APS AU4 Feature Review: Introducing Round Robin Distribution
This is our second post covering the new features and improvements included in the SQL APS PDW AU4 release.
Starting with AU4, tables in PDW now have a third distribution option. In addition to the familiar Hash distribution and replicated formats, AU4 introduces the Round Robin distribution type. This distribution type was first introduced in the SQL Data Warehouse Service and has now made it into the PDW workload. I would recommend reviewing the content on the SQL Data Warehouse Documentation site, as almost everything applies to the appliance as well and is a great explanation of the feature. The main difference being for the appliance the default distribution method will remain to be replicated. Tables will only utilize the ROUND ROBIN distribution type if it is specified in the create command.
While Round Rodin distribution will get even distribution and eliminate skew, it will not be distribution compatible for a large number of scenarios. It is very likely that joins and aggregations to distributed tables or other round robin tables will require data movement. It is recommended to use this distribution type when you are beginning a migration to PDW and do not yet know of a good distribution key, or when no good distribution key exist for a large table. If queries against round robin tables contain predicates which greatly reduce the amount of data to be moved, it is possible the even distribution may be enough benefit to warrant permanent use.
Example Syntax:
CREATE TABLE round_robin_test
(
col1 int,
col2 varchar,
col3 int
)
WITH
(DISTRIBUTION = ROUND_ROBIN)
Like other distribution formats, it will take testing in your specific environment to find the best combination of distribution methodologies to get optimal performance.
The first post in this series can be found at:
Please check back for future posts!