August 8, 2014 at 2:41 am
I have business process that we report on - Fitting room appointments. stores setup fitting rooms on their systems, book in appointments types on their calendar and assign staff to the fitting rooms to help with dress try-ons.
The dimensions for these are very simple ie. 3-4 columns max but have about 1000 rows each as the data for all the store. Is it over kill to go through the process creating staging and live tables and building etl packages to populate StoreUsers, AppointmentTypes and FittingRooms dimensions?
I thought about using a Junk dimension but if i understand there purpose correctly they would hold every combination of the the 3 tables possible.
Thanks
August 13, 2014 at 2:41 am
How were you thinking of populating your dimension tables without creating ETL packages? Whether you need to create staging tables depends on a number of things such as:
- how frequently your ETL process runs.
- how much data is captured form your source systems.
- what sort of resource pressure the ETL process places on your source systems. (Staging tables are often used for just the EL part of ETL, i.e. get the data out of the source systems as quickly as possible therefore minimising the impact on those source systems).
- do you have Slowly Changing Dimensions and if so, which type?
Another benefit of using staging tables is that in the event of a failure in loading data into your Data Warehouse, you don't have to back and extract data from the source systems again - you just go back to the staging tables.
It's perfectly possible and normal to have dimension tables with far less than 1000 rows.
Regards
Lempster
August 13, 2014 at 8:02 am
Hi
Thanks for the reply. When i say creating etl packages im really talking about complex ones with lookups, conditional splits derived colums inserts and updates etc. This business process im talking about is just one of a few processes, the rest of which properly use etl and staging tables for the exact reasons you mention i.e. load and scd.
Im new to DW so im just not sure if you have to create complex etl package and analyse scd's requierments on all data needed. or is it sufficient to just have an etl that drops and recreates the data in some dimensions for example for an appointment status dimension that only has 10 rows do you really need to create a complex package that track type 1 & 2 changes?
August 14, 2014 at 2:48 am
Whether you need to track changes or not should stem from the business requirements regardless of how many rows are in a dimension table. During the dimensional modelling phase, those requirements need to be captured. Of course, the situation that often arises is that when asked the question 'do you need to track history for this attribute' the business will say 'yes' to everything!
This is why it is key for a DWH project to be successful that you really get your business users engaged in the process. Having a good Business Analyst on board can really help in that regard.
There is no definitive answer to your question, but hopefully what I've said helps a little bit...? 😉
Regards
Lempster
August 15, 2014 at 6:44 am
Is it over kill to go through the process creating staging and live tables and building etl packages to populate StoreUsers, AppointmentTypes and FittingRooms dimensions?
No, it's not overkill. The ETL package(s) [some might disagree, but all three could be part of the same package] might not all have the same features that one for daily use would have, but you need something. It doesn't even have to be part of the automated download.
November 14, 2014 at 5:21 am
If you are going to drop and rebuild you dimensions, you need to be really careful that the keys are always alocated in the same order or you could end up associating the facts with the wrong dimension records.
I would say that all tables in the DW model should be populated using SSIS packages as you then have a single point of entry. The packages might just call a stored procedure, but at least you know where to look for the start point. If some of it runs from SSIS, others from agent jobs, some is inserted directly by the source system through a linked server then maintenance can become a nightmare. Also if it is all done by SSIS you can build a framework that does logging a failure handling which you can't do if something else is drivent the ETL process.
Viewing 6 posts - 1 through 5 (of 5 total)
You must be logged in to reply to this topic. Login to reply