ClickHouse is fast, proven by some of the biggest names in tech, and has built its reputation as a top-tier analytical database. But when we ran identical workloads on identical hardware against Exasol, the gap was striking: Exasol delivered 10.7x better median performance across the industry-standard TPC-H benchmark at 100GB scale. On complex analytical queries – the kind with multi-table joins and correlated subqueries that show up in real enterprise workloads-Exasol was 207x faster on the worst ClickHouse queries.

This isn’t theoretical. This is the difference between sub-second interactivity and waiting 48 seconds for a single query to finish.

What ClickHouse Does Well

ClickHouse didn’t become one of the most popular analytical databases by accident. When Uber, Cloudflare, and Bloomberg run your database in production at massive scale, you’ve built real credibility. ClickHouse performs well on the workloads it was designed for – scanning billions of rows for aggregations, time-series analytics, and log data with excellent storage efficiency.

The open-source model helped, too. No vendor lock-in, transparent development, an active community, and years of production deployments gave engineering teams confidence to deploy it for their specific needs. For straightforward analytical queries on denormalized data – the kind powering real-time dashboards and event analytics – ClickHouse delivers what it promises.

The Performance Gap Emerges

Being optimized for fast aggregations on denormalized data isn’t the same thing as being optimized for complex analytical queries across normalized schemas. That’s where the performance story changes dramatically.

TPC-H isn’t just another benchmark – it’s specifically designed to test the hard parts of real data warehouses: eight-table joins, correlated subqueries, NOT EXISTS predicates across multiple dimensions, complex filtering with OR conditions. These patterns stress the optimizer’s ability to generate efficient execution plans, choose optimal join orders, and manage memory under complex query loads.

ClickHouse excels when queries are simple and data is pre-aggregated. But when you throw complex multi-table joins and subqueries at it – the kind of queries that power BI dashboards, financial reports, and exploratory analytics in enterprise environments – the performance profile shifts.

We benchmarked both databases on identical AWS hardware (r6id.8xlarge: 32 vCPUs, 247GB RAM, local NVMe storage) using the latest stable releases: Exasol 2025.1.0 and ClickHouse 25.9.5.21. Every query executed with proper warmup runs and 7 measured runs for statistical validity.

Key Findings

Overall Performance: 10.7x Faster Median Runtime

Exasol delivered a median query runtime of 238.1ms versus ClickHouse’s 2,546.6ms across all 22 TPC-H queries. That’s not a marginal difference – it’s an order of magnitude gap that translates directly to user experience, dashboard responsiveness, and infrastructure efficiency.

Performance on Complex Queries

The gap widens significantly on complex analytical queries:

• 207x faster on Q17 (subquery with aggregation – Exasol: 28ms, ClickHouse: 5,822ms)
• 121x faster on Q21 (NOT EXISTS with correlated subqueries – Exasol: 385ms, ClickHouse: 46,707ms)
• 107x faster on Q08 (8-table join – Exasol: 74ms, ClickHouse: 7,909ms)
• 86x faster on Q19 (complex OR predicates – Exasol: 26ms, ClickHouse: 2,208ms)
• 44x faster on Q05 (6-table join – Exasol: 203ms, ClickHouse: 8,885ms)

ClickHouse took 46.7 seconds to execute Q21, a query that Exasol finished in 385 milliseconds. That’s the difference between interactive analytics and multi-second delays.

Performance Consistency

Beyond raw performance, Exasol demonstrated greater predictability. The maximum query runtime across all 22 queries was 946ms on Exasol versus 47,985ms (nearly 48 seconds) on ClickHouse. For production analytics systems with SLAs, that 50x difference in worst-case performance matters for system reliability and user experience.

Benchmark Configuration

Systems Under Test

Exasol 2025.1.0

Exasol is a purpose-built analytical database management system designed for high-performance data warehouse and analytical workloads. This benchmark used:

• Deployment: Native c4 cluster manager (v4.28.2) installation

• Database RAM: 220GB allocated from 247.7GB system memory

• Optimizer Mode: Analytical (optimized for complex queries over single-row operations)

• Storage: Local NVMe SSD with direct data placement

• Custom Parameters:

• writeTouchInit=1: Initialize pages on write for consistent performance

• cacheMonitorLimit=0: Maximize buffer cache utilization

• maxOverallSlbUsageRatio=0.95: Allow 95% of SLB memory usage

• useQueryCache=0: Disable query cache to measure actual execution performance

• joinOrderMethod=0: Use cost-based join order optimization

ClickHouse 25.9.5.21

ClickHouse is an open-source column-oriented database management system for online analytical processing. This benchmark used:

• Deployment: Native APT installation on Ubuntu 22.04

• Memory Limit: 230GB total memory limit (~93% of system memory)

• Max Threads: 32 (matching instance vCPU count)

• Max Memory Per Query: 200GB for large join operations

• External Processing Thresholds:

• max_bytes_before_external_group_by: 100GB

• max_bytes_before_external_sort: 100GB

• Storage: Local NVMe SSD with MergeTree engine tables

• Query Variant: Standard TPC-H queries without ClickHouse-specific optimizations

Infrastructure

Both database systems ran on dedicated, identically configured AWS EC2 instances:

• Instance Type: r6id.8xlarge (memory-optimized with local NVMe storage)
• Region: eu-west-1
• CPU: Intel Xeon Platinum 8375C @ 2.90GHz, 32 vCPUs (16 physical cores with hyperthreading)
• Memory: 247.7GB RAM
• Storage: 1.7TB NVMe local SSD (up to 650,000 IOPS)
• Operating System: Ubuntu 22.04 LTS
• Network: 25 Gbps network bandwidth

The r6id.8xlarge instance type was chosen for its balance of memory capacity, CPU performance, and high-speed local storage – important factors for analytical database performance. Local NVMe SSDs provide consistent, low-latency I/O without the network overhead of EBS volumes.

Workload Configuration

• Benchmark: TPC-H
• Scale Factor: 100 (approximately 100GB of data across 8 tables)
• Data Generator: dbgen (official TPC-H data generator)
• Data Format: CSV files loaded natively into each system
• Queries: All 22 TPC-H queries (Q01-Q22)
• Warmup Runs: 1 per query (excluded from results)
• Measured Runs: 7 per query for statistical validity
• Total Query Executions: 308 queries (2 systems x 22 queries x 7 runs)
• Metric Reported: Median execution time (robust to outliers)
• Execution Date: October 31, 2025

The TPC-H benchmark represents realistic analytical workloads including complex joins, aggregations, subqueries, and a mix of data access patterns. Scale factor 100 provides a substantial dataset size that exercises database optimizers and execution engines while remaining practical for comprehensive testing.

Overall Performance Results

System-Wide Performance Summary

Across all 22 TPC-H queries, Exasol demonstrated consistently higher performance:

Metric	Exasol	ClickHouse	Exasol Advantage
Median Runtime	238.1ms	2,546.6ms	10.7x faster
Average Runtime	301.1ms	5,378.4ms	17.9x faster
Minimum Runtime	25.2ms	162.6ms	6.5x faster
Maximum Runtime	946.5ms	47,984.7ms	50.7x faster

The average runtime gap (17.9x) exceeded the median gap (10.7x) due to ClickHouse’s extremely low performance on specific queries like Q21 (46.7 seconds) and Q09 (12.5 seconds). These outliers significantly skewed the average while having less impact on the median.

Performance Consistency

Beyond raw speed, Exasol exhibited remarkably stable performance:

• Exasol’s runtime range: 25.2ms to 946.5ms (37.6x spread)
• ClickHouse’s runtime range: 162.6ms to 47,984.7ms (295x spread)

The slowest query (Q21) took 946 milliseconds on Exasol versus nearly 48 seconds on ClickHouse. This difference in worst-case performance makes Exasol far more predictable for production workloads where service level agreements (SLAs) matter.

Performance by Query Category

The benchmark queries fall into three categories based on their analytical patterns:

Aggregation Queries (Q01, Q06, Q12, Q14, Q15, Q19, Q20) These queries focus on data reduction through grouping and aggregation operations.

• Exasol median: 82.2ms
• ClickHouse median: 390.0ms
• Exasol advantage: 4.7x faster

Join-Heavy Queries (Q02, Q05, Q08, Q09, Q10, Q11, Q21, Q22) These queries involve multiple large table joins with complex predicates.

• Exasol median: 178.0ms
• ClickHouse median: 5,596.1ms
• Exasol advantage: 31.4x faster

Complex Analytical Queries (Q03, Q04, Q07, Q13, Q16, Q17, Q18) These queries combine multiple operations including joins, aggregations, and sorting.

• Exasol median: 324.7ms
• ClickHouse median: 4,894.5ms
• Exasol advantage: 15.1x faster

Exasol showed particular strength in join-heavy workloads, where it outperformed ClickHouse by more than 30x on average. This reflects Exasol’s design: sophisticated join optimization and cost-based query planning for complex analytical queries.

Query-Level Performance Analysis

Largest Performance Advantages for Exasol

The following queries showed the most performance differences:

Q17: Small-Quantity Order Revenue (207.9x faster)

• Exasol: 28.0ms median
• ClickHouse: 5,821.7ms median

This query involves a correlated subquery with aggregation. Exasol’s optimizer efficiently decorrelates the subquery and generates an optimal join plan. ClickHouse’s execution plan appears to materialize intermediate results multiple times, leading to excessive computation.

Q21: Suppliers Who Kept Orders Waiting (121.5x faster)

• Exasol: 384.5ms median
• ClickHouse: 46,707.2ms median (46.7 seconds)

Q21 uses multiple NOT EXISTS subqueries with correlated predicates across four tables. This pattern stresses the optimizer’s ability to transform subqueries into efficient join operations. Exasol’s subquery optimization handles this elegantly. ClickHouse’s execution took nearly 48 seconds – problematic for interactive analytics.

Q08: National Market Share (107.3x faster)

• Exasol: 73.7ms median
• ClickHouse: 7,908.5ms median

Q08 joins all eight TPC-H tables with complex filtering and grouping. Exasol’s cost-based optimizer selected an efficient join order and execution strategy. ClickHouse struggled with the join complexity.

Q19: Discounted Revenue (86.2x faster)

• Exasol: 25.6ms median
• ClickHouse: 2,207.7ms median

This query features complex OR predicates with multiple conditions on different columns. Exasol’s predicate pushdown and filter optimization reduced the data volume processed. ClickHouse’s execution suggests less effective predicate evaluation.

Q05: Local Supplier Volume (43.8x faster)

• Exasol: 203.1ms median
• ClickHouse: 8,884.8ms median

A six-table join with filtering on region and date. Exasol’s join order optimization and parallel execution excelled. ClickHouse’s performance suggests suboptimal join ordering.

Other Significant Advantages:

• Q04: 42.6x faster (60.7ms vs 2,586.4ms) – EXISTS subquery optimization
• Q07: 18.5x faster (265.1ms vs 4,894.5ms) – Multi-table analytical query
• Q09: 13.3x faster (942.8ms vs 12,545.6ms) – Complex multi-year analysis
• Q03: 12.5x faster (324.7ms vs 4,059.9ms) – Joins with ordering and top-N

Where ClickHouse Was Competitive

ClickHouse performed comparably on a few queries:

Q15: Top Supplier Query (ClickHouse 1.1x faster)

• Exasol: 379.6ms median
• ClickHouse: 346.4ms median

Q15 creates a temporary view with aggregation and then joins it. ClickHouse’s materialization strategy worked well for this pattern. This was the only query where ClickHouse was faster.

Q16: Parts/Supplier Relationship (Exasol 1.1x faster)

• Exasol: 457.7ms median
• ClickHouse: 517.5ms median

Close performance on this grouping query with high cardinality. Both systems handled it efficiently.

Q20: Potential Part Promotion (Exasol 1.4x faster)

• Exasol: 277.6ms median
• ClickHouse: 390.0ms median

Another competitive result on an IN subquery pattern.

ClickHouse achieved competitive performance (within 2x of Exasol) on only 3 of 22 queries. On the remaining 19 queries, Exasol’s advantage ranged from 3x to 207x.

Technical Insights: Why Exasol Excels

Join Optimization

The most striking performance differences appeared on join-heavy queries, where Exasol was 31x faster on average. Exasol’s sophisticated join optimization includes:

1. Cost-Based Join Ordering: Exasol’s optimizer uses detailed statistics and cost models to evaluate multiple join orders and select the optimal execution plan. On Q05 (6-table join), Q08 (8-table join), and Q21 (4-table join with subqueries), this produced better plans than ClickHouse’s optimizer.
2. Parallel Hash Joins: Exasol distributes hash join operations across all CPU cores with intelligent data distribution. The system minimizes data movement while maximizing parallelism.
3. Join Method Selection: Exasol dynamically chooses between hash joins, merge joins, and nested loop joins based on data characteristics, available memory, and join selectivity.

Subquery Processing

Queries with correlated subqueries (Q04, Q17, Q21) showed Exasol advantages ranging from 42x to 207x. Exasol’s subquery handling features:

1. Subquery Decorrelation: The optimizer transforms correlated subqueries into efficient join operations where possible, eliminating repeated subquery execution.
2. EXISTS/NOT EXISTS Optimization: Particularly evident in Q21’s 121x advantage, Exasol converts EXISTS predicates into semi-joins and NOT EXISTS into anti-joins with optimal execution strategies.
3. Predicate Pushdown: Exasol aggressively pushes predicates into subqueries and joins, reducing intermediate result sizes early in execution.

Memory-Efficient Execution

Despite both systems having access to 200GB+ memory per query, Exasol’s maximum query runtime was 946ms versus ClickHouse’s 47,984ms. Key factors:

1. Better Memory Management: Exasol efficiently manages buffer pools and intermediate results without spilling to disk or thrashing caches.
2. Streaming Execution: Rather than materializing large intermediate results, Exasol uses pipelined execution where possible, reducing memory pressure.
3. Adaptive Execution: Exasol monitors memory usage during execution and adapts strategies dynamically.

Predicate and Filter Optimization

On queries with complex filtering (Q19: 86x faster), Exasol demonstrated advanced predicate evaluation:

1. Filter Ordering: Exasol evaluates the most selective predicates first, minimizing unnecessary computation.
2. SIMD Vectorization: Modern CPU SIMD instructions accelerate predicate evaluation on large data volumes.
3. Bloom Filters: Runtime bloom filters eliminate non-matching data early in join processing.

Consistent Performance

The coefficient of variation (standard deviation / mean) was consistently lower for Exasol across almost all queries. Contributing factors:

1. Stable Query Plans: Exasol’s optimizer produces consistent execution plans across runs.
2. Predictable Resource Usage: Memory and CPU utilization remained stable without unexpected spikes.
3. Effective Caching: Warmup runs successfully primed caches, and subsequent runs showed minimal variation.

Key Takeaways

When to Choose Exasol

This benchmark demonstrates Exasol’s clear advantages for:

1. Join-Heavy Analytical Workloads: If your queries regularly join multiple large tables (5+ tables), Exasol’s 31x advantage on join-heavy queries translates to faster dashboards, reports, and ad-hoc analysis.
2. Complex Subquery Patterns: Workloads with correlated subqueries, EXISTS/NOT EXISTS patterns, or multi-level nesting benefit from Exasol’s advanced optimizer. The 121x advantage on Q21 exemplifies this.
3. Predictable Performance Requirements: When SLAs matter and worst-case performance affects user experience, Exasol’s consistency (946ms max vs 48 seconds max) provides production-grade reliability.
4. Mixed Query Complexity: Real-world analytical workloads combine simple aggregations with complex multi-table analysis. Exasol’s across-the-board performance advantage (10.7x median) means faster performance regardless of query complexity.
5. Enterprise Data Warehousing: Organizations needing robust, predictable, high-performance analytics at scale will appreciate Exasol’s combination of speed and consistency.

Performance vs. Cost Considerations

On identical AWS infrastructure, Exasol delivered:

• 10.7x faster median performance – completing workloads in 1/10th the time
• 50x better worst-case performance – ensuring consistent user experience
• Lower infrastructure costs – achieving the same throughput with fewer resources

The performance advantage means Exasol can handle the same workload on smaller infrastructure or process 10x more queries on the same hardware. Organizations with hundreds of concurrent users or thousands of daily queries can achieve substantial cost savings.

Benchmark Validity and Reproducibility

This benchmark followed rigorous methodology:

• Identical hardware – eliminating infrastructure as a variable
• Latest stable releases – representing current production capabilities
• Standard workload – TPC-H queries used industry-wide
• Statistical rigor – 7 runs per query with median reporting
• Warmup runs – ensuring fair cache states
• Full transparency – complete results and configuration available

The results are reproducible and represent real-world performance characteristics both systems would exhibit on production analytical workloads.

Resources and Full Results

Complete Benchmark Report

The full interactive benchmark report with all query-by-query results, visualizations, and detailed statistics is available at:

https://exasol.github.io/benchkit/exa_vs_ch_100g_1to1/reports/2-results/REPORT.html

The report includes:

• Interactive Visualizations: Performance heatmaps, box plots, CDF charts, and speedup comparisons
• Query-by-Query Tables: Detailed statistics for every query including median, mean, standard deviation, min, max
• System Specifications: Complete hardware and software configuration details
• Raw Result Data: All 308 query execution times available for independent analysis

Benchmark Methodology

All queries were executed sequentially on a single connection to measure single-user performance. Each query executed with:

• 1 warmup run (excluded from results)
• 7 measured runs (median reported to minimize outlier impact)
• Identical data across both systems
• Same query SQL where possible (standard TPC-H queries)

The median metric provides robust performance measurement resistant to occasional system variance or background process interference.

System Versions Tested

• Exasol: 2025.1.0 with c4 cluster manager v4.28.2
• ClickHouse: 25.9.5.21 (official stable release)
• Operating System: Ubuntu 22.04 LTS
• Infrastructure: AWS eu-west-1, r6id.8xlarge instances
• Test Date: October 31, 2025

Conclusion

This comprehensive benchmark demonstrates Exasol’s substantial performance advantage over ClickHouse on analytical workloads. The 10.7x faster median performance, 207x advantage on complex queries, and 50x better worst-case consistency make Exasol the clear choice for enterprise analytical workloads where performance, predictability, and reliability matter.

Organizations evaluating analytical database systems for data warehousing, business intelligence, or advanced analytics can use these results for decision-making. The performance differences observed aren’t marginal – they represent fundamental architectural advantages in query optimization, join processing, and execution planning that translate directly to faster insights, better user experience, and lower infrastructure costs.

For more information about Exasol or to discuss how these performance characteristics apply to your specific workload, visit www.exasol.com or contact our team directly.

All benchmark results, configurations, and raw data are available for independent verification at the links provided above. This benchmark was conducted independently using standard TPC-H methodology on identical cloud infrastructure.