02 — Exam Map & High-Yield Targets
Chapter weight: 6%. Expect terminology discrimination, architecture-pattern recognition, lifecycle sequencing and scenario choices rather than deep implementation coding.
Must know cold
- Integration: movement/consolidation of data.
- Interoperability: systems’ ability to communicate.
- ETL / ELT: transform before load vs load raw then transform in target.
- Latency: time between source generation/change and target availability.
- CDC: detect/transmit only changes; data-based or log-based.
- Interaction: point-to-point, hub-and-spoke, publish-subscribe.
- Coupling: tight/synchronous dependency vs loose/asynchronous independence.
- Canonical model: common exchange format that reduces pairwise transformations.
- Lifecycle: Plan & Analyze → Design → Develop → Implement & Monitor.
- Governance: business owns meaning; Metadata/lineage preserve trust.
Must distinguish
| Pair / family | Deciding clue |
|---|---|
| ETL vs ELT | Where transformation occurs relative to load |
| Snapshot vs Delta | Point-in-time set vs changes since prior transfer |
| CDC vs Replication | Detect/pass changes vs keep synchronized copies |
| Batch vs Micro-batch | Scheduled larger sets vs same model at higher frequency |
| Async vs Sync | Sender continues vs waits for confirmation |
| Point-to-point vs Hub | Direct pairwise vs exchange through common hub |
| Mapping vs Orchestration | What becomes what vs what runs when/depends on what |
| Discovery vs Profiling | Find candidate sources vs inspect actual contents/fitness |
| Lineage vs Mapping | Where data came/went/changed vs source-target conversion rule |
| Federation vs Virtualization | Combined access vs one logical DB/view over heterogeneous stores |
| EAI / ESB / SOA / CEP | App-interface integration / message bus / reusable services / event-pattern intelligence |
| Migration vs Integration | Controlled conversion for app change vs recurring exchange |
| Virtualization vs Warehouse | Virtual access vs persisted integrated history |
| Sharing Agreement vs Interface Spec | Responsibilities/use/service obligations vs technical structure |
Must apply
- Use batch when scheduled high-volume movement satisfies latency.
- Use micro-batch/CDC/event when freshness requires it.
- Prefer async/loose when a producer must continue through receiver outage.
- Accept sync/tight when the transaction truly cannot complete without immediate downstream confirmation.
- Move toward hub/canonical/reusable services when interface proliferation and semantic inconsistency are the problem.
- Profile actual source/target data before design assumptions become production defects.
- Treat mapping and transformation-rule changes as business-governed Metadata.
- Establish Data Sharing Agreements/MOUs for governed cross-boundary exchange.
- Choose virtualization/federation when unified access is needed without mandatory physical consolidation.
- Choose a physical hub/warehouse when persisted history, source-independent availability or predictable performance require it.
Selective memorization
- Four business-rule categories: definitions, facts, constraints/action assertions, derivations.
- Five Plan & Analyze activities: requirements, discovery, lineage, profiling, business rules.
- Five Develop activities: data services, data flows, migration, publication, CEP.
- Seven tool categories: ETL/transformation, virtualization server, ESB, rules engine, data/process modeling, profiling, Metadata repository.
- Metric families: availability; volumes/speed; cost/complexity.
60-second route
- ETL vs ELT.
- Snapshot vs delta vs CDC vs replication.
- Batch vs micro-batch vs event/streaming.
- Async vs sync.
- Point-to-point vs hub vs pub-sub.
- Mapping vs orchestration.
- Discovery vs profiling vs lineage.
- EAI vs ESB vs SOA vs CEP.
- Federation/virtualization vs physical persistence.
- Business ownership of transformations + sharing agreement + Metadata/lineage.
Changed-fact drills
- ETL → ELT: transform before target load; change fact to raw landing first with target-side transformation.
- Snapshot → Delta/CDC → Replication: all current rows; change to only inserts/updates/deletes; change again to “keep the same dataset synchronized elsewhere.”
- Batch → Micro-batch → Event-driven: overnight deadline; change to five-minute freshness; change again to immediate action as events arrive.
- Async → Sync: producer must continue through downstream outage; change to transaction cannot complete without immediate confirmation.
- Point-to-point → Hub/canonical: two simple systems; change to many systems with inconsistent pairwise mappings.
- Mapping → Orchestration:
STATUS 1 → ACTIVE; change to “Customer must finish before Order.” - Discovery → Profiling → Lineage: locate candidate fields; change to actual null/pattern statistics; change again to origin and downstream path.
- ESB → CEP: route messages; change to interpret multiple streams and trigger fraud action.
- Virtualization → Physical hub/warehouse: unified view without copying; change to persisted history/source independence/performance.
- Sharing Agreement → Technical spec: permitted use/responsibility/SLA; change to fields, mappings, endpoint/protocol.
Readiness rule: recognition is not mastery. Explain the fact that changes the architecture answer.