protocols · 4 min read

NTP stratum levels and accuracy budgets

NTP stratum runs from 0 to 15: the further from the source, the larger the time deviation. How do you plan the stratum hierarchy for your network, and what accuracy budgets apply per layer?

Choose your reading level

Stratum is a rank number that indicates how far a clock is from the real time source. Stratum 0 is the source itself (an atomic clock or GPS), stratum 1 is a server aligned directly with it, stratum 2 aligns with stratum 1, and so on. The higher the number, the lower the accuracy.

For a network that means: place one or two stratum-1 servers as the source, have your important servers align with them, and the rest align with those. Let every device work with several time servers at once, so that the failure of one source is never a problem.

In NTP architectures you will hear terms like "stratum 1" or "stratum 3". That is a hierarchical ranking of clocks based on distance from the original time source, and it strongly determines how accurate each layer is.

What does each stratum level mean?

Stratum 0: the fundamental time source, an atomic clock in a national laboratory, a GPS satellite, a GNSS receiver. Not directly accessible over the network.
Stratum 1: an NTP server that synchronises directly with a stratum 0 source via hardware (GNSS antenna, IRIG cable, etc.). Very accurate, typically sub-100 μs.
Stratum 2: synchronises with one or more stratum 1 servers over the network. Accuracy: 1–10 ms on LAN, tens of ms over WAN.
Stratum 3: synchronises with stratum 2. Accuracy: typically 10–100 ms.
Stratum 4–15: derived even further. Rarely used for IT purposes in practice.
Stratum 16: unsynced (the server does not trust its own time).

What does that mean for your network?

In a typical office or data-centre deployment:

Place one or two stratum 1 servers (an NTP100-GNSS plus GMR redundant for example) as the source
Let key application servers (database, monitoring, SIEM) synchronise against that stratum 1, they become stratum 2
Workstations, IoT devices, clocks synchronise against the stratum 2 servers, they become stratum 3
Redundancy: every client can synchronise against 3+ NTP servers at once, so that a single failure is not a disaster

How much accuracy do you lose per stratum layer?

Every hop adds uncertainty. A rough rule of thumb:

Layer	Typical LAN accuracy
Stratum 1 (your own GNSS server)	10–100 μs
Stratum 2 (server in this data centre)	0.5–5 ms
Stratum 3 (client PC)	5–50 ms
Stratum 4+	depends

Enough for 99% of business applications, TLS certificate validation, audit-trail sorting, database replication.

When is it not enough?

High-frequency trading: requires sub-millisecond, so stratum 1 directly plus PTP
Broadcast frame sync: stratum 1 plus PTP or even SMPTE 2110
Digital substation IEC 61850: PTP grandmaster instead of an NTP hierarchy
Scientific timing: direct atomic clock or GPS disciplined Rubidium

In those cases you are talking about a fundamentally different architecture, PTP, not NTP. See our article NTP vs PTP.

In practice via Daylight

For most customers, 1 stratum 1 server plus redundancy is sufficient. Too many customers buy 3 stratum 1 servers because they "want more". Call Daylight and we will sketch the architecture out together, less hardware can work perfectly fine.

Sources

RFC 5905 — NTPv4 specification section 3 on stratum
NTP Pool Project, an example of a worldwide stratum 2/3 distribution

The stratum number in NTP is a topological distance measure, not a direct accuracy value, but in practice they correlate. Stratum 0 is the physical reference (atomic clock, GNSS, IRIG), not network-addressable. Stratum 1 synchronises with it directly via hardware and typically reaches sub-100 µs. Each further layer adds up to a maximum of stratum 15; stratum 16 marks an unsynchronised clock.

The accuracy budget stacks per layer: a client's error is the error of its source plus its own measurement uncertainty (path asymmetry, jitter, polling interval). That is why deeper in the hierarchy typically yields 1–10 ms on a LAN and more over a WAN. Design with redundancy: each client preferably configures three or more independent sources, so NTP's selection and clustering algorithm can vote out an outlier (falseticker).

For critical environments one combines two or more stratum-1 servers with independent GNSS antennas and, where needed, holdover oscillators, so that neither a source outage nor an antenna fault drags down the whole hierarchy.

Standard: IETF RFC 5905.