This page is for people who plan or run an installation with many devices: hundreds or thousands in one organisation. It explains which limits apply to the organisation as a whole, how to keep devices from all connecting or reporting in the same second, and how to change many devices safely. Each feature has its own page; this page links to them instead of repeating them.
Limits that apply to the whole organisation
Every device has its own rate limits from its device profile (see Device profiles). Your plan adds limits for the organisation as a whole. With many devices, these are usually the ones you reach first.
| Limit | What happens when it is reached |
|---|---|
| Devices | A new device is refused, whichever way it is added: New device, import, claim, self-registration or a gateway announcing a new child. Quarantined devices count. Existing devices keep working. |
| Messages per second, data points per minute | Messages over the rate are acknowledged and dropped. Over HTTP the answer is 429 with a Retry-After header. The connection stays open: an organisation limit never disconnects a device. Owners and administrators get at most one notice per hour. |
| Devices connected at once (only when your plan sets it) | A new connection is refused with CONNACK 5. A device that reconnects, or takes over its own session, is not counted twice. |
| Messages per month | Nothing stops. Messages over the allowance are accepted and billed as overage. |
The values for your plan are on Plans and the IoT add-on and Limits and quotas. The monthly count is on the Usage page; see Usage and quotas and Message quota and usage.
Spread reports over time
The organisation’s messages per second are counted in one-second windows, not as an average. Six hundred devices that each report once a minute average 10 messages a second. If they all report at second 0 of the minute, 600 messages arrive in the same second, and every message above the organisation’s per-second limit is dropped.
- Give each device its own offset inside its reporting interval, chosen once (for example derived from its serial number), instead of reporting on the full minute or hour of the clock.
- Batch readings. One message with several records counts as one message; separate messages count one each. Data points still count per value.
- Keep the per-device limits in mind when buffered data is sent after an outage. Replayed readings must also fit the profile’s replay rate, which by default is half the per-second telemetry limit. See MQTT.
Stagger connections
Many devices connect at the same moment after a power cut at a site, a network outage, or a firmware rollout that reboots a wave of devices. Hermetry protects itself against connection floods, and the same protections slow down devices that all arrive at once.
| Protection (defaults) | What the device sees |
|---|---|
| On each broker node, at most 64 connections from one public address may be in the middle of logging in at the same time (256 from one /24 IPv4 or /48 IPv6 network) | Extra connections are closed at once, before the TLS handshake |
| A connection must complete TLS and log in within 5 seconds | The connection is closed |
| More than 10 failed logins (wrong token or password) from one address within 5 minutes block that address for 60 seconds | CONNACK 3, or a closed connection, for every device behind that address |
| A device whose session is taken over more than 5 times within 60 seconds from two different networks | The newer network gets CONNACK 3 for 60 seconds |
| The platform is not ready, or is under load and holds back the organisations sending far more than their share | CONNACK 3 (HTTP 503); connected devices are slowed down, not disconnected, and keep their data |
Good practice:
- Wait a random 0 to 30 seconds before the first connection after a boot or a firmware update. A whole wave of devices then does not arrive in one spike.
- Back off with jitter on every refused or lost connection. The schedules are on Reconnection and back-off.
- Fix wrong credentials quickly. Behind one NAT address, a few misconfigured devices can get the address blocked for every device behind it.
- Give every device its own credential. Two devices sharing one credential take the session from each other; see Device credentials.
- Watch the Log tab of a few devices after a large reconnect: it shows each refusal and disconnect with its reason.
Add devices in bulk
Add many devices from a CSV file on the Imports page. A file holds up to 1,000 rows. Your plan’s device limit is checked for the whole file before anything is written, so a file that does not fit is refused whole.
The new devices’ access tokens and MQTT passwords are offered once, in one file, for 15 minutes after the import, and only to the person who ran it. Decide before you import how those credentials will reach the devices. The steps are in Importing devices in bulk.
Devices that register themselves with a profile’s provisioning key land in quarantine and count toward your device limit while they wait. For devices made in volume, prefer pre-provisioned devices. See Quarantine for self-registered devices and Adding and claiming devices.
Roll out firmware in waves
An assignment to every device of a profile, or to a list of devices, can be sent in waves: a small canary wave of online devices first, then larger waves, each with a success gate and a waiting time. The rollout pauses by itself when a wave goes worse than you allow. See Rolling out firmware in waves.
- Test a new image first on a few devices you can reach physically, with One device assignments.
- Keep the waiting time long enough for problems to show. A device that reboots in a loop needs time to fail.
- Every device in a wave reboots after installing and reconnects. The random start delay in the firmware (above) keeps that from becoming one spike.
- Devices on metered links may wait instead of updating. They do not count as failures when a wave closes.
Change settings in waves
Devices that take remote settings are changed in bulk with a baseline and a settings job. A job also works in waves with a success gate. See Remote device settings and Remote settings good practice.
| Limit | Value |
|---|---|
| Jobs running or paused at the same time | 5 per organisation |
| Devices in one job | 5,000 |
| Devices a job sends changes to per minute | 200 |
| Default waves | 1 %, then 10 %, then the rest |
| Single settings changes (Settings tab and API) | 600 per minute per organisation |
The first wave of a settings job is a fixed random cross-section of the selected devices, not the online ones first. Its devices answer only when they are online, so start a job when most of them are connected.
Before you scale up
- Check your plan’s device, message rate and connected-device limits against the number of devices and how often each one reports.
- Make sure the firmware spreads its reports, waits a random time before its first connect, and backs off with jitter.
- Set the device profiles’ rate limits to what the devices really send, with some headroom.
- Set the quota alert thresholds on the Usage page, and subscribe to the
usage.thresholdandusage.overagewebhooks if your own systems should know. - Plan firmware and settings changes as waves, starting with devices you can reach.

