In real-time bidding (RTB), every bid request must complete within 100 milliseconds — missing that window forfeits the ad impression. This case study describes how an interactive advertising company eliminated the operations and maintenance (O&M) burden of a self-managed HBase cluster and cut P99.9 response latency by 90% by migrating to Lindorm in two stages.
Challenges
The RTB system faced three compounding pressures:
Strict latency SLA: Bid requests must be submitted within 100 ms. Any response that exceeds this threshold forfeits the ad impression and directly impacts revenue.
High-concurrency reads: Precise ad targeting requires reading large volumes of user persona data at high concurrency within the bidding window.
Real-time write and analytics: The system must ingest user behavior data generated from ad responses immediately, then aggregate it across multiple dimensions to surface advertising effectiveness in real time so that bidding strategies can be adjusted.
Solution
The company migrated in two stages.
Stage 1 — Self-managed HBase to ApsaraDB for HBase Standard Edition
Moving from an on-premises data center to ApsaraDB for HBase Standard Edition reduced resource overhead and improved P99.9 response time, while offloading hardware provisioning and day-to-day O&M to Alibaba Cloud.
Stage 2 — ApsaraDB for HBase Standard Edition to Lindorm
Migrating from ApsaraDB for HBase Standard Edition to Lindorm instances of equivalent specifications reduced P99.9 response latency by a further 90%.
The following diagram shows the overall migration path.
Benefits
Reduced latency: After migrating to Lindorm, response time (RT) for business requests became more stable. P99.9 response latency dropped by 90%, keeping bid requests reliably within the 100 ms window and protecting ad revenue.
Managed O&M: Alibaba Cloud provides service level agreement (SLA) guarantees and a complete portfolio of data management services, freeing the company from O&M.
Expert support: A professional Lindorm expert team provides free technical support, allowing the company to focus on business growth rather than database operations.
Business continuity: The dual service architecture keeps primary and secondary clusters in the same region. On a primary cluster failure, workloads on the failed primary node are immediately switched to a secondary node, improving the stability of the bid submission system.