Meteor: A Real-Time Computation Platform Based on Storm for Ctrip Marketing

Marketing scenarios at Ctrip require diverse real‑time computation resources, leading to complex topology management, high deployment cost, and security risks.

To address these challenges, Ctrip’s Marketing R&D built Meteor, a platform that wraps Apache Storm (and Spark) with a unified driver layer, data‑operation layer, graph‑computation layer, service layer and application layer.

The core of Meteor is a directed acyclic graph (DAG) of computation nodes (spouts, bolts) that can be created, compiled, packaged and deployed automatically; the Meteor Service registers, schedules and monitors these graphs via REST APIs.

Key features include high availability of topologies and containers, secondary scheduling, resource isolation, and automated build‑publish‑deploy pipelines, enabling rapid creation, modification and scaling of marketing scenarios.

In practice, Meteor reduces development time from days to hours, supports tens of thousands of scenarios, achieves sub‑50 ms node response latency, and handles billions of daily queries, greatly improving marketing efficiency.

Overall, Meteor demonstrates how a customized stream‑processing platform can transform marketing data pipelines into a scalable, automated, and maintainable system.