Case Study: Building a High‑Performance Advertising Platform with ClickHouse Enterprise

Introduction – With the rapid growth of the digital advertising industry, constructing an efficient ad‑delivery platform has become crucial for market share and brand visibility. EasyPoint, a leading digital advertising solutions provider, leveraged ClickHouse Enterprise to build a high‑performance, stable platform.

Company Overview – EasyPoint is a technology‑driven international smart‑marketing service company serving over 2,000 clients worldwide, including Alibaba, Tencent, ByteDance, and others, across e‑commerce, tools, content distribution, and gaming.

Platform and Business Introduction – The ad‑delivery platform aggregates real‑time and offline traffic, generates cost and performance reports, recommends bidding decisions, and manages media and DSPs, providing essential data support for operations.

Existing Data Architecture – Data flows through Kafka pipelines, with MirrorMaker replicating from Singapore to a central US cluster. From there, data is split: part stored in Alibaba Cloud OSS for long‑term archiving, and part fed into the Druid OLAP engine for analytics. When Druid data becomes inaccurate, a Lambda architecture using Spark cleanses data from OSS and reloads it into Druid.

Problems with Druid – Complex maintenance, limited SQL support, restrictive data model definitions, windowing constraints, and difficult scaling lead to performance bottlenecks and OOM issues.

Advantages of ClickHouse over Druid – ClickHouse offers diverse storage engines, materialized views, columnar storage for superior OLAP performance, vectorized and parallel query execution, high compression ratios, and excellent high‑concurrency write capabilities.

ClickHouse Enterprise Architecture – The enterprise edition adopts a cloud‑native compute‑storage separation design, providing strong data consistency, horizontal scalability, and simplified node addition without downtime. Serverless resource management automatically scales resources based on load, reducing operational complexity.

Performance Tests

OSS Data Import – Parallel client imports achieve up to 2 million rows per second.

ClickHouse Enterprise vs. Spark – For a 60 billion‑row dataset with 10 dimensions and JSON parsing, ClickHouse consumes far less resources while delivering comparable aggregation speed.

Community vs. Enterprise CK – Enterprise CK is 2–3× faster across aggregation, distinct‑count, and multi‑dimensional aggregation scenarios, and avoids OOM occurrences.

Join Performance – Enterprise CK processes a 175 billion‑row table join with a 47 billion‑row table in 27 seconds without OOM; materialized‑view joins achieve sub‑second latency.

Summary – In advertising workloads, performance, stability, and resource efficiency are paramount. ClickHouse Enterprise delivers superior data ingestion, aggregation, and join performance, automatic serverless scaling for stability, and consumes only ~7.5 % of the resources required by Spark, offering significant cost savings while meeting high‑throughput demands.

Future plans include full commercial rollout of ClickHouse Enterprise in production environments to further enhance ad‑business performance and cost efficiency.