Data Platforms

Alexander Polomodov and Nikolay Golov discuss how analytical platforms turn data into value. The story starts with a centralized warehouse and one team building marts for the organization. This model controls architecture but becomes a bottleneck, leading companies toward hybrid self-service.

Small companies can rely on analytical replicas of production databases until load reaches a limit. Architecture then moves from OLTP to MPP and columnar engines. Batch inserts, parallel queries, and scaling cost matter: a database that can store the volume may still fail under concurrent use by many analysts.

The next transition separates storage from compute. Data Mesh distributes ownership, but chains of data products add latency and cognitive load; a base domain can stabilize shared entities. The platform stores data in object storage and connects compute engines for different workloads.

Athena demonstrates serverless queries over S3, while ClickHouse, Trino, DuckDB, and Snowflake show forms of independent compute. Personal compute adds flexibility but raises consistency and cost questions. Five-year-old constraints cannot be trusted forever: hardware, engines, and economics change, so architecture must be revisited.