Dremio is a batch processor, not a stream processor. The fundamental difference is that a batch processor will need to recompute a query from scratch whenever the input data changes, while a stream processor can incrementally update the existing query result based on the change to the input.

This can make a huge difference when making small changes to large datasets. Materialize can incrementally compute small changes to very complicated queries in just a few milliseconds, while with batch processors you're looking at latency in the hundreds of milliseconds, seconds, or minutes, depending on the size of the data.

Another way of looking at it is that in batch processors, latency scales with the size of the total data, while in stream processors, latency scales with the size of the updates to the data.