Explain the internal working of HANA's in-memory computing engine?

HANA's engine keeps table data resident in RAM, organized primarily in columnar format, and distributes query processing across multiple CPU cores in parallel — a single large aggregation query can be split into pieces, each processed by a different core simultaneously, then combined into the final result.

flowchart TD
    A[Query received] --> B[Query optimizer builds execution plan]
    B --> C[Work split across multiple CPU cores in parallel]
    C --> D[Each core processes its portion of in-memory columnar data]
    D --> E[Partial results combined into final result set]

Persistence is handled asynchronously: changes are written to a redo log immediately (for durability against a crash) and periodically checkpointed to disk in the background, but the actual working data a query touches stays resident in RAM rather than being fetched from disk on demand. This combination — in-memory data, columnar organization, and massively parallel execution across cores — is what lets HANA process queries that would take minutes on a traditional database in a fraction of a second.