational database has not adapted to this huge storage and calculation requirements, is basically out of the game (is it? Too sensational, I doubt it). Among several known big data processing software, Hadoop’s HBase uses column storage, MongoDB is a document type of row storage, and Lexst is a binary type of row storage.”

\

Storage:

\

Storage:

\

The advantage of row storage is fast insertion and modification;

Column storage is faster when querying a single column or a small number of columns. As it is organized by column, the data type of the same column is consistent, so it is easy to store and parse.

“If the priority is data integrity and reliability, then row storage is the right choice, and column storage will only come close with the addition of disks and improved software design. If you store data primarily, the write performance of row storage is much higher than that of column storage. Column storage is most appropriate in applications that require frequent reads of single-column collections. If more than one column is read at a time, two options are available: when row storage is used, the design should consider reducing or avoiding redundant columns; In the column storage scheme, data in each column is stored on different disks as far as possible to ensure I/O efficiency. Multiple threads read and write data in parallel, avoiding disk race and improving processing efficiency. Regardless of which solution you choose, clustering together content data is a must, which is an effective way to reduce head movement across disk and improve data read time.”

\

Reference article:

www.infoq.com/cn/articles… \

\

\

\

\

\