The short answer
Azure Synapse Dedicated SQL Pool is a massively parallel processing architecture that spreads database storage across 60 underlying distribution databases managed by control and compute nodes. Table distribution determines how rows are assigned across those distributions.
Distribution strategies compared
Selecting the wrong distribution causes severe data movement across compute nodes during query execution, leading to slow query performance and cluster bottlenecks:
- Hash distribution: Uses a deterministic hash function on a single column to assign rows across the 60 distributions. Best for large exceeding two gigabytes or 60 million rows. Colocating 2 tables on the same join key allows joins to execute locally on each node with zero data movement.
- Replicated distribution: Copies the entire table to every compute node. Best for small lookup or under two gigabytes. Joins between a hash-distributed fact table and a replicated dimension require no network shuffling.
- Round-Robin distribution: Distributes rows evenly in circular sequence across all distributions. Best for temporary staging tables during initial ingest or when no clean join key exists.
Hash (Fact): Distributes by key across 60 buckets (Fast joins on key)
Replicated (Dim): Full copy cached on each Compute Node (Zero network hop)
Round-Robin (Stage): Spreads rows evenly (Fast write, slow joins)Choosing the right hash column
A good hash distribution key must possess high , avoid skewed values such as default null placeholders, and represent the column most frequently referenced in join and aggregation clauses.