Integration Services :: DefaultBufferMaxRows - Is It Determined By Row Length Of Data Flow Task Source Or Destination

We have a single generic SSIS package that is used to import several hundred iSeries tables into SQL. I am not looking to rewrite the process. But I am looking for ways to improve performance.

I have tried retain same connection, maximum insert commit size, lock table (tablock), removed some large columns, played with the log file location and size, and now I am working to tweak the defaultbuffermaxrows.

To describe the data flow task - there are six data flows tasks (dft)  working at the same time. Each dtf has their own list of iSeries tables and columns and the corresponding generic SQL table names. Each dtf determines their list of tables based on the number of columns to import. So there is dft30 (iSeries table has 1-30 columns to import), dtf60 (iSeries table has 31-60 columns to import), etc. The destination SQL tables are generically called Staging30, Staging60, etc. Each column in the generic Staging tables are varchar(100). The dtfs are comprised of an OLE DB Source and an OLE DB Destination.

The OLE DB Source uses a SQL Command from Variable to build a SELECT statement. The OLE DB Source uses a connection manager that uses an IBM iAccess IBMDA400 provider.  The SQL Command ends up looking like this for the dtf30. This specific example is importing from the iSeries table TDACLR and it only has two columns so it will be copied to the Staging30 table.

select TCREAS AS C1,TCDESC AS C2,0 AS C3,0 AS C4,0 AS C5,0 AS C6,0 AS C7,0 AS C8,0 AS C9,0 AS C10,0 AS C11,0 AS C12,0 AS C13,0 AS C14,0 AS C15,0 AS C16,0 AS C17,0 AS C18,0 AS C19,0 AS C20,0 AS C21,0 AS C22,0 AS C23,0 AS C24,0 AS C25,0 AS C26,0 AS C27,0 AS
C28,0 AS C29,0 AS C30,''TDACLR'' AS T0 from Store01.TDACLR

The OLD DB Source variable value looks like the following, but I am not showing the full 30 columns

select cast(0 AS varchar(100)) AS C1,cast(0 AS varchar(100)) AS C2,cast(0 AS varchar(100)) AS C3,cast(0 AS varchar(100)) AS C4,cast(0 AS varchar(100)) AS C5, ... cast(0 AS varchar(100)) AS C30.

The OLE DB Destination uses OpenRowSet Using FastLoad From Variable. The insert into Staging30 ends up looking like this.

insert bulk STAGE30([C1] varchar(100) ,[C2] varchar(100) ,[C3] varchar(100) ,[C4] varchar(100) ,[C5] varchar(100) , ...  ,[C30] varchar(100) ,[T0] varchar(20)

Of course we then copy and transform the Staging30 data to the SQL table that equals T0.

But back to defaultbuffermaxrows. Previously the dtfs had default values of 10000 for DefaultBufferMaxRows and 10485760 for DefaultBufferSize. I added a SQL task to SUM the iSeries column sizes, TCREAS and TCDESC in this example, and set the DefaultBufferMaxRows by dividing the SUM of the columns max_length into 10485760. But I did not see a performance improvement. Do you think that redefining the columns as varchar(100) for the insert is significant? Should I possibly SUM the actual number of columns (2) as 2x100 or SUM the 30x100?