Helpful information and examples on how to use SQL Server Integration Services.
As a follow-up to my previous post about SCD processing in SSIS, I thought I’d go deeper into using the built-in Slowly Changing Dimension Wizard. While there are definitely more efficient ways to do SCD processing in SSIS, there are some optimizations you can apply to the components that the wizard outputs that might make it more acceptable for your environment.
First, let’s take a look at the Wizard’s strengths. Besides the advantage of not having to write any SQL by hand, there are two key scenarios the SCD Wizard is optimized for:
If your scenario doesn’t match the above, you might want to consider just using one of the alternate approaches directly. If it does, or if you don’t want to use any custom components (or hand craft the SQL required for a MERGE statement, for example), consider making the following optimizations:
The first transform does not cache any lookup results from the reference dimension, so every incoming row results in a query against the database. By default, the wizard will open a new connection to the database on each query. For a performance gain (and less resource usage), you can set the RetainSameConnection property of the wizard’s connection manager to True to re-use the same connection on each query.
The wizard will output three separate OLE DB Command transforms (which perform row-by-row updates). You will get a big performance boost by placing the rows in a staging table, and performing the updates in a single batch once the data flow completes. Another option is to use the Batch Destination component, which is available on Codeplex.
The default destination that the Wizard outputs will have Fast Load disabled (to avoid locking issues). In many cases (mostly depending on the number of rows you’re processing), you can enable Fast Load for an immediate performance gain. To avoid potential deadlocking issues, another option is to use a staging table, and move the data over to the final destination table once the data flow is complete using a INSERT INTO … SELECT statement.
Using the above optimizations, I was able to bring down the processing time of a 200k row change set (against a 100k row dimension table) from 60 minutes to 14 minutes. Note however, that processing a similar change set using the SQL MERGE statement took under 3 minutes to complete.
I go into more detail about these optimizations (and the difference between SCD processing approaches) in my Performance Design Patterns talk.
I like SCD transformation simplicity and I often use it on small dimensions but performance should be much better....I can see you can largely improve performance of SCD but I go with SCD for simplicity reasons so I doubt I would decide to do staging tables and most likely would go with merge which is not so easy to read and folow.
I think the biggest drawback of SCD is that it re-creates everything so I lose all bespoke changes which are needed and are not supported by SCD transformation.
I wished Microsoft made some more work on it as I don't generally like using third party components and would prefer to have everything available. It seems nothing changed in SCD Transformation in SQL 2012 which is a shame.