How does Teradata handle Skew in Joins? The skewed workload is one of the two most important problems we have to solve in Performance Tuning. This is especially important for join steps, as large amounts of data may be copied between the AMPs. The optimizer has techniques to execute joins separately for skewed primary index

Read More

Teradata can choose from various join methods. The most common ones are Merge Join, Hash Join, Product Join, or for PPI tables, the Merge Join variants Rowkey Based Merge Join, or Sliding Window Merge Join. Less often, however, we see the Teradata nested join, since specific prerequisites must be met before it can be used.

Read More

Joining partitioned tables can be quite expensive, especially if partitions don’t match or one of the tables is not partitioned. Today I discovered an interesting trick that can be applied to reduce resource usage if joining two tables with matching Primary Indexes and Partitions. The test setup consists of the following two tables (both tables

Read More

Sometimes natural skew can become a huge problem, causing bad join performance. The Partial Duplication & Partial Redistribution (PDPR)  feature on Teradata 14 (and above) helps to reduce this issue, but will not always be able to detect all possible applications (for example, if statistics are not revealing the skewed values). If you are stuck

Read More

One of a kind: DWH Pro gives you a detailed description of the Teradata sliding window merge join. There is almost no information available about how this join process works. Therefore we decided to immerse ourselves deeply on this topic and present you with our insights. Before explaining the details of the Teradata sliding window

Read More

Dealing with data skew can be considered a principal activity in Teradata performance optimization. Although skew on table level usually can be avoided by choosing a proper Primary Index, skew arising during query execution is often a major issue (so-called spool skew). Spool skew is often emerging when joining tables (or spools). On a typical Teradata

Read More