ClickHouse Crushes Databricks in Real-Time Analytics Cost

Real-Time Analytics: A Cost Showdown

The article presents a compelling case for ClickHouse Cloud's superior performance and cost-efficiency in real-time analytics, particularly highlighting its integrated approach to data preparation and query serving. The benchmark's methodology, focusing on continuous ingestion and a holistic 'performance-per-dollar' metric encompassing preparation, query cost, and runtime, provides a robust foundation for its claims. The detailed breakdown of costs for both platforms, from ingestion and clustering to materialized view refreshes and query execution, is particularly valuable for potential adopters. The core innovation lies in how ClickHouse Cloud appears to tightly couple these stages, minimizing overhead compared to Databricks' more segmented architecture with separate services for ingestion, clustering, and materialized views.

However, a crucial limitation to acknowledge is the exclusion of Databricks' Lakehouse//RT, which is explicitly stated as being in beta and unavailable for the test. This new offering is positioned by Databricks as their specific solution for real-time workloads, and its absence means the comparison is against Databricks Serverless SQL, which may not be the direct competitor to ClickHouse Cloud's real-time focus. While the article acknowledges this, it significantly impacts the perceived completeness of the comparison. Furthermore, while the benchmark is well-defined, real-world scenarios can introduce complexities like varying data quality, unpredictable query patterns, and diverse integration requirements that might alter the observed performance differentials. The pricing model used is based on list prices, and actual cloud bills can fluctuate due to reserved instances, regional pricing variations, and specific usage patterns. Therefore, while the 752x figure is striking, users should conduct their own tailored benchmarks.

Key Points

• ClickHouse Cloud demonstrated 752x better end-to-end performance per dollar compared to Databricks Serverless SQL in a continuous ingestion benchmark of 113.2 billion stock quotes.
• The performance advantage stems from ClickHouse Cloud's integrated approach to data preparation (columnar storage, ordering, pre-aggregation) running on the same ingest nodes, minimizing overhead.
• Databricks' architecture, with separate services for ingestion (Zerobus), clustering (liquid clustering/predictive optimization), and materialized view refreshes, incurs higher preparation and query costs.
• Key cost components analyzed include preparation cost, normalized query cost, and accumulated query runtime, with ClickHouse significantly outperforming Databricks in all three.
• The benchmark excluded Databricks' beta Lakehouse//RT, which is designed for real-time workloads, meaning the comparison is against a potentially less optimized Databricks offering for this specific use case.

---

📖 Source: Why Databricks can’t match ClickHouse Cloud for real-time analytics