Design and Challenges of Meizu Real-Time Message Push System

The presentation, originally delivered at Meizu's technical open day, describes a real‑time push system serving roughly 25 million online users, handling about 5 billion page views per day and achieving a push throughput of 600 k messages per minute.

The architecture is divided into four logical layers: device access, message distribution (including upstream routing and downstream delivery with a user routing table), subscription management, and storage for both offline and subscription messages.

Key challenges discussed include mobile power consumption, which the team addressed by designing a lightweight IDG protocol (10× faster encoding/decoding than XMPP/SIP) to reduce traffic by 50‑70 % and by implementing adaptive heartbeat intervals to lower battery drain.

Network instability in mobile environments leads to duplicate messages; the solution involves sequence‑based acknowledgments, DNS/IP fallback lists, and client‑side probing to select the optimal server, while also applying delayed responses on overloaded servers.

To support massive concurrent connections (up to 4 million long‑lived connections per machine), the system uses C++ with epoll, memory pools, tcmalloc, and kernel tuning (CPU affinity for NIC interrupts and increased TCP RTO).

Load balancing is achieved without a single LVS point by having clients fetch a prioritized IP list, probe multiple servers, and select the fastest response, complemented by server‑side latency throttling when load thresholds are exceeded.

Monitoring covers error counts, send/receive queue depths, request rates, interface latency, and service availability, with alerts to detect early signs of overload or failure.

Finally, a gray‑release process is described, allowing incremental, user‑transparent deployments and gradual traffic migration to ensure stability before full rollout.