2023 vivo Developer Conference: Internet Technology Session Technical Practices

On November 1, 2023, the vivo Developer Conference Internet Technology Session was successfully held. Eight technical experts from vivo Internet shared their technical practice achievements and insights in microservices, big data, cloud computing, DevOps, event platforms, and audio/video areas.

Microservices Architecture: vivo's microservice platform serves over 500 million users in more than 60 countries/regions, with tens of thousands of microservices covering hundreds of thousands of machines, handling 800 billion RPC calls daily with peak QPS reaching tens of millions. The team formed a matrix of nearly 30 architectural capabilities and adopted a "open-source + self-research" development path.

Big Data Solutions: vivo built a stream-batch dual-active architecture with configurable multi-component output. They introduced async request frameworks to solve millions-level concurrent real-time computing performance issues, achieving second-level latency during Redis dynamic expansion. The dynamic micro-batch computing architecture improved batch computing performance by over 60% through Shuffle Hash Join optimization.

Container Mixed Deployment: vivo's mixed deployment platform achieved powerful scheduling and isolation capabilities. The "three-step" strategy of making tasks run smoothly, stably, and intelligently enabled nearly 20,000 schedulable tasks with over 40,000 daily scheduling executions, providing an additional 20,000 cores and 50TB memory during peak hours.

Low-Code Platform (Wukong System): The self-developed Wukong system uses SaaS+PaaS model to cover 100% of vivo's business scenarios. The BCA model (Behavior, Condition, Action) combined with the playmaker editor enables dynamic activity definition generation through a low-code system.

Short Video Optimization: vivo implemented multiple optimization strategies including connection multiplexing, keep-alive, HTTPDNS, multi-threaded fragment downloading, and pre-rendering for zero-latency first frame. They explored fixed, dynamic, and intelligent pre-loading strategies while establishing a BI layered monitoring system. For cost optimization, they used PCDN, adaptive encoding, and utilization governance to reduce playback costs while maintaining quality.