Follow spaceflight for long enough and you notice that the vehicle alone never tells you much. What matters just as much is the chain that carries its readings home: antennas, relays, ground stations, archives and the screens in the control room. When that chain degrades, operators lose sight of a perfectly healthy vehicle. Software teams face the identical problem, and the practice of looking after that chain goes by the name ObservabilityOps. This comparison sets three common ways of running a telemetry chain side by side, using orbital operations as the yardstick, so you can recognise which one your own team resembles.
A quick definition first. As the Ops Atlas entry on ObservabilityOps explains, the label is not an established standard term. The Atlas gives it a narrow reading: operating the observability system itself, as opposed to using observability data to run a service. Think of it as the people who keep the downlink healthy, not the flight controllers who read what arrives on it.
Why the orbit you choose changes the telemetry problem
Different orbits impose different habits on a ground team. A craft parked in a high fixed position, the arrangement covered in this plain walkthrough of geostationary orbit, stays in view of the same dish all day, so one antenna and one steady stream will do. A low-flying satellite, by contrast, zips past each station in minutes, which is why the explainer on low Earth orbit for beginners stresses short contact windows and many handovers. The telemetry chain for the second case needs buffering, retries and a clear idea of what must be sent first.
Software estates behave the same way. A single monolith talking to one dashboard is the geostationary case. A swarm of short-lived containers emitting logs, metrics and traces is the low-orbit case: the data is bursty, the contact windows are brief, and whoever runs the chain must decide what to keep and what to drop before the window closes.
Setup one: the improvised ground station
Picture a university team with a single rooftop dish and a spreadsheet of pass times. Whoever happens to be in the lab tunes the receiver, saves files into a shared folder and eyeballs the signal. It works while the team is small and the satellite is simple.
The Atlas describes the software equivalent as the lowest rung of its maturity ladder, labelled ad hoc. Every team installs its own agents, picks its own label names and builds its own dashboards. Nobody owns the pipeline, so nobody notices when it quietly stops delivering. The failure mode is not dramatic; it is silence. A dashboard that is flat could mean calm or could mean the feed died.
Strengths: zero coordination cost, fast to start, and each team gets exactly the view it wants. Weaknesses: duplicated effort, no shared vocabulary, and no one watching the watchers. Costs creep upward because nobody can see the total.
Setup two: the single mission-control room
The next step up is a dedicated control room with a named owner and a heartbeat. Every few seconds a known test signal is sent through the whole path, and if it fails to arrive, someone is paged. The Atlas calls the corresponding level centralised, and it ties it to two concrete things: an accountable owner, and a watchdog-style alert that fires when the alerting path itself goes quiet.
This is a genuine improvement, and for a handful of teams it may be all you need. The trap is scope creep. When one room is made responsible for everybody’s alerts and everybody’s dashboards, it becomes a bottleneck and a convenient place to dump blame. The Atlas leans on a point from DORA research here: monitoring should not be confined to one individual or team. Mission control can run the downlink, but flight dynamics, power and thermal specialists still need to own their own alarm limits.
Strengths: clear accountability, a tested alert path, quicker fixes when the chain breaks. Weaknesses: the owning room can turn into a ticket queue, and its expertise is a single point of dependence on one handful of people.
Setup three: the standardised network
The third arrangement resembles an agency-wide ground network. Antennas differ, but every station speaks the same frame format, uses the same naming for channels and enforces the same limits on what a mission may transmit. In the Atlas’s ladder this is the standardised level, where a common instrumentation standard (OpenTelemetry is the example it leans on), shared templates and hard limits replace one-off choices.
The limits matter more than they sound. One of the Atlas’s recurring warnings concerns label cardinality: every distinct combination of label values creates a new time series, so putting something unbounded such as a user identifier into a metric label can multiply storage without anyone deciding it should. A ground network would never let a mission add ten thousand telemetry channels overnight without review; a standardised observability setup applies the same discipline to labels.
Strengths: teams move faster because the paved road exists, costs become predictable, and onboarding a new service takes hours rather than weeks. Weaknesses: the standards need maintaining, and a rigid template can annoy a team whose spacecraft genuinely is unusual.
A side-by-side view
- Ownership: improvised stations have none; the control room has one named group; the network spreads ownership across a standards function and the producing teams.
- Watching the watchers: only the second and third setups run a continuous heartbeat that proves the path is alive.
- Cost visibility: the first hides it, the second sees totals, the third attributes spend to each team.
- Change speed: the first is fast but chaotic, the second is slowed by a queue, the third is fast on the paved road and slower off it.
- Typical size: one or two teams, a few teams, and many teams respectively.
The Atlas gives a rough decision rule that mirrors this table. With one or two teams, it suggests a named on-call owner only if you host the stateful back ends yourself or work in a regulated setting. Once three or more teams are fighting cost and noise problems, it points towards a platform team if you run those back ends, or a small standards function if you buy them as a service.
Sampling: choosing what to downlink
Every satellite operator knows the downlink budget is finite, so low-value readings are summarised on board and only the interesting ones are sent in full. Distributed tracing has the same dilemma. The Atlas notes that OpenTelemetry guidance begins suggesting sampling at around a thousand traces per second. Sampling at the start of a request is cheap but cannot promise that failing requests are kept, whereas deciding after the whole trace has arrived keeps the errors but is stateful and described as difficult to operate.
The control-room analogy fits neatly. Deciding up front resembles dropping low-priority channels on board; deciding afterwards resembles recording everything to a buffer and discarding the dull stretches once you know what happened. The second costs more buffer and more machinery, so the Atlas suggests introducing it for the largest producer first and not everywhere at once.
Which setup should you pick?
Use the staging of a flight as a guide. Before launch, you can tolerate improvisation; the stakes are small. During ascent and early operations, you want the single control room with a clear owner, because things change quickly and one accountable group can react. During long cruise, with many missions flying, a standard network pays for itself. A useful companion read is the overview of how a spacecraft moves through its flight phases, because the same logic of matching the support model to the phase applies to telemetry as much as to propulsion.
A practical way to decide is to ask three blunt questions:
- If the pipeline went silent tonight, how long before a human noticed? If the honest answer is “whenever someone checks the graph,” you are in setup one.
- Can you name the one person or group who is accountable for the chain, and does a test signal prove it still works end to end?
- Could a new team start sending data this week using templates and limits you already publish?
If you can answer yes to the third, you are close to the standardised network. If you can only answer yes to the second, you are in the control-room stage, which is a respectable place to be.
Common mistakes when moving between setups
Teams often jump straight from improvised to standardised by writing a long policy document. Nothing changes because nothing enforces it. The Atlas pairs standards with gateway limits, scrape limits and redaction rules so that the rules live in the pipeline instead of a wiki page. Another mistake is assuming self-hosting is free; the stateful back ends need patching, capacity planning and an on-call rota, which is just the dish-maintenance bill under a different name.
A third mistake is treating the label as a rebrand. The Atlas is explicit that this is neither the same thing as site reliability engineering nor a renamed monitoring team. Reliability targets and error budgets sit elsewhere. The job here is narrower and more mechanical: keep the pipe honest, bounded and affordable.
Bottom line
Compared on the three dimensions that matter, ownership, proof that the path is alive, and cost visibility, the improvised ground station loses, the control room is a solid starting point, and the standardised network is where growing organisations should aim. None of this needs new vocabulary beyond what the Atlas already lays out, and none of it needs to be adopted in one go. Pick the next rung, add one heartbeat, name one owner, cap one runaway label, and you will have moved your chain further up the ladder than most teams manage in a quarter.