Designing OLAP Systems

Engine-level architecture decisions, assuming modeling theory from S7. The core pre-aggregation-vs-raw tradeoff, choosing rollup granularity, materialized-view strategies and their consistency models, denormalization on columnar engines, late-arriving data and restatements, upsert architectures across Pinot/ClickHouse/Iceberg, designing for federation, designing for real-time with per-layer latency budgets, multi-engine architectures, and plotting cost-vs-latency curves. Ends with a capstone designing three valid OLAP architectures for one workload.