IoT Platform / Connectivity
IoT connectivity management
across a whole fleet
Which bearer the product should use, what the SIM estate costs to run, which protocol earns its overhead, how data survives an outage, and how the transport is secured. These decisions set the running cost of a connected product for its entire life.
Bearer Choice
The bearer decision is
made once and paid for forever.
How a device connects determines the module on the board, the antenna, the power budget, the enclosure, the install procedure, and the monthly bill. Changing it later means a hardware revision and a fleet replacement, so it deserves more than an afternoon.
The question we start with is who controls the site. That single answer eliminates more options than any technical comparison, because a bearer that depends on somebody else keeping their network friendly to your device will eventually stop working for reasons no engineer can fix.
| Signal or link | Cellular | Wi-Fi | LoRa |
|---|---|---|---|
| Range | Wide area | One building | Site or campus |
| Power draw | High in bursts | Moderate | Very low |
| Cost | Running data bill | Low, site owned | Gateway upfront |
| Bandwidth | High | High | A few bytes |
Cellular
Where it fits
Devices that move, devices on sites you do not control, and anything where the install has to work without a network engineer present.
What it costs
A running data bill per device, module and antenna cost on the board, and exposure to operator technology retirement.
What to watch
Coverage at the exact mounting point, power draw during transmit bursts, and how quickly a reconnection loop can burn a monthly allowance.
Wi-Fi
Where it fits
Fixed indoor devices where the site owner is cooperative and a network already exists with capacity to spare.
What it costs
No data bill, but an onboarding step at every install and an ongoing dependency on somebody else changing their password.
What to watch
Guest network isolation, captive portals, firewall egress rules, and what happens to your fleet when the site replaces its router.
Ethernet
Where it fits
Cabinet-mounted equipment in plant rooms and data closets where a cable run is already part of the install.
What it costs
Lowest running cost and the most stable link, at the price of physical installation work and a dependency on site IT.
What to watch
VLAN and firewall policy, static addressing versus DHCP, and having a fallback for when a switch port gets reassigned.
LoRa and LPWAN
Where it fits
Many low-rate sensors across one site or campus, especially where devices run for years on a battery.
What it costs
Gateway hardware and backhaul, plus a hard limit on payload size and message frequency you cannot design around later.
What to watch
Duty cycle limits, whether the application really can live on a few bytes per message, and who owns the gateway estate.
Once the bearer is chosen, the board-level engineering follows: module selection, band support, power sequencing, and antenna work. That is covered on our cellular connectivity design page. This page stays with the fleet-level question of running that choice across thousands of units.
SIM and Data Plan Management
Connectivity is a bill
that grows with the fleet.
At a hundred devices the SIM estate is a spreadsheet. At ten thousand it is a cost centre with its own failure modes. Binding SIM records to device records turns an operator invoice into something you can actually act on.
- SIM identifier bound to device serial
- Data usage tracked against each plan
- Suspend on theft or decommission
- Alerts on sudden usage jumps
- Cost rolled up by product and customer
- Roaming and profile visible per region
SIM to Device Binding
Every SIM identifier is linked to a device serial in the registry, so a data anomaly points at a specific unit and a returned unit points at a specific SIM to suspend.
Usage Against Plan
Data consumed per device compared against the plan it sits on, with the outliers surfaced. A unit using ten times the fleet average is almost always a reconnection loop or a firmware bug.
Lifecycle Control
Activate on first commission, suspend on decommission or theft report, and reactivate on refurbishment. Idle active SIMs are a slow leak that only shows up on the invoice.
Profile and Roaming
Which profile a device is registered on, how often it changes, and whether a roaming agreement is doing what it was bought to do in each region.
Cost Attribution
Data cost rolled up by product, lot, customer, and region, so connectivity spend can be read against the contracts that are meant to cover it.
Anomaly Alerts
Alerts on sudden usage jumps, on SIMs registering from an unexpected country, and on devices that have gone quiet while still consuming data.
Give us your payload size and report interval. We will size the bearer, the protocol, and the monthly data cost.
Book a Technical Session →Protocol Choice
MQTT, HTTP, or bytes
you framed yourself.
Protocol overhead is invisible on a lab bench and obvious on a monthly invoice. The right answer depends on whether the server ever needs to talk first, how often the device reports, and how tight the data budget is.
MQTT
Strengths
Persistent session, low per-message overhead, server-initiated commands, quality of service levels, and a natural fit for many devices publishing small messages often.
Costs
Keeping a session open costs power and keepalive traffic. Broker capacity has to be sized for concurrent connections as well as message rate.
Where we use it
Devices that need to be told something rather than only to report. Charge points, lift controllers, anything with a remote command surface.
HTTP
Strengths
Passes through corporate networks and proxies with the least argument, easy to debug, and simple to batch several readings into one request.
Costs
Header overhead per request is significant on a metered link, and there is no clean path for the server to initiate anything.
Where we use it
Wake, send a batch, sleep. Battery devices on long report intervals and anything installed behind restrictive site IT.
Binary over TCP
Strengths
The smallest frames possible, full control of the wire format, and the lowest data cost per reading at high message rates.
Costs
You own everything: framing, acknowledgements, retransmit, session handling, and a decoder that has to be versioned alongside the firmware.
Where we use it
Very high report rates or very tight data budgets, which is why so much of the tracker world runs on custom binary protocols.
Where a custom wire format is the right call, the framing, acknowledgement, and retransmit design is engineering in its own right. That work is described on our device-to-cloud protocol engineering page.
Buffering and Store-and-Forward
An outage should be
a delay, not a gap.
Field devices lose connection constantly. Lifts sit in shafts, trackers drive through tunnels, plant rooms sit behind a metre of concrete. If the firmware only sends live readings, every one of those events becomes a hole in the record that nobody can reconstruct later.
Store-and-forward is cheap to design in and effectively impossible to add once a fleet is shipped. It also changes what the backend has to accept, since a device coming back online sends historical data interleaved with current data, and the pipeline has to order it correctly rather than treat late arrivals as noise.
Local queue sized against the worst realistic outage, not the average one
Readings stamped at capture time, never at upload time
Retention policy that decides what to drop first when storage fills
Backlog uploaded at a controlled rate so it does not starve live data
Duplicate handling on the server so a retried batch does not double count
Out-of-order ingestion, because a backlog and live data arrive interleaved
Backpressure signalling so a struggling backend can slow devices down
Buffer depth reported as telemetry, so a growing queue is visible before data is lost
Transport Security
Encrypted is not the same
as authenticated.
A surprising number of shipped devices open a TLS connection and then skip certificate validation because the flash was tight or the clock was wrong. That gives an attacker on the path everything they need. These are the layers we hold to.
Transport Encryption
TLS on every link with certificate validation performed on the device. A device that skips validation to save flash has encryption without authentication, which is close to no security at all.
Mutual Authentication
The server proves who it is to the device and the device proves who it is to the server. One-directional trust invites anything that can spoof a hostname.
Per-Device Credentials
No shared fleet key. One compromised unit is a revocation, not a firmware campaign across the installed base.
Independent Firmware Trust
Update images are signed and verified on the device separately from the transport, so compromising the link does not mean controlling what the device runs.
Least Privilege on Topics
A device may publish its own telemetry and subscribe to its own commands. Nothing more. Fleet-wide subscribe permissions are how one device becomes a window into all of them.
Time and Replay Protection
Messages carry sequence and timestamp so a captured payload cannot be replayed later to fake a reading or repeat a command.
FAQ
Connectivity questions
that decide the product.
Start with who controls the site. If the device sits on a customer network, Wi-Fi means depending on their password policy and their firewall, which turns every install into a negotiation. Cellular carries its own connectivity and works anywhere with coverage, at a running data cost. LoRa suits many low-rate devices across one site where you can own a gateway. Payload size, report interval, mobility, and power budget then narrow it further.
Related
Go deeper.
Settle connectivity
before you settle the board.
Bring the payload, the report interval, the environment, and the power budget. We will work the bearer, the protocol, and the running cost with you.