Intelligent Operations: Challenges and Solutions with the IoTDB Time‑Series Database

Introduction: In the transition from mobile/consumer internet to industrial internet, data challenges arise in intelligent operations (AIOps), a big‑data‑driven industry. Cloudwise, China’s largest unicorn in this field, shares real‑world practices on selecting and designing a time‑series database.

Intelligent Operations Overview: Describes the evolution from traditional network management to modern AIOps, the market leaders, and the importance of metrics, logs, and tracing data. Highlights the massive growth of IoT and time‑series data in digital transformation.

Data Challenges: Lists six key challenges—large volume, missing data, peak‑valley fluctuations, out‑of‑order data, granularity inconsistency, and single‑point explosion—and explains why they matter for IT/OT monitoring.

IoTDB Value: Presents a benchmark comparing ClickHouse, IoTDB, and TDengine on a low‑spec machine (Intel i7‑16C, 32 GB RAM, CentOS 7.4, HDD). IoTDB achieves 2.332 M points/s write speed with 81 % compression, handles out‑of‑order data, supports missing‑value filling, and offers efficient query and aggregation.

Architecture: Shows Cloudwise’s intelligent operations platform architecture—Kafka for data ingestion, MySQL for metadata, IoTDB and a proprietary DODB for time‑series storage, and upper‑layer services for metric management, anomaly detection, prediction, log and event analysis, plus a TensorFlow‑based AI engine.

Open‑Source Contributions: Describes contributions to Apache IoTDB, including a Prometheus long‑term storage backend that maps Prometheus labels to IoTDB tags, enabling longer retention and higher compression.

Case Studies: (1) Real‑time anomaly detection for a national bank handling millions of metrics, featuring self‑learning change detection, trend adaptation, batch adaptation, and physical‑world metric support. (2) Anomaly detection for a telecom operator’s number‑portability service, illustrating log‑to‑metric conversion and root‑cause analysis.

Q&A: Answers whether IoTDB can replace Prometheus (it complements it for long‑term storage), outlines the anomaly‑detection algorithms (statistical, STL‑based, machine‑learning), and confirms the use of a gateway‑based distributed IoTDB deployment.

Conclusion: Emphasizes the team’s expertise in AIOps, collaborations with top universities and research institutes, and invites further contact.