Architecture and Indexing Mechanism of the Car Smart Investment Advertising System

The article begins by outlining the core business workflow of the Car Smart Investment advertising system, including configuration of ad placements, storage of ads in an index library, request handling, real‑time bidding, and reporting.

It then introduces the index system, explaining the data‑投放 model where ad plans contain groups, each mapped to a numeric docid using an auto‑increment scheme stored in a hash table.

The index architecture employs multiple threads—main, Kafka, MySQL, and Dump—to handle request processing, real‑time updates, full‑data synchronization, and periodic backups.

The logical index structure consists of four parts: (1) a docid‑to‑ad‑group mapping via a hash table; (2) a forward table (array) storing detailed ad information for fast random access; (3) an inverted hash table combined with bitmap bitmaps for rapid condition matching; and (4) a docid status bitmap indicating valid or invalid entries.

Index creation is illustrated with a step‑by‑step example, showing docid assignment, bitmap updates for targeting attributes (region, keyword, device), and insertion of ad data into the forward table.

Index modification is described as a real‑time process that disables old bitmap entries, recomputes new attribute bitmaps, and re‑enables the status bitmap to keep the index consistent with business changes.

The search procedure is explained: the ad engine forwards targeting conditions to the index service, which performs inverted bitmap retrieval, forward‑table filtering, and returns matching ad groups.

Finally, performance testing shows the system can respond within 10 ms under billions of requests, sustain up to 80 k QPS, and handle real‑time updates of hundreds of thousands of ad records.