Compared to pure ETL or pure ELT, I find mixed-mode ETL (mixing some pushdown capabilities into the overall ETL process) more appealing. Whereas ETL usually brings better organization of the data load process, in-database processing is generally faster than ETL, so the ability to send part of the process (part of an Informatica "mapping," for example) into the database is attractive. The standard work-around today, when we need to do some heavy lifting in the database, is to write, say, stored procedures invoked from an ETL process. The down side of this approach - besides having to develop logic in two environments, ETL and database SQL - is that often it is not easy making the mid-stream data available to the stored procedure, leading to convoluted architecture and ETL design compromises. If you can code uniformly in one environment and then just send part of that code down to the database, you gain flexibility as well as performance.
It's getting increasingly hard to stay excited about ETL (Indisputable Need + Near Ubiquity ≠ Enthrallment) but pushdown seems like the best thing that has happened to ETL since data quality and EII, and I can see mixed-mode options, such as that offered by Informatica, giving ETL solution architects and designers a needed boost in designing sophisticated ETL solutions and containing data load jobs within processing windows. I also fully expect that pushdown will become a new frontier in the battle for ETL supremacy - and once again, Informatica seems to have the edge.
But please, can we just continue calling all that ETL without coining a new acronym for a clever improvisation?There's no doubt that ELT - yes, that's extract-load-transform (also called "pushdown") not conventional extract-transform-load (ETL) - is now a mainstream capability. Informatica's inclusion of pushdown optimization in the recently released PowerCenter version 8.5 brings ELT the legitimacy it deserves... I fully expect pushdown will be come a new frontier in the battle for ETL supremacy.