Network Time Protocol

The Network Time Protocol (NTP) is a networking protocol for clock synchronization between computer systems over packet-switched, variable- latency data networks. In operation since before 1985, NTP is one of the oldest Internet protocols in current use. NTP was designed by David L. Mills of the University of Delaware.

NTP is intended to synchronize all participating computers to within a few milliseconds of Coordinated Universal Time (UTC). It uses the intersection algorithm, a modified version of Marzullo's algorithm, to select accurate time servers and is designed to mitigate the effects of variable network latency. NTP can usually maintain time to within tens of milliseconds over the public Internet, and can achieve better than one millisecond accuracy in local area networks under ideal conditions. Asymmetric routes and network congestion can cause errors of 100 ms or more.

The protocol is usually described in terms of a client–server model, but can as easily be used in peer-to-peer relationships where both peers consider the other to be a potential time source. Implementations send and receive timestamps using the User Datagram Protocol (UDP) on port number 123. They can also use broadcasting or multicasting, where clients passively listen to time updates after an initial round-trip calibrating exchange. NTP supplies a warning of any impending leap second adjustment, but no information about local time zones or daylight saving time is transmitted.

The current protocol is version 4 (NTPv4), which is backward compatible with version 3.

Network Time Security (NTS) is a secure version of NTP with TLS and AEAD.

History

In 1979, network time synchronization technology was used in what was possibly the first public demonstration of Internet services running over a trans-Atlantic satellite network, at the National Computer Conference in New York. The technology was later described in the 1981 Internet Engineering Note (IEN) 173 and a public protocol was developed from it that was documented in . The technology was first deployed in a local area network as part of the Hello routing protocol and implemented in the Fuzzball router, an experimental operating system used in network prototyping, where it ran for many years.

Other related network tools were available both then and now. They include the Daytime and Time protocols for recording the time of events, as well as the ICMP Timestamp messages and IP Timestamp option (). More complete synchronization systems, although lacking NTP's data analysis and clock disciplining algorithms, include the Unix daemon ''timed'', which uses an election algorithm to appoint a server for all the clients; and the Digital Time Synchronization Service (DTSS), which uses a hierarchy of servers similar to the NTP stratum model.

In 1985, NTP version 0 (NTPv0) was implemented in both Fuzzball and Unix, and the NTP packet header and round-trip delay and offset calculations, which have persisted into NTPv4, were documented in . Despite the relatively slow computers and networks available at the time, accuracy of better than 100 milliseconds was usually obtained on Atlantic spanning links, with accuracy of tens of milliseconds on Ethernet networks.

In 1988, a much more complete specification of the NTPv1 protocol, with associated algorithms, was published in . It drew on the experimental results and clock filter algorithm documented in and was the first version to describe the client–server and peer-to-peer modes. In 1991, the NTPv1 architecture, protocol and algorithms were brought to the attention of a wider engineering community with the publication of an article by David L. Mills in the IEEE Transactions on Communications.

In 1989, was published defining NTPv2 by means of a state machine, with pseudocode to describe its operation. It introduced a management protocol and cryptographic authentication scheme which have both survived into NTPv4, along with the bulk of the algorithm. However the design of NTPv2 was criticized for lacking formal correctness by the DTSS community, and the clock selection procedure was modified to incorporate Marzullo's algorithm for NTPv3 onwards.

In 1992, defined NTPv3. The RFC included an analysis of all sources of error, from the reference clock down to the final client, which enabled the calculation of a metric that helps choose the best server where several candidates appear to disagree. Broadcast mode was introduced.

In subsequent years, as new features were added and algorithm improvements were made, it became apparent that a new protocol version was required. In 2010, was published containing a proposed specification for NTPv4. The protocol has significantly progressed since then, and , an updated RFC has yet to be published. Following the retirement of Mills from the University of Delaware, the reference implementation is currently maintained as an open source project led by Harlan Stenn.

Clock strata

NTP uses a hierarchical, semi-layered system of time sources. Each level of this hierarchy is termed a ''stratum'' and is assigned a number starting with zero for the reference clock at the top. A server synchronized to a stratum ''n'' server runs at stratum ''n'' + 1. The number represents the distance from the reference clock and is used to prevent cyclical dependencies in the hierarchy. Stratum is not always an indication of quality or reliability; it is common to find stratum 3 time sources that are higher quality than other stratum 2 time sources. A brief description of strata 0, 1, 2 and 3 is provided below.

; Stratum 0
: These are high-precision timekeeping devices such as atomic clocks, GNSS (including GPS) or other radio clocks. They generate a very accurate pulse per second signal that triggers an interrupt and timestamp on a connected computer. Stratum 0 devices are also known as reference clocks. NTP servers cannot advertise themselves as stratum 0. A stratum field set to 0 in NTP packet indicates an unspecified stratum.RFC 5905, p. 21
; Stratum 1
: These are computers whose system time is synchronized to within a few microseconds of their attached stratum 0 devices. Stratum 1 servers may peer with other stratum 1 servers for sanity check and backup. They are also referred to as primary time servers.
; Stratum 2
: These are computers that are synchronized over a network to stratum 1 servers. Often a stratum 2 computer queries several stratum 1 servers. Stratum 2 computers may also peer with other stratum 2 computers to provide more stable and robust time for all devices in the peer group.
; Stratum 3
: These are computers that are synchronized to stratum 2 servers. They employ the same algorithms for peering and data sampling as stratum 2, and can themselves act as servers for stratum 4 computers, and so on.

The upper limit for stratum is 15; stratum 16 is used to indicate that a device is unsynchronized. The NTP algorithms on each computer interact to construct a Bellman-Ford shortest-path spanning tree, to minimize the accumulated round-trip delay to the stratum 1 servers for all the clients.

In addition to stratum, the protocol is able to identify the synchronization source for each server in terms of a reference identifier (refid).

Timestamps

The 64-bit binary fixed-point timestamps used by NTP consist of a 32-bit part for seconds and a 32-bit part for fractional second, giving a time scale that rolls over every 2³² seconds (136 years) and a theoretical resolution of 2⁻³² seconds (233 picoseconds). NTP uses an epoch of January 1, 1900. Therefore, the first rollover occurs on February 7, 2036.

NTPv4 introduces a 128-bit date format: 64 bits for the second and 64 bits for the fractional-second. The most-significant 32-bits of this format is the ''Era Number'' which resolves rollover ambiguity in most cases. According to Mills, "The 64-bit value for the fraction is enough to resolve the amount of time it takes a photon to pass an electron at the speed of light. The 64-bit second value is enough to provide unambiguous time representation until the universe goes dim."University of Delaware Digital Systems Seminar presentation by David Mills, 2006-04-26

Clock synchronization algorithm

A typical NTP client regularly polls one or more NTP servers. The client must compute its time offset and round-trip delay. Time offset ''θ'' is positive or negative (client time > server time) difference in absolute time between the two clocks. It is defined by

$\theta = \frac ,$
and the round-trip delay ''δ'' by
$\delta = ,$
where
*''t''₀ is the client's timestamp of the request packet transmission,
*''t''₁ is the server's timestamp of the request packet reception,
*''t''₂ is the server's timestamp of the response packet transmission and
*''t''₃ is the client's timestamp of the response packet reception.

To derive the expression for the offset, note that for the request packet,
$t_0 + \theta + \delta/2 = t_1$
and for the response packet,
$t_3 + \theta - \delta/2 = t_2$
Solving for ''θ'' yields the definition of the time offset.

The values for ''θ'' and ''δ'' are passed through filters and subjected to statistical analysis. Outliers are discarded and an estimate of time offset is derived from the best three remaining candidates. The clock frequency is then adjusted to reduce the offset gradually, creating a feedback loop.

Accurate synchronization is achieved when both the incoming and outgoing routes between the client and the server have symmetrical nominal delay. If the routes do not have a common nominal delay, a systematic bias exists of half the difference between the forward and backward travel times.

Software implementations

Reference implementation

The NTP reference implementation, along with th