With columnar storage, the CPU can become a bottleneck. SQL Server solves these problems with batch mode processing. In batch mode processing, SQL Server processes data in batches rather than processing one row at a time. A batch represents roughly 900 rows of data. Each column within a batch is stored as a vector in a separate memory area, meaning that batch mode processing is vector-based. Batch mode processing interrupts a processor with metadata only once per batch rather than once per row, as in row mode processing, which lowers the CPU burden substantially. This means that batch mode spreads the metadata access costs over all of the 900 rows in a batch.
Batch mode processing is orthogonal to columnar storage. This means ...
