Proven at scale
See how far TBMQ scales — 100 million connections, millions of messages per second, and single-digit-millisecond latency. Every test is open and reproducible.
TBMQ is proven across the three core MQTT messaging patterns — fan-in, fan-out, and point-to-point.
Fan-in · Scale
A 25-node cluster sustained 100M+ concurrent MQTT clients emulating smart trackers while processing 6M messages per second.
Fan-out · Throughput
One 32-vCPU TBMQ node handled a 3M msg/sec fan-out — 100 publishers to 3,000 subscribers — with CPU headroom to spare.
Point-to-point · Delivery
Persistent DEVICE clients scaled linearly from 200k to 1M msg/sec with QoS 1 guaranteed delivery, backed by Redis and Kafka.
The numbers are a consequence of the architecture — not a lucky benchmark run.
No node-to-node communication. Add nodes and throughput scales linearly — the 1M msg/sec P2P test proved it phase by phase.
Messages are persisted in Kafka, not just buffered in memory. The 100M test moved 11.4 billion messages with 100% delivery.
Persistent DEVICE clients get reliable, ordered delivery even after disconnects — without overloading the database.
A Netty-based, event-driven I/O layer holds millions of connections per node without a thread for each — the reason one cluster reaches 100M.
The full 100-million-connection cluster runs on AWS for roughly $26,500 a month — before Savings Plans and reserved-instance discounts. That is a cost efficiency few brokers can match.
Illustrative AWS us-east-1 pricing. See the full breakdown in the 100M connections report.
We publish the test scripts, load configurations, and exact hardware specs behind every result. Run them against your own workload.
Spin up TBMQ and benchmark it against your own workload. The same broker that handles 100 million connections runs on a single laptop for development.
