Usually relational databases are designed to record, modify and query the data itself (add, delete, change) and are not concerned with the business implications contained in the data itself. Simply put, they are for recording.

While the data warehouse is focused on data and data between the business meaning, such as different statistical caliber and analysis of the significance of the difference between the values. Simply put, it's for analysis.

So in order to ensure the accuracy and effectiveness of the analysis results, it is necessary to decompose and refine the analysis of caliber, precise positioning of the factors affecting the dimensions of the table and metrics is to define the analysis angle and the factors affecting a way.

The volume of data does not reach a certain level, and business needs do not require, for example: news topics table, a few hundred G normal, it may not be necessary to split, but companies like Sina must be split. In other words, it depends on the amount of data, business development trends, data access needs, data access concurrency and other considerations; large data volumes, high concurrency and other scenarios, it should be accurate to say that the general: vertical split + horizontal split, is certainly to provide the performance, system load capacity, support for business incremental and so on.