Meitu Technology Salon: Evolution of the Big Data Platform, Distributed Bitmap (Naix), and Apache Kylin

The Meitu Internet Technology Salon gathered senior big‑data experts to discuss the challenges and solutions of building and operating large‑scale data platforms in the mobile‑internet era.

Key questions raised: common pitfalls in big‑data platform construction, issues when adopting open‑source technologies, deeper user exploration via big data, and the current stage of Meitu’s data platform.

Speakers and topics:

Lu Rongbin, Meitu’s Director of Big Data Technology, presented the end‑to‑end architecture of Meitu’s data platform, covering data collection (collector → Kafka → Hadoop), a Lambda architecture with DataBus for real‑time processing, and the “Data Workshop” that exposes data services such as the user‑profile platform (Van Gogh), Data Talk, and channel tracking.

He highlighted platform stability measures, including Hive‑SQL validation, cluster‑level and permission‑level safeguards, and offered practical advice: prioritize data quality, build the platform early, and strengthen fault‑tolerance.

Future focus areas include enhancing OLAP capabilities, advancing platform intelligence and stability, and implementing resource isolation and dynamic scaling.

Yang Yaqiang, Meitu’s Senior Big Data Architect, introduced Meitu’s distributed bitmap solution “Naix”. He explained why bitmap uses minimal storage, demonstrated performance advantages over Hive, and described Naix’s three‑layer architecture (external call layer, core node layer, external storage layer) along with its three core data structures: index group, index, and dictionary.

Naix enables millisecond‑level device/user positioning, multi‑dimensional cross‑analysis, and high‑concurrency query handling, and is already deployed in dozens of Meitu apps.

Huang Zhenxian, Senior Architect at Meizu, shared the Meizu User Insight Platform. He described the evolution from a basic user‑profile system (200+ dimensions) to a comprehensive insight platform (1000+ dimensions) supporting offline/real‑time tags, statistical/algorithmic tags, and single/multi‑value tags. Storage choices include Elasticsearch for fast queries, HBase for general profiling, and Redis for performance‑critical queries.

Li Dong, Apache Kylin PMC and Kyligence partner, gave an overview of Apache Kylin’s OLAP capabilities, ANSI‑SQL interface, and integration with BI tools. He cited large‑scale use cases such as eBay, Baidu, JD.com, and telecom operators, emphasizing Kylin’s high performance, concurrency, and query push‑down.

He also demonstrated Apache Superset, an open‑source visualization front‑end that integrates seamlessly with Kylin, showcasing features like sorting, time filtering, HAVING clauses, pagination, and Docker deployment.

The salon concluded with a recap of the four speakers’ insights and an invitation to view the full video replay and PPT (search “Meitu Salon0811” on the Meitu Tech WeChat public account).