Case Study: Pandora achieves sub-second, real-time dashboard queries over hundreds of billions of rows with MemSQL

Using MemSQL to Query Hundreds of Billions of Rows in a Dashboard

Pandora’s Ad sales team needed a usable, interactive dashboard to explore current and historical campaign data, but their existing system imposed a two-day lag, relied on pre-computed answers that prevented ad-hoc queries, and required long nightly processing windows. The project demanded sub-second, real-time queries on a rolling multi-year window (hundreds of billions of rows), standard SQL and drivers, production maturity and support, scalability, and willingness to invest in a reliable solution.

After evaluating finalists (Redshift, CitusDB) Pandora selected MemSQL and worked closely with the vendor to build ingestion paths from Hive, Postgres and files, tune columnstore vs. rowstore usage, and size hardware for SSD/CPU and RAID10 redundancy. The result: sub-second interactive queries over hundreds of billions of rows, sustained high ingest rates (150k inserts/sec from one machine, ~500k/sec with several writers), ~85–90% columnar compression, and a production cluster now holding nearly four years of data with no observed performance degradation; MemSQL has since been used on additional projects at Pandora.

Pandora