SSIS OLEDB Provider has an option fast loading ... I would prefer using all manipulations (REPLACE,SUMs and etc) on SQL Server when data is already moved Best Regards, Uri Dimant SQL Server MVP http://dimantdatabasesolutions.blogspot.com/ http://sqlblog.com/blogs/uri_dimant/

Hi there, The view will execute every time you request data unless you persist it. Plus with a table you can add indexes and possibly normalize some data. I'd go for table... Hope this helps. Cheers, Michel

Thanks, Gentlemen. I appreciate it. -Karen

Just to compliment Uri's reply: 1st - your goal is to suck the data out to the local storage, this way when you will later manipulate on the data you can use shared memory (when you connect using SQL Server Source with tab lock) or a conn string involving the . , LocalHost or (local) notation, thus using a staging table is beneficial the most indeed. 2nd to load the data fast (not sure if it from another db or file) use the "fast load" option of the OLEDB destination connection as here: http://www.simple-talk.com/sql/reporting-services/using-sql-server-integration-services-to-bulk-load-data/ (btw this is your guide). Also try to maximize the "Maximum insert commit size" property on the Conn Manager property page may be start with 100,000 and revise up.Arthur My Blog

A good package design is one where majority of sorts, aggregations and other blocking transformations are handled by writing SQL queries and not using the tasks. http://www.bidn.com/blogs/Daniel/ssas/1361/ssis-blocking-non-blocking-and-partially-blocking-transformationsPlease vote as helpful or mark as answer, if it helps Cheers, Raunak | t: @raunakjhawar | My Blog

I have an SSIS package that imports a large table into a SQL 2008 table. The table has over 17 million records (FYI, I know the real answer here is to cut down the number of records - working on the end-users to convince them of that...). I then need to do a bunch of manipulation on it - using "replace" on some fields, parsing out some date information, some sums, etc. Is it faster/better to do a "Select Into" a new table or to create a view in this scenario?

Hi there, The view will execute every time you request data unless you persist it. Plus with a table you can add indexes and possibly normalize some data. I'd go for table... Hope this helps. Cheers, Michel