InMobi Turbocharges Reporting: ClickHouse Delivers 20x Speed, 80% Cost Cut

From Lag to Lightning: InMobi's ClickHouse Transformation

The InMobi case study powerfully demonstrates the transformative potential of ClickHouse for real-time analytics, particularly in high-throughput, low-latency scenarios common in ad-tech. The shift from a costly serverless warehouse to a more optimized ClickHouse architecture resulted in a dramatic 20x query speed improvement (P99 latency from over 60 seconds to under 3 seconds) and an 80% cost reduction, saving $32,000 monthly. This is a compelling validation of ClickHouse's capabilities in handling massive datasets (10TB) and high query volumes (400,000+ daily) while meeting stringent SLAs. The article effectively highlights the challenges faced with the previous system – latency issues, linear cost scaling with traffic, and infrastructure bloat – and how ClickHouse addressed them through its efficient data compression (5:1 ratio), reduced scan rows (70% reduction), and more predictable infrastructure requirements (60 vCPU/240GB RAM vs. 448 vCPU/3.5TB RAM). The implementation details, such as the separation of read/write services and the atomic ingestion pipeline using staging and verification, provide valuable insights into building robust, production-ready systems on ClickHouse. The proactive approach to downtime fallback and the step-function cost scaling model further underscore a mature engineering effort. Furthermore, the vision of extending this ClickHouse-based platform as a reusable analytical foundation for other InMobi use cases is a significant strategic advantage, indicating scalability and future-proofing.

However, while the article paints a glowing picture, there are nuances to consider. The '80% cost reduction' is impressive, but the article mentions the previous serverless warehouse cost was around $40,000/month, implying a new cost of $8,000/month. This figure is for a specific reporting API and might not represent the total data infrastructure cost for InMobi. The complexity introduced by the staging/verification pipeline for atomic ingestion, while necessary for data integrity, adds operational overhead and requires careful management. The shift to a managed ClickHouse instance also introduces vendor dependency, though this is a common trade-off for reduced operational burden. The success hinges on the team's ability to maintain the optimized ClickHouse configuration and leverage its features effectively. The article doesn't delve deeply into the specific query patterns or the optimization strategies beyond general observations like reduced scan rows, which could be beneficial for other users trying to replicate this success. The reliance on Spark for ingestion, while standard, also represents a component of the overall data pipeline that requires its own maintenance and optimization.

This case study is highly beneficial for ad-tech companies, real-time analytics platforms, and any organization dealing with large-scale data ingestion and querying under strict performance requirements. Developers and architects looking to improve query performance, reduce infrastructure costs, and build scalable analytical systems will find this article particularly valuable. It serves as a strong endorsement for ClickHouse as a viable alternative to traditional data warehouses or other analytical databases for specific high-performance workloads. The lessons learned regarding atomic ingestion, asynchronous connection management, and predictable cost scaling are universally applicable to data engineering challenges. The article's success in demonstrating tangible business outcomes – speed, cost, and reliability – makes it a significant contribution to the discourse on modern data architectures.

Key Points

• InMobi migrated its real-time reporting API from a serverless warehouse to ClickHouse, achieving significant performance and cost improvements.
• P99 latency dropped from over 60 seconds to under 3 seconds, a 20x speed increase.
• Monthly costs were reduced by 80%, from $40,000 to $8,000.
• The system now handles over 400,000 queries daily on 10TB of data.
• Key technical solutions included atomic ingestion pipelines (staging/verification), asynchronous connection management (gevent/greenlets), and a read/write service separation.
• ClickHouse's data compression and efficient scanning contributed to infrastructure reduction and cost savings.
• The project is being leveraged as a reusable analytical foundation for other InMobi use cases.
• Lessons learned emphasize data accuracy over speed in ingestion, the criticality of asynchronous connection management, cost predictability, and essential operational guardrails.

---

📖 Source: InMobi が ClickHouse で実現したクエリ20倍高速化と80%のコスト削減