Ways to get time into a datacenter
Time has to get into the datacenter from somewhere. There are more options than most people realize, and none of them are exotic.
The options
GNSS. An antenna with a view of the sky, a receiver, and a cable between them. The signals carry time traceable to UTC, and the equipment is commodity — this is the route most datacenters take, for reasons below.
That “receiver” is two quite different purchases, and it is worth deciding which one you are making before reading any further:
- a GPS network clock — a 1RU box you rack, feed an antenna, and give an IP address, which hands you NTP and PTP. You are buying the clock.
- a GNSS receiver module or geodetic receiver, whose PPS and serial stream you wire into yourself, supplying the oscillator, the discipline loop and the protocols. You are building the clock.
Almost every datacenter wants the first. The rest of this page applies to both; which capabilities matter to which is where they part company.
A commercial time service over a dedicated link. Increasingly available and increasingly good. Quincy Data, for instance, announced a sub-nanosecond synchronization service in 2024, delivered over the same low-latency infrastructure the trading industry already buys. If somebody is already selling you a fiber, they may be able to sell you time down it.
Fiber from a national lab. Several NMIs will deliver a traceable timescale over dedicated fiber to institutions near enough to reach. Superb accuracy, and a short list of eligible addresses.
NTP over the internet. Free, universal, good to milliseconds. Worth naming rather than dismissing — for a great many applications it is genuinely sufficient, and organizations spend money escaping it before establishing that they needed to.
Every one of these arrives through a distribution protocol. Acquisition and distribution are the same technology pointed in different directions: what makes something acquisition is that the source is outside your control and authoritative, not that the mechanism differs.
Why GNSS usually wins below 100 ns
Not because it is the most accurate — it is not — but because of where it sits on the cost curve.
Below about 100 ns, and beyond what NTP can reach, GNSS occupies a sweet spot that is hard to argue with:
- The signal is free and available everywhere with a view of the sky, which is a sentence no other option can claim.
- The equipment is commodity. Timing receivers are a competitive market with well-understood products.
- Its chain to UTC is short and free — through the constellation operators, with no contract, no provider and nobody to depend on commercially. A short chain is not the same as traceability, which needs calibration and a document; this is the raw material for one. And how well each operator manages its end is measured and published monthly, which turns the choice of constellation into an evidence question. The short version: on predicting UTC, GPS, Galileo and BeiDou are effectively tied under a nanosecond of mean bias, and GLONASS is an order of magnitude worse.
- Its error budget is well documented, which matters more than it sounds: you can find out what your accuracy actually is, and much of this site is about how.
The alternatives beat it on accuracy or on independence, and cost considerably more, or are unavailable at your address, or both.
Most of what is written here concerns acquiring time from GNSS, because that is what most datacenters do and what I have most experience of. That is a focus, not a claim of superiority — the other methods are complementary, and as the next section argues, you probably want at least two.
Why you want more than one
GNSS is a single point of failure with a large blast radius. A failed antenna, a cut cable, a jammer in the parking lot, a receiver firmware bug or a constellation problem takes out everything downstream at once — and, as the silently wrong theme running through this site keeps demonstrating, timing failures are frequently quiet.
Two arguments for a second source. The second is much stronger than the first:
Availability. If one source fails, the other carries you. Standard resilience reasoning, and the argument people make first.
Comparison. Two sources let you check. A single source can be wrong with complete confidence and nothing in the system will disagree with it. Two sources that ought to agree, and do not, are telling you something no single source ever could.
Which is why the useful thing is not merely a spare but a mechanism for comparing sources against each other, and something that complains when they diverge. A backup nobody compares against inherits exactly the same silence as the primary — and what your second source shares with the first decides what it can catch, which is a page of its own.
Can you trust the time you acquired?
Accuracy and authenticity are different questions, and the second one is easier to forget. A spoofed source can be exquisitely precise and completely wrong, and nothing about the accuracy figures on this page detects that.
Both major routes now have a free answer.
On the network side: NTS. Network Time Security is NTP with cryptographic authentication — same protocol, same accuracy, with assurance the time came from who it claims and was not altered on the way. Plain NTP is trivially spoofable by anyone able to inject packets.
On the GNSS side: OSNMA. Galileo’s Open Service Navigation Message Authentication signs the navigation message, so a receiver can tell an authentic Galileo signal from a fabricated one. It is broadcast free alongside the ordinary open service; receiver support is uneven but growing, and it is worth asking about when buying.
Neither defends against everything. OSNMA authenticates the navigation message rather than the ranging signal, so it raises the cost of spoofing without eliminating it, and neither helps at all against plain jamming — which denies service rather than falsifying it. But both are free, and the alternative is trusting an unauthenticated broadcast.
This is also the strongest practical argument for the two-source setup above. Authentication tells you a source is genuine; a second source tells you it is right.
Where to go next
- Matching acquisition and distribution — how much of each you need, and why buying one without the other wastes money.
- How does GNSS holdover work? — the other reason to have a second source, and how long the first one’s absence actually costs you.
- What do I put in for my antenna’s position? — if you take the GNSS route, this is where the accuracy is won or lost.
- Datacenter GNSS time best practices — the eight-item checklist for once you have chosen. Most of them cost nothing but attention.
- Why do you want precise time? — worth answering before any of the above, because it decides how much of this you actually need.