The art of SQL Warehouse sizing

Warehouse sizing requires calibrating the cluster size and the scaling range. Simply put, we must configure the speed and concurrency with which the SQL Warehouse will process queries submitted by users.

That said, it is important to understand that speed and concurrency are not entirely independent metrics. For example, if you process your queries faster, then the overall throughput of queries will be higher and the amount of time a query spends in a queued state will be shorter. This will avoid having to scale the concurrency of the SQL Warehouse by increasing the cluster size.

So, let’s start by understanding the mechanics of query routing, queuing, and cluster autoscaling.

Rules for query routing, queuing, and cluster autoscaling

The control processes in the SQL Warehouse follow a very simple decision tree for performing query routing, query queuing, and cluster autoscaling.

The 10-Query Rule

The core rule (at the time of...