eArk WP6 - reference implementation: generate CSV from archived database
Provide some kind of DWH functionality for archived databases. There are various formats of archived databases, e.g. SIARD. These are still sort of relational by providing meta-data on the original relations. The first step/idea is to inline all relations into a huge, de-normalised table which can be exported as CSV or loaded into HBase later.
See also Implementation Notes.
- HBase export
- id column
- try multiple FKs recursive
Unsorted Notes fom E-Ark Technical Meeting
- smallest table is (ID, value), always inline these tables.
- Foreign key is first candidate for a dimension
- consider problems of inconsistent data (which we always have in DWH)
- use fuzzy dimensions for data that is like 90% a proper dimension according the other rules
- for determining dimensions, calculate the strength of the discriminator of the values of the column(s).
- e.g. gender is a strong discriminator (it separates 50/50)
- e.g. 10/90 is maybe not as good
- good dimensions are strong hierarchical structures, a dimension has at least 2 to 3 levels, a strong hierarchical structure might have 6-7 levels - these are worth extracting and working with
- try to identify dimensions by data, e.g. timestamps, geo coordinates, etc (by regex)
- "factless fact tables" exist and only contain counts of occurances
- get archived database formats, start with Estonian buildings database
- understand given formats (XML/CSV) and load
- create data model using given sample data
- experiment with ways to automatically follow foreign keys
- define requirements for AIP level 2 (Manfred/HKI) to be useful
- understand other formats like SIARD and EPP and HKI defined one.
Future Ideas for Providing OLAP
- see if zohmg or Mondrian can help analysing the data
- how to we need to put data into it?
- maybe run a Mondrian on each node using a different table as facts
- send queries to CHX and Solr and HBase parallel and reduce results for user
- See Model
- See Sakila