ETL AND ELT

ETL AND ELT

By W H Inmon

Once upon a time people wanted to build a data warehouse. In order to do this they had to bring together many different data bases and files. The problem was that these files and data bases were never designed to be combined or integrated together. But if a data warehouse was to be built, it was nevertheless necessary to bring these files and data bases together.

The first attempt at this massive integration effort was to write individual programs that merged the files. During the merger the data in the files and data bases were integrated.

Then one day someone noticed that the mainline logic of these integration programs was all the same. Someone had the bright idea to write a universal integrator, where only the specifics of integration were done. This program was affectionately called ETL.

Soon there appeared whole companies that were building and selling their own versions of ETL.

And soon data warehouses began to pop up like weeds in the springtime.

The large vendors never supported data warehouse (IBM, et al). But soon their customers thrust data warehouses upon them. The large vendors and consulting companies begrudgingly began to build data warehouses.

But in short order the vendors and consultants discovered that data had to be integrated in order to build a data warehouse. And integrating data required that they get their hands dirty, kind of like planting tomatoes in the springtime. (Is there anyone who doesn’t get their hands dirty when planting tomatoes in their back yard?)

And vendors and consultants don’t like to get their hands dirty. They looked at ETL and proclaimed – the hard part of ETL is the T part. If we just didn’t have to do the T part, we could handle this with clean hands,

So the large vendors invented ELT. Now ELT sounds like ETL. An unsuspecting customer might not know the difference. And so the large vendors shuffled off the T part to someone else. Not my job. And my hands don’t get dirty.

And guess what? The T part never got done. It was just inconvenient to do integration of data.

It became someone else’s job, but they never could find the person to do the job.

The upshot is – if you wanted to just move some data around, then ELT is your thing. But if you want believable data, then you have to do ETL.

So the choice is yours. Do you want data quickly and easily, that may be essentially unreliable? Or do you want data that forms a firm foundation for – AI, analytics, data mesh, ML, et al?

No free enchiladas here.

?

Bill Inmon lives in Denver with his wife and his two Scotty dogs – Jeb and Lena. Jeb and Lena have had their grooming, taking off their winter coats. They like to show off how pretty they look.

?