Apache Doris, a popular real-time data warehouse, ranked first in the latest RTABench results, setting a new benchmark for real-time analytics performance. In standardized tests, Doris delivered up to 6 times the performance of ClickHouse, 30 times that of PostgreSQL, and 100 times that of MongoDB.
RTABench (Real-Time Analytics Benchmark) is specifically designed for real-time analytics, offering a more realistic evaluation of database performance than traditional benchmarks. It features normalized multi-table models, real-world operational data sets, and queries that are more representative of real analytical scenarios. Apache Doris once again proved its ability to handle complex, high-concurrency analytical queries with better speed and efficiency.
What is RTABench?
RTABench is an open-source benchmarking tool designed specifically for real-time analytics (RTA). Built on the ClickBench framework, it introduces more realistic data models and query sets that closely mirror real-world business workloads. The benchmark offers a more relevant standard for evaluating the capabilities of real-time analytics databases, addressing the gaps in traditional benchmarks when it comes to testing real-time performance.
Official website: https://rtabench.com/
The Results: Apache Doris Leads the Real-Time Analytics Benchmark
In the latest RTABench test, Apache Doris delivered:
On a c6a.4xlarge instance with 500GB gp2 storage, Apache Doris achieved a query performance score of just ×1.28, ranking number one. Its performance was nearly 4× faster than TimescaleDB, 6× faster than ClickHouse, 30× faster than PostgreSQL, and 100× faster than MongoDB.
When tested using the same machine type, Apache Doris outperformed many traditional databases by more than 10 times. Even compared to ClickHouse, one of the popular real-time analytics databases in recent years, Doris still shows a clear performance advantage.
Why Apache Doris Leads Real-Time Analytics: The Tech Behind Performance
Continuous innovation in system architecture, execution engine, and query optimization was the key to Apache Doris' success in the RTABench, outperforming TimescaleDB, ClickHouse, MongoDB, PostgreSQL, and other systems.
-
MPP Architecture and Columnar Storage Doris adopts a Massively Parallel Processing (MPP) architecture combined with columnar storage and efficient compression algorithms. This design distributes workloads across nodes for parallel execution, while reading only the necessary columns during queries. The result: far lower I/O overhead and a strong foundation for real-time analytics at scale.
-
Pipeline Execution Engine Doris breaks down queries into smaller pipeline tasks that run concurrently, taking full advantage of multi-core CPUs computing power to improve query performance.
- Limits query thread count to avoid the thread explosion problem common in traditional engines
- Reduces data copying and sharing, lowering system overhead
- Targeted optimization on core operations like sorting and aggregation, significantly boosts query efficiency and overall throughput.
-
Vectorized Execution Doris supports batch data processing (vectorized execution), using the full power of modern CPU instruction sets to reduce function call overhead. Combined with the Pipeline Execution Engine, it further accelerates the response time of complex analytical queries.
-
Smart Optimizer (CBO + RBO) The optimizer in Doris blends Rule-Based Optimization (RBO) with Cost-Based Optimization (CBO). This layered strategy ensures Doris consistently generates the most efficient execution plans, even for highly complex queries.
How RTABench Works
RTABench is designed to mirror real-world analytical workloads using a multi-table model, realistic datasets, representative queries, and diverse database categories, making it a reliable and practical benchmark for real-time performance.
1. Normalized Multi-Table Database Model
Unlike traditional benchmarks that rely on a single-table design, RTABench simulates an order tracking system in an e-commerce scenario. It uses a multi-table relational schema with five core tables:
| Table Name | Description |
|---|---|
| Customers | Stores customer details, including name, location, and signup date |
| Products | Contains product catalog information, including pricing and stock levels |
| Orders | Tracks orders placed by customers |
| Order_Items | Records the products included in each order |
| Order_Events | Tracks order status changes (e.g., created, shipped, delivered) |
This structure better reflects real-world applications where cross-table joins are common, avoiding the oversimplification of single-table designs and providing a more accurate evaluation of database performance in real-time workloads.
2. Realistic and Moderate Dataset
The RTABench dataset strikes a balance between realism and practical test execution. Its scale is:
- ~171 million order event records (high-frequency core data)
- 1,102 customers
- 9,255 products
- ~10.01 million orders
This dataset is large enough to stress-test databases' real-time processing capabilities, yet not so big that it makes benchmarking too expensive or difficult to scale.
3. Query Types and Scenario Simulation
RTABench provides 31 representative queries, covering the most typical query patterns seen in real-time analytics. They fall into four categories:
| Query Type | Main Goal | Example Query |
|---|---|---|
| Raw Event Queries | Validate efficiency of indexes/partitions | Count daily "Departed" shipments for a device |
| Selective Filters | Validate efficiency of indexes/partitions | Get the latest status of a specific order |
| Multi-Table Joins | Simulate complex business logic & analysis | Calculate revenue per customer over the past 30 days |
| Pre-Aggregated Queries | Compare raw vs. pre-computed query performance | See how many shipments were delayed each month, using pre-aggregated results |
This design provides a clear comparison of databases in terms of both flexibility (in handling raw queries) and performance (speed of pre-aggregations), aligning benchmarks more closely with real-world business needs.
Apache Doris: A New Leader in Real-Time Analytics
RTABench has shown that Apache Doris stands out as one of the most reliable databases for real-time analytics. It excels at:
- Handling both multi-table joins and single-table lookups
- Delivering consistently outstanding query performance
- Offering needed scalability to process massive volumes of real-time data
Whether you're a fast-moving startup or a large enterprise, Apache Doris can help your business to gain data insights (on a millisecond level) and stay ahead of the competition. Further explore Apache Doris by joining the Apache Doris community on Slack.
For enterprise or cloud deployments of Apache Doris, VeloDB Cloud offers all the performance benefits highlighted above, plus compute-storage separation, fully-managed SaaS, multi-cloud support, and enterprise-grade security and compliance. For more information, please get in touch with the VeloDB team.
