TL;DR:
A leading telecommunication company's enterprise big data platform, StreamCloud, chose Apache Doris as its core database solution for ingesting and querying trillions of incremental data daily. Currently, this solution has been deployed in more than ten production scenarios. The largest cluster is deployed on 117 high-performance server nodes and has been operating stably for over six months. Its single table contains over 13 petabytes of raw data and 534 trillion records, maintaining a daily data ingestion of about 145TB and peak loads of about 158TB during holidays.
The Challenges: High Costs and Insufficient Capabilities of the Previous OLAP System
The previous architecture:
For high write throughput, sub-second response to billions of records, and JOIN queries for massive data, the ClickHouse-powered OLAP system faced significant challenges:
- Unstable Data Ingestion: ZSTD with high compression ratio led to frequent "too many parts" errors and data ingestion backlogs.
- High Storage Costs: To ensure stable operations, LZ4 with lower compression ratio was adopted, leading to higher storage costs.
- High Operation and Maintenance Costs: uneven data distribution across nodes and manual data migration increased costs.
- Poor Concurrent Query Performance: query performance degraded in high-concurrency queries.
- Limited JOIN Capabilities: ClickHouse's architectural constraints blocked multi-table/large-table joins.
The Search for A Solution: The Migration to Apache Doris
Apache Doris vs. ClickHouse Comparison Tests
We used three physical servers to simulate production workloads. The results:
- Prefix Index: Doris achieved over 2x faster query speeds compared to ClickHouse.
- Secondary Index (BloomFilter Index & Inverted Index): Doris achieved 2x faster query speeds with BloomFilter Index and over 5x faster query speeds with Inverted Index, compared to ClickHouse.
- Full Table Scan: Doris had comparable performance.
In summary, Doris would accelerate queries by over 2x.
Apache Doris Replacement Practice
We replaced the ClickHouse-powered log storage and analytics engine with Apache Doris. During stress testing, we solved batch write failures, Compaction overloads, and import errors, and saved 28% ingest servers with Broker Load.
The Results: A Stable, Efficient OLAP System with Apache Doris
This company has replaced ClickHouse with Apache Doris to build a new OLAP system, achieving 100+ servers and a peak throughput of 158TB/day. Doris improved the compression ratio per replica by 4x and has been operating for over 6 months. The improvements:
- Reduced Hardware Costs: Doris Broker Load achieved 28% savings in hardware costs, and ZSTD's high compression ratio reduced storage usage by 6% without write bottlenecks compared to ClickHouse's LZ4 compression.
- Superior Query Performance: Doris accelerated multi-table joins, achieving 2x faster SQL queries and nearly 30% improvements in batch query performance.
- Simplified Operation: Doris has high availability during failures. The built-in Web UI and Grafana efficiently detected abnormal nodes and disk failures.
Building on these improvements, this company will leverage Doris further:
- Advancing Doris-Powered Lakehouse: Doris Hive Catalog unifies data warehouse resources and data access interfaces, achieving queries and analysis across all datasets.
- Accelerating Complex Queries: Doris accelerates complex queries, such as multi-dimensional analytics and aggregation computations.
- Reducing Storage Costs: with tiered storage and other functions, Doris reduces storage costs while improving query performance.
Talk to Us
If you want to bring similar (or even higher) performance improvements and benefits to your data platform, or just explore further on Apache Doris, you are more than welcome to join the Apache Doris community, where you can connect with other users facing similar challenges and get access to professional technical advice and support.
If you're exploring fully-managed, cloud-native options, you can reach out to the VeloDB team!
