Many common M queries can be written for both import and DirectQuery datasets, but with widely different implications for the source system resources utilized and the performance of the analytical queries from Power BI. It's essential that the mode of the dataset (import or DirectQuery) has been determined in advance of the development of the data access queries and that this decision is reflected in the M queries of the dataset.
The M queries supporting a Power BI dataset import mode should exclude, or possibly split, columns with many unique values, such as a Transaction Number column, as these columns consume relatively high levels of memory. A standard design technique for import mode models is to exclude derived fact table columns with relatively more unique values when these values can be computed via simple DAX measure expressions based on columns of the same table with fewer unique values.
In the following example, the SUMX() DAX function is used to compute...