Columnstore Parallelism

Sunil Agarwal shows how clustered columnstore indexes take advantage of parallelism:

Do you need to be concerned about that a delta rowgroup is scanned single threaded? The answer is NO for two reasons (a) most columnstore indexes have very few delta rowgroups (b) if you have multiple delta rowgroups, they can be scanned in parallel with one thread per delta rowgroup

I have a beef with (a), at least for SQL Server 2014, but that’s a story for another day.

Sunil has a follow-up post on parallel bulk import:

Recall that on rowstore tables (i.e. the tables organized as rows not as columnstore), SQL Server requires you to specify TABLOCK for parallel bulk import to get minimal logging and locking optimizations. One key difference for tables with clustered columnstore index is that you don’t need TABLOCK for getting locking/logging optimizations for bulk import. The reasons for this difference in behavior is that each bulk import thread can load data exclusively into a columnstore rowgroup. If the batch size < 102400, then the data is imported into a delta rowgroup otherwise a new compressed rowgroup is created and the data is loaded into it. Let us take two following interesting cases to show this bulk import behavior. Assume you are importing 4 data files, each with one bulk import thread, concurrently into a table with clustered columnstore index

The “don’t use TABLOCK” is interesting in comparison to rowstore tables.

Related Posts

Linked Servers And Columnstore

Niko Neugebauer continues his columnstore series by looking at how they interact with linked servers: Lets us make sure everything is fine for data transfer and as we are using our source server (SQL Server 2014) with Linked Server to SQL Server 2016, let us insert a couple of ObjectIds to the T1 table that we have […]

Read More

Columnstore Segment Elimination Pains

Niko Neugebauer shows us two pain points with row group (segment) elimination on clustered columnstore indexes: Still this gives us processing of just 1 Segment with 253 lob logical reads vs 6 Segments and 2669 lob logical reads in the original query. But wait, that’s not all! It can get significantly worse! Let’s use the […]

Read More


February 2016
« Jan Mar »