Push Notification Fan-Out at Scale

Campaign

Devices to notify 250,000

Fan-Out Shape

Worker shards (W) 8

Push rate per shard 1000 msg/s

Delivery failure / retry rate 8%

Live Status

0 / 0 delivered

Backlog queued

Effective throughput

0 /s

Elapsed

0.0 s

ETA to finish

—

How it works

A campaign of N devices is split evenly across W worker shards, each capped at a push-gateway rate limit of R messages/second (this is exactly how FCM/APNs-backed fan-out services throttle senders). A fraction p of attempts fail (offline device, expired token, transient 5xx) and are re-queued on the same shard instead of leaving the system:

ideal time      T_ideal = N / (W · R)
retry overhead  attempts per device ≈ 1 / (1 − p)
effective time  T_eff ≈ N / ((1 − p) · W · R)

Each simulated tick, every shard drains up to R·dt messages from its queue; (1 − p) of that leaves as delivered, p flows straight back into the same queue. That is why the backlog bars shrink slower — or plateau — as the failure slider climbs: more of every processed message re-enters the queue instead of exiting the system.

Worker shards — more parallel shards linearly cuts completion time, exactly like adding more sender workers behind a mobile backend's notification service.
Push rate per shard — models the per-connection rate limit a push gateway enforces on each sending worker.
Failure / retry rate — models unreachable devices and gateway errors; the backlog bars and the streaming particles both slow down as this rises, showing why retry storms are the real scalability risk in fan-out systems, not raw device count.

Particles are a proportional visualization of message flow (density scales with active shard count and current throughput), not a literal one-particle-per-device rendering — real campaigns push far more messages per second than a screen can draw individually.

Dispatcher → shard Worker shard (bar = backlog) High backlog / retry pressure

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.