StarRocks was designed as an OLAP which means it was designed to answer analytical queries. While it’s true that it has some features like primary key table, which are a common feature in OLTP systems, StarRocks’ primary key table was implemented to better support real time streaming inserts and upserts OLAP use cases. StarRocks was not designed to execute large volume inserts/upserts in the 10 to 30ms range that are common in OLTP.
| Characteristic | OLTP | OLAP (AWS Redshift) | OLAP (Snowflake GCP Big Query) | OLAP (StarRocks) |
|---|---|---|---|---|
| Query complexity | Simple | Complex | Complex | Complex |
| Purpose | Transaction processing | Data analysis | Data analysis | Data analysis |
| Performance requirements | High throughput, low latency | High throughput, high latency | High throughput, high latency | High throughput, high latency |
| Data model | Relational | Multidimensional | Multidimensional | Multidimensional |
| Typical workload | Insert, update, delete | Select, aggregate, group by | Select, aggregate, group by | Select, aggregate, group by |
| Deployment Architecture | Compute with attached storage | Separated compute and storage architecture | Separated compute and storage architecture | |
| ACID | Commit or rollback on a record level | Commit or rollback on a batch level | ||
| Comments | Support JOIN queries in normalized and denormalized data, ACID ingestion, mysql wire protocol emulation, Data warehouse with local disk or Data Warehouse on top of data lake (S3 with Apache Iceberg, Apache Hudi, Apache Hive or Delta Lake), mutable data support (update, upsert, delete) through the primary key table, immutable data support through aggregate or duplicate key table |