A Query Engine for Aggregates
Cube Store Architecture and Performance Benchmark
This ebook covers Cube Store's architecture and a TPC-H-derived performance benchmark. Learn how the engine serves pre-aggregated data, then explore all 43 reported query-and-scale measurements, the methodology, and what the results establish.
Architecture and benchmark results
How the engine works, what the benchmark measured, and how to evaluate your own workload.
Aggregate awareness
How measures, dimensions, filters, and time granularity determine whether a query can reuse a rollup.
Sorted indexes
Why physical order matters for filtering and grouping, and what each additional index costs to build and store.
Pruning and local data
How partition ranges and Parquet statistics narrow reads, while warmup makes data available locally to workers.
Execution and refresh
The roles of the router, workers, and coordinator, and the resources needed to keep prepared data current.
Performance benchmark
All 43 reported query-and-scale combinations from 11 adapted TPC-H queries, including setup and measurement limits.
Workload evaluation
A practical sequence for checking aggregate coverage, inspecting indexes and partitions, and testing concurrent traffic.
For teams building interactive analytics
Internal dashboards, embedded analytics, and agent investigations all need prepared data to stay available and respond quickly. This guide explains the engineering behind Cube Store and how to evaluate it against the query patterns and freshness requirements of your application.
18 pages, free. The benchmark reports medians of 173–620 ms and p90 values of 291–879 ms, including network time. It measures individual query latency, not concurrent throughput.
Get the ebook
Read the architecture, full benchmark tables, and workload evaluation guide.