ClickHouse Cloud: TPC-H Under a Penny!

Cost-Performance Revolution in Analytics

The article makes a compelling case for ClickHouse Cloud's cost-performance advantage in TPC-H benchmarks, particularly at smaller scales (SF10), demonstrating remarkable cost efficiency for analytical workloads. The raw performance on SF100 is competitive with major cloud data warehouses, even when using a single node against potentially larger or more distributed configurations of competitors. The transparency in sharing benchmark scripts and methodology is commendable, fostering trust and reproducibility. The focus on join engineering as the driver for these improvements is a key takeaway, highlighting ClickHouse's evolution beyond its historically perceived limitations in complex analytical queries.

However, the article's strength in showcasing cost-performance at SF10 might inadvertently raise questions about scalability for truly massive datasets (SF1000 and beyond), although the authors acknowledge this as the next engineering frontier. The comparison, while detailed, relies on specific configurations chosen by ClickHouse, and while they attempt to find comparable hardware, the inherent differences in cloud warehouse architectures (e.g., serverless auto-scaling vs. fixed nodes) mean direct, apples-to-apples hardware comparisons are inherently difficult. The emphasis on 'hot runs' and disabling result caching is crucial for measuring raw compute, but real-world performance can be influenced by these factors. The 'less than a cent' claim for SF10 is a powerful marketing point, but it's specifically for compute cost and excludes data ingestion, storage, and other potential associated costs, which users will need to factor in.

This benchmark is highly beneficial for organizations seeking to optimize their analytical spending without compromising significantly on query performance. Developers and data engineers working with join-intensive analytical queries, especially those on a budget or facing cost pressures, will find this information particularly valuable. The article also serves as a strong indicator of ClickHouse's advancements in handling complex analytical tasks, making it a more viable contender in the cloud data warehousing space. For businesses evaluating cloud data warehouse solutions, this provides a data-driven perspective that strongly favors ClickHouse Cloud for specific use cases focused on cost-efficiency and competitive raw performance.

Key Points

• ClickHouse Cloud demonstrates competitive raw runtime on TPC-H SF100 against major cloud data warehouses using a single 59-core node.
• It achieves the top rank in cost-performance for TPC-H SF100, significantly outperforming competitors.
• At TPC-H SF10, the entire workload runs in 2.9 seconds for a compute cost of $0.009 (less than one cent), leading in both speed and cost-performance.
• The improvements are attributed to two years of focused join engineering, including faster hash joins, better planning, and runtime filters.
• Benchmark methodology is transparent, with scripts and results available on GitHub for reproducibility.

---

📖 Source: TPC-H for less than a cent: ClickHouse Cloud vs. Snowflake, Databricks, BigQuery, and Redshift