Must Know Cold
Tier 1
- BI vs DW vs Data Warehousing — analysis/technologies vs integrated historical store vs processes that build/maintain it.
- Inmon/CIF — subject-oriented, integrated, time-variant, non-volatile; central normalized enterprise DW is the integration point.
- Kimball — dimensional business-process fact areas/marts; facts measure, dimensions describe; conformed dimensions/facts and bus create enterprise integration.
- Staging vs ODS vs DW vs Mart vs Cube — prepare vs current-integrated vs historical-enterprise vs targeted analytical presentation vs multidimensional access.
- Atomic vs summarized — preserve useful detail, summarize/optimize after requirements are known.
- Historical load vs recurring update — baseline/backfill vs repeated delta maintenance.
- Five CDC methods — timestamp, log table, transaction log, message delta, full comparison.
- Low-latency patterns — Trickle = source; Messaging = bus; Streaming = target.
- Six activities — Requirements → Architecture → Develop DW/Marts → Populate → BI Portfolio → Maintain Data Products.
- Mapping vs remediation vs transformation — document route/rules vs fix defect vs intentionally convert/derive.
- Metadata/dictionary vs lineage vs impact — meaning vs provenance/path vs change consequence.
- Operational/tactical vs strategic BI — current workflow/short horizon vs historical trend/long-term performance.
- Governed self-service — user flexibility inside trusted semantic/security/support guardrails.
- Usage vs coverage vs performance vs satisfaction — use it? cover it? run it? value it?
Tier 2 — understand deeply
- Why ODS is lower latency, shorter history, and more volatile than DW.
- Why Inmon and Kimball have the same integrated analytical objective but different integration mechanisms.
- Why conformed dimensions + bus make Kimball marts enterprise-integrated rather than independent.
- Why population is often a major share of DW effort.
- Why requirements begin with business questions/KPIs rather than source tables.
- Why source-to-target mappings are lineage evidence.
- Why optimistic loading requires reconciliation controls and pessimistic loading requires recycle/reload controls.
- Why self-service requires Metadata, security, governance, training, and support.
- Why actual connected/query activity is better adoption evidence than licenses.
Tier 3 — selective memorization
- CIF components: applications, staging, integration, ODS, DW, marts/OpDM, BI/reporting, supporting Master/Reference/external data.
- ROLAP / MOLAP / HOLAP.
- CDC trade-offs and delete behavior.
- Readiness, roadmap, configuration management, culture/business commitment, support.
- Reporting strategy dimensions.
Common traps
| Trap | Correction |
|---|---|
| ODS is just another DW | ODS is current/near-current, low latency, shorter history, more volatile. |
| Kimball = independent marts | Conformed dimensions/facts + bus provide integration. |
| Summarize first for performance | Preserve required atomic detail; optimize based on actual requirements/usage. |
| Self-service means no governance | Governed self-service depends on trusted data, shared semantics, access, Metadata and support. |
| Registered users prove adoption | Actual connections/query activity are stronger. |
| Warehouse exists only for BI | Historical data also supports operational analysis, compliance/evidence and other uses. |
Source anchor: pp. 361–393.