## Metadata - Author: [[airbyte.com|Airbyte]] - Full Title:: The Rise of the Semantic Layer - Category:: #🗞️Articles - Document Tags:: [[pagesmetrics-layer|Pages/metrics Layer]], - URL:: https://airbyte.com/blog/the-rise-of-the-semantic-layer-metrics-on-the-fly - Finished date:: [[2023-06-12]] ## Highlights But in the other hand, queries may be super slow/expensive > In the transformation layer, you must balance low and high [granularity](https://glossary.airbyte.com/term/granularity). What level do you aggregate and store (e.g., [rollups](https://glossary.airbyte.com/term/rollup) hourly data to daily to save storage), or what valuable dimensions to add. With each dimension and its column added, rows will [explode](https://www.ibm.com/docs/en/ida/9.1.1?topic=phase-step-identify-measures#c_dm_design_cycle_4__c_dm_4_step7) exponentially, and we can’t persist each of these representations to the filesystem. > A semantic layer is much **more flexible** and makes the most sense on top of transformed data in a data warehouse. Avoid extensive reshuffles or reprocesses of large amounts of data. Think of [OLAP](https://glossary.airbyte.com/term/olap-online-analytical-processing) cubes where you can dice-and-slice ad-hoc on significant amounts of data without storing them ahead of time. ([View Highlight](https://read.readwise.io/read/01h2mxgw34awy5rb529dsna84v))