[Docs] [txt|pdf] [Tracker] [WG] [Email] [Nits]

Versions: 00 01

Network Time Protocol Working                          W. Kasch (Editor)
Group                                                            JHU/APL
Internet-Draft                                        J. Burbank, Editor
Expires: April, 2006                                             JHU/APL
                                                           October, 2005


      The Network Time Protocol Version 4 Algorithm Specification
                     <draft-ietf-ntp-ntpv4-algorithms-00>

Status of this Memo

   This document is an Internet-Draft and is subject to all provisions
   of Section 3 of RFC 3978.  By submitting this Internet-Draft, each
   author represents that any applicable patent or other IPR claims of
   which he or she is aware have been or will be disclosed, and any of
   which he or she becomes aware will be disclosed, in accordance with
   Section 6 of BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as
   Internet-Drafts.

   Internet-Drafts are draft documents valid for a maximum of six
   months and may be updated, replaced, or obsoleted by other
   documents at any time.  It is inappropriate to use Internet-Drafts
   as reference material or to cite them other than as "work in
   progress."

   The list of current Internet-Drafts can be accessed at
   http://www.ietf.org/ietf/1id-abstracts.txt.

   The list of Internet-Draft Shadow Directories can be accessed at
   http://www.ietf.org/shadow.html.

   This document is a submission of the IETF NTP WG.  Comments should
   be directed to the NTP WG mailing list, ntpwg@lists.ntp.isc.org.

   This Internet-Draft will expire on April, 2006.

Abstract

   The Network Time Protocol (NTP) is widely used to synchronize
   computer clocks in the Internet.  This memorandum describes
   the algorithms used by Version 4 of the NTP (NTPv4) to calculate
   time values.







Kasch, et al.               Expires April, 2006                 [Page 1]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


Table of Contents

   1.  Introduction . . . . . . . . . . . . . . . . . . . . . . . . .  3

   2.  Clock Filter Algorithm . . . . . . . . . . . . . . . . . . . .  5

   3.  Clock Selection Algorithm  . . . . . . . . . . . . . . . . . .  7

   4.  Clustering Algorithm . . . . . . . . . . . . . . . . . . . . .  8

   5.  Clock Combining Algorithm  . . . . . . . . . . . . . . . . . .  9

   6.  Polling Algorithm  . . . . . . . . . . . . . . . . . . . . . . 10

   7.  Clock Discipline Algorithm . . . . . . . . . . . . . . . . . . 10
       7.1 Poll Interval Control  . . . . . . . . . . . . . . . . . . 12
       7.2 State Machine  . . . . . . . . . . . . . . . . . . . . . . 13

   8. Security Considerations . . . . . . . . . . . . . . . . . . . . 15

   9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 15

   10. Acknowledgments  . . . . . . . . . . . . . . . . . . . . . . . 15

   11. References . . . . . . . . . . . . . . . . . . . . . . . . . . 15
      11.1   Normative References . . . . . . . . . . . . . . . . . . 15
      11.2   Informative References . . . . . . . . . . . . . . . . . 15

   12. Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . 16

       Intellectual Property and Copyright Statements . . . . . . . . 17





















Kasch, et al.                Expires April, 2006                [Page 2]


Internet-Draft          NTPv4 Algorithms Specification      October 2005


1.  Introduction

   The Network Time Protocol Version 3 (NTPv3) specified in RFC 1305
   [MIL92] has been widely used to synchronize computer clocks in the
   global Internet.  It provides comprehensive mechanisms to access
   national time and frequency dissemination services, organize the NTP
   subnet of servers and clients and adjust the system clock in each
   participant. In most places of the Internet of today, NTP provides
   accuracies of 1-50 ms, depending on the characteristics of the
   synchronization source and network paths.

   NTP is designed for use by clients and servers with a wide range of
   capabilities and over a wide range of network jitter and clock
   frequency wander characteristics.  Many users of NTP in the Internet
   of today use a software distribution available from www.ntp.org.  The
   distribution, which includes the full suite of NTP options,
   mitigation algorithms and security schemes, is a relatively complex,
   real-time application.  While the software has been ported to a wide
   variety of hardware platforms ranging from personal computers to
   supercomputers, its sheer size and complexity is not appropriate for
   many applications.  This facilitated the development of the Simple
   Network Time Protocol Version 4 (SNTPv4) as described in RFC 2030
   [MIL96].

   Since the standardization of NTPv3, there has been significant
   development which has led to Version 4 of the Network Time Protocol
   (NTPv4).  This document describes NTPv4, which introduces new
   functionality to NTPv3 as described in RFC 1305, and functionality
   expanded from that of SNTPv4 as described in RFC 2030 (SNTPv4 is a
   subset of NTPv4).

   When operating with current and previous versions of NTP and SNTP,
   NTPv4 requires no changes to the protocol or implementations now
   running or likely to be implemented specifically for future NTP or
   SNTP versions.  The NTP and SNTP packet formats are the same and the
   arithmetic operations to calculate the client time, clock offset and
   roundtrip delay are the same.  To a NTP or SNTP server, NTP and SNTP
   clients are indistinguishable; to a NTP or SNTP client, NTP and SNTP
   servers are indistinguishable.

   NTP usually operates simultaneously with multiple servers and may
   have multiple clients of its own. NTP employs several algorithms
   that together allow the calculation of time from messages that come
   from an NTP or SNTP server.  The overall organization of the
   algorithms is illustrated in Figure 1. For every server there are two
   processes, a peer process which receives and processes each packet,
   and a companion poll process which sends packets to the server at
   programmed intervals. State variables and data measurements
   are maintained separately for each pair of processes in a block of
   memory called the peer variables. The peer and poll processes


Kasch, et al.               Expires April, 2006                 [Page 3]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


..........................................................
. Remote   ..   Peer/Poll   ..              System       .
. Servers  ..   Processes   ..              Process      .
.          ..               ..                           .
.----------..-------------..--------------               .
.|        |->|           |..|            |               .
.|Server 1|..|Peer/Poll 1|->|            |               .
.|        |<-|           |..|            |               .............
.----------..-------------..|            |               .   Clock   .
                                                         .Discipline .
.          ..       ^     ..|            |               .. Process  .
.          ..       |     ..|            |               ..          .
.----------..-------------..|            |  |-----------|..          .
.|        |->|           |..| Selection  |->|            ..--------  .
.|Server 2|..|Peer/Poll 2|->|    and     |  | Combining |->| Loop |->.--
.|        |<-|           |..| Clustering |  | Algorithm |..|Filter|  . |
.----------..-------------..| Algorithms |->|           |.------------ |
.          ..       ^     ..|            |  |-----------|.             |
.          ..       |     ..|            |               .             |
.----------..-------------..|            |               .             |
.|        |->|           |..|            |               .             |
.|Server 3|..|Peer/Poll 3|->|            |               .             |
.|        |<-|           |..|            |               .             |
.----------..-------------..|------------|               .             |
....................^.....................................             |
                    |                                                  |
                    |                                                  |
                    |                                ...............   |
                    |                                .   /-----\   .   |
                    '----------------------------------<-| VFO |-<-.---+
                                                     .   \-----/   .
                                                     . Clock Adjust.
                                                     .   Process   .
                                                     ...............

                Figure 1. NTPv4 Algorithm Interactions


   together with their variables collectively belong to an association.
   Associations can be either temporary or permanent. Permanent
   associations are described as persistent, while temporary
   associations are referred to as preemptable or ephemeral.

   As each NTP packet arrives, the server time is compared to the system
   clock and an offset specific to that server is determined. The system
   process refines these offsets using the selection, clustering and
   combining algorithms and delivers a correction to the clock
   discipline process, which functions as a lowpass filter to smooth the
   data and close the feedback loop. The clock adjust process runs at
   one-second intervals to amortize the corrections in small adjustments


Kasch, et al.               Expires April, 2006                 [Page 4]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   that approximate a continuous, monotonic clock. The output of the
   combining algorithm represents the best estimate of the system clock
   offset relative to the server ensemble.  The discipline algorithm
   adjusts the frequency of the variable frequency oscillator (VFO) to
   minimize this offset. Finally, the timestamps of each server are
   compared to timestamps derived from the VFO in order to calculate the
   server offsets and close the feedback loop.

   This document is organized as follows.  Section 2 describes the clock
   filter algorithm.  Section 3 describes the clock selection algorithm.
   Section 4 describes the clustering algorithm.  Section 5 describes
   the clock combining algorithm.  Section 6 describes the polling
   algorithm.  Section 7 describes the clock discipline algorithm.

   NTPv4 is hereafter referred to simply as NTP, unless explicitly
   noted.


2.  Clock Filter Algorithm

   The NTP clock filter algorithm selects the most appropriate sample
   data while rejecting noise spikes due to packet collisions and
   network congestion.  The clock offset (?) and roundtrip delay (d)
   samples are computed from the four most recent timestamps.  Without
   making any assumptions about the delay distributions, but assuming
   the frequency difference or skew between the server and peer clocks
   can be neglected, let (?,d) represent the offset and delay when the
   path is otherwise idle; thus (?,d) represents the true offset and
   delay values.  The clock filter algorithm essentially acts as an
   accurate estimator and produces an estimate of the time, known as
   (?hat,dhat), from a sample sequence (?i,di), where i denotes a
   particular sample at some time, collected for the path over an
   appropriate interval under ambient traffic conditions.

   The design of the clock filter algorithm was suggested by the
   observation that packet switching networks are most often operated
   well below the knee of the throughput-delay curve, which means that
   packet queues are mostly small with relatively infrequent bursts. In
   addition, the routing algorithm most often operates to minimize the
   number of packet-switch hops and thus the number of queues. Not only
   is the probability that an NTP packet finds a busy queue in one
   direction relatively low, but the probability of packets from a
   single exchange finding busy queuesin both directions is even lower.
   Therefore, the best offset samples should occur with the lowest
   delays.

   Upon arrival of an NTP packet resulting from some poll interval at
   time t=0, a shift register containing four variables (?i,di,ei,ti) is
   populated with the 0th sample, (?0,d0,e0,t0).  Here, e is the error
   (in seconds), which is initially set to precision and grown at a rate
   r=15 ppm for each epoch.  If a packet has not arrived for three

Kasch, et al.               Expires April, 2006                 [Page 5]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   successive poll intervals, then the sample (0,0,16,t) is shifted into
   the register, where t is the last current known time.  Missing data
   samples that force this condition are never used in subsequent filter
   calculations, but do prevent very old (i.e. stale) samples from being
   used.

   Next, the register contents are copied to a temporary list and sorted
   by the metric ? designed to avoid missing data and devalued samples
   older than the compromise Allan intercept ssuby(x) = 1500 s. The
   Allan intercept is the intersection coordinate (x, y) of the phase
   and frequency lines.  It characterizes each particular timing source
   and clock oscillator. A useful statistic is the x value, which
   specifies the optimum time constant for the particular source and
   oscillator combination.  The x value ranges from about 500 s to 2000
   s. Above this value the performance is limited by oscillator wander,
   while below this value the performance is limited by system jitter.
   For comparison, the NTPv4 clock discipline time constant is about
   1000 s at a poll interval of 64 s. The y statistic represents the
   best stability that can be achieved for the particular source and
   oscillator, but is not useful for performance optimization. For this
   reason, the term Allan intercept applies to the x value at the
   intercept point.

   If ej = infinity, then ?j = infinity; else, if tj-t > ssuby(x) then
   ?j=Ksubd + ej; else, ?j=dj, where Ksubd=1 s is the selection
   threshold.  The algorithm essentially sorts the data by exchanging
   sets; however, an exchange is not made unless to do so would reduce
   the metric by at least the value of the precision. In other words, it
   does not make sense to change the order in the list, which might
   result in the loss of otherwise good samples, unless the metric
   change is significant. The first entry (?0,d0,e0,t0) on the temporary
   list represents the lowest delay sample, which is used to update the
   peer offset ?=?0 and peer delay d=d0.  The peer dispersion e is
   calculated from the temporary list:

   e=sum from k=0 to k=n-1 of [esubk/2^(k+1)].

   Finally, the temporary list is trimmed by discarding all entries
   where ?j= infinity and all but the first devalued entry ?j >= Ksubd,
   if one is present, leaving m (0<=m<n) surviving entries on the list.
   The peer jitter psi is used by the clustering algorithm as a quality
   metric and in the computation of the expected error:

   psi=[(1/(m-1))*sum from k=1 to k=m-1 of [(?subk-?sub0)^2)]]^(1/2).

   A 'popcorn spike' is a transient outlier, usually only a single
   sample, that is typical of congested Internet paths. The popcorn
   spike suppressor is designed to detect and remove them. Let ?' be the
   peer offset determined by the previous message and psi the current
   peer jitter. If |?-?'|>Ksubs*psi, where Ksubs is a tuning parameter
   that defaults to 3, the sample is a popcorn spike and is discarded.

Kasch, et al.               Expires April, 2006                 [Page 6]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   Note that the peer jitter will increase to protect a legitimate step
   change.

   As demonstrated by simulation and practical experience, it is prudent
   to avoid using samples more than once. Let tp be the epoch the peer
   variables were last updated and t0 the epoch of the first sample on
   the temporary list. If t0<=tp, the new sample is a duplicate or
   earlier than the last one used. If this is true, the algorithm exits
   without updating the system clock; otherwise, tp=t0 and the offset
   can be used to update the system clock. The components of the tuple
   (?, d, e, psi, tp) are called the peer variables.


3.  Clock Selection Algorithm

   In order to provide reliable synchronization, NTP uses multiple
   redundant servers and multiple disjoint network paths whenever
   possible.  When a number of associations are established, it is not
   clear beforehand which are truechimers and which are falsetickers.
   A 'truechimer' is a clock that maintains timekeeping accuracy to a
   previously published (and trusted) standard, while a 'falseticker'
   is a clock that do not maintain that level of timekeeping accuracy.
   Crucial to the success of this approach is a robust algorithm which
   finds and discards the falsetickers from the raw server population,
   since the timekeeping accuracy of a particular server may not be
   known a priori.  The clock selection algorithm determines from among
   all associations a suitable subset of truechimers capable of
   providing the most accurate and trustworthy time using principles
   similar to [DOL95].

   The true offset ? of a correctly operating clock relative to UTC must
   be contained in a computable range, called the confidence interval,
   equal to the root distance defined below. Marzullo and Owicki [BER00]
   devised an algorithm designed to find the intersection
   interval containing the correct time given the confidence intervals
   of m clocks, of which no more than f are considered incorrect. The
   algorithm finds the smallest intersection interval containing points
   in at least (m-f) of the given confidence intervals.

   The clock selection algorithm operates as follows:

   1.  For each of m associations, construct a correctness interval
       [? - rootdist(), ? + rootdist()].

   2.  Select the lowpoint, midpoint and highpoint of these intervals.
       Sort these values in a list from lowest to highest. Set the
       number of falsetickers f = 0.





Kasch, et al.               Expires April, 2006                 [Page 7]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   3.  Set the number of midpoints d = 0. Set c = 0. Scan from lowest
       endpoint to highest. Add one to c for every lowpoint, subtract
       one for every highpoint, add one to d for every midpoint.
       If c = m - f, stop; set l = current lowpoint.

   4.  Set c = 0. Scan from highest endpoint to lowest. Add one to c for
       every highpoint, subtract one for every lowpoint, add one to d
       for every midpoint. If c = m - f, stop; set u = current
       highpoint.

   5.  Is d = f and l < u?

   if yes, then follow step 5y, else, follow step 5n.

   5y.  Success:  the intersection interval is [l,u].

   5n.  Add one to f. Is f < m / 2?  If yes, then go to step 3 again.
        If no, then go to step 6.

   6.  Failure; a majority clique could not be found.  Stop algorithm.


4.  Clustering Algorithm

   NTP configurations usually include several servers in order to
   provide sufficient redundancy for the selection algorithm to
   determine which are truechimers and which are not. When a sizeable
   number of servers are present, the individual clock offsets for each
   are not always the same, even if each server is closely synchronized
   to UTC by one means or another. Small systematic differences in the
   order of a millisecond or two are usually due to interface and
   network latencies. Larger differences are due to asymmetric delays
   and in the extreme due to asymmetric satellite/landline delays.

   The clustering algorithm sifts the truechimers of the selection
   algorithm to identify the survivors providing the best accuracy. In
   principle, the sift could result in a single survivor and its offset
   estimate used to discipline the system clock; however, a better
   estimate usually results if the offsets of a number of survivors are
   averaged together. So, a balance must be struck between reducing the
   selection jitter by casting off outlyers and improving the offset
   estimate by including more survivors in the average.

   The clustering algorithm steps follow:

   1.  Let (?, ?, ?) represent a candidate peer with offset ?, jitter j
       and a weight factor ? = stratum * MAXDIST + rootdist().

   2.  Sort the candidates by increasing ?. Let n be the number of
       candidates and NMIN the minimum number of survivors.


Kasch, et al.               Expires April, 2006                 [Page 8]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   3.  For each candidate compute the selection jitter jsubS (RMS peer
       offset differences between this and all other candidates).

   4.  Select jsubmax as the candidate with maximum jsubS.

   5.  Select jsubmin as the candidate with minimum jsubS.

   6.  Does the condition (jsubmax < jsubmin OR n = NMIN) hold true?

   If yes, go to step 6y.  If no, go to step 6n.

   6y.  Done.  The remaining cluster survivors are correct. The
        survivors are in the v. structure sorted by ?.

   6n.  Delete the outlyer candidate with jsubmax; reduce n by one, and
        go back to step 3.


5.  Clock Combining Algorithm

   The selection and clustering algorithms operate to select a single
   system peer based on stratum and root distance. The result is that
   the NTP subnet forms a logical tree with the primary servers at
   the root and other servers at increasing stratum levels toward the
   leaves. However, since each server on the tree ordinarily runs the
   NTP protocol with several other servers at equal or lower stratum,
   these servers can provide diversity paths for backup and cross
   checking. While these other paths are not ordinarily used directly
   for synchronization, it is possible that increased accuracy can be
   obtained by averaging their offsets according to appropriately chosen
   weights.

   The result of the clustering algorithm is a set of survivors (there
   must be at least one) that represent truechimers, or correct clocks.
   If only one peer survives or if the prefer peer is among the
   survivors, that peer becomes the system peer and the combining
   algorithm is not used. Otherwise, the final clock correction is
   determined by the combining algorithm.

   Let the tuples (?subi,psisubi,?subi)represent the peer offset, peer
   jitter, and root distance for the ith survivor. Then the combined
   peer offset and peer jitter is, respectively:

   T=a*sum over all i of [?subi/?subi] and psisubr=a*sum over all i of
   [(psisubi)^2/?subi]^(1/2),

   where a is a normalization constant:

   a=1/[sum over all i of [1/?subi].

   The result T is the system offset processed by the clock discipline

Kasch, et al.               Expires April, 2006                 [Page 9]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   algorithm. Note that the root distance cannot be less than
   the precision in order to avoid divide exceptions.

   Let psisubs represent the selection jitter associated with the system
   peer and psisubr as above. Then the system jitter is defined as:

   sj=[(psisubr)^2+(psisubs)^2]^(1/2).

   The system jitter represents the best estimate of error in computing
   the clock offset. It is interpreted as the expected error statistic
   available to application program.


6.  Polling Algorithm

   The poll process determines whether and when to send a poll message
   to the server. Ordinarily, polls are sent at regular intervals
   determined by the clock discipline time constant. In some cases
   where justified by network load, performance can be improved and
   network jitter reduced by sending several messages instead of just
   one. This can be done when the server is unreachable, when it is
   reachable or both. The most common cases where this is advisable is
   when using very large poll intervals in the order of several hours or
   more.

   The poll interval starts out normally at about one minute. If the
   offset is less than a tuning constant times the system jitter for
   some number of polls, it is increased, but usually not above 1024 s.
   Otherwise, it is decreased, but usually not below 64 s. The limits
   can be changed to a lower limit of 16 s and/or to an upper limit of
   36 h. In order to minimize network traffic, when a server has not
   been heard for some time, the poll interval is increased in stages to
   1024 s.


7.  Clock Discipline Algorithm

   The clock discipline algorithm synchronizes the computer clock with
   respect to the best time value from each server and the best
   combination of servers.  This algorithm automatically adapts to
   changes in operating environment without manual configuration or
   real-time management functions.  The clock discipline algorithm is
   implemented as the feedback control system shown in Figure 2.









Kasch, et al.               Expires April, 2006                [Page 10]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


                     ---------
              ?r +  |         \        +----------------+
      NTP --------->|  Phase   \   Vd  |                |  Vs
              ?c -  | Detector  ------>|  Clock Filter  |-----+
          +-------->|          /       |                |     |
          |         |         /        +----------------+     |
          |          ---------                                |
          |                                                   |
        -----                                                 |
       /     \                                                |
       | VFO |                                                |
       \     /                                                |
        -----    +-------------------------------------+      |
          ^      |            Loop Filter              |      |
          |      |                                     |      |
          |      | +---------+   x  +-------------+    |      |
          |      | |         |<-----|             |    |      |
          +------|-|  Clock  |   y  | Phase/Freq  |<---|------+
                 | | Adjust  |<-----| Prediction  |    |
                 | |         |      |             |    |
                 | +---------+      +-------------+    |
                 |                                     |
                 +-------------------------------------+

             Figure 2. Clock Discipline Algorithm

   The variable ?subr represents the combined server reference phase and
   ?subc the control phase of the VFO. Each update received from a
   server produces a signal Vsubd representing the instantaneous phase
   difference ?subr - ?subc. The clock filter for each server functions
   as a tapped delay line, with the output taken at the tap selected by
   the clock filter algorithm. The selection, clustering and combining
   algorithms combine the data from multiple filters to produce the
   signal Vsubs. The loop filter, with impulse response F(t), produces
   the signal Vsubc which controls the VFO frequency ?subc.  Thus, its
   phase ?subc follows:

   ?subc = integral over t of (?subc(t)dt)

   which closes the loop.  The Vsubc signal is generated by an
   adjustment process which runs at intervals of one second in the NTP
   daemon or one tick in the kernel.

   The NTPv4 discipline includes both PLL and FLL capabilities. The
   selection of which mode to use, FLL or PLL and in what combination,
   is made on the basis of the poll exponent tau. In the NTPv4 design,
   PLL mode is used at smaller values of tau, while FLL mode is used at
   larger values. In between, a combination of PLL and FLL modes is
   used. This improves the clock accuracy and stability, especially for
   poll intervals larger than the Allan intercept.


Kasch, et al.               Expires April, 2006                [Page 11]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


                  x <------(Phase Correction)<--.
                                                |
                        ysubfll                 |
                         .-(FLL Predict)<-------+<--Vsubs
                         |                      |
                        \|/                     |
                  y <--(Sum)                    |
                         ^                      |
                         |                      |
                         '-(PLL Predict)<-------'
                       ysubpll

                  Figure 3. FLL/PLL Prediction Functions

   In PLL mode y is a time integral over all past values of Vs, so the
   PLL frequency adjustment required is:

   ysubPLL = Vsubs*mu/(64Tsubc)^2.

   where Tsubc is the time constant.  In FLL mode, is an average of past
   frequency changes, as computed from Vsubs and mu. The goal of the
   algorithm is to reduce Vsubs to zero; so, to the extent this
   has been successful in the past, previous values can be assumed zero
   and the average becomes:

   ysubFLL = (Vsubs-x)/(8mu).

   where x is the residual phase error computed by the clock adjust
   process.

   Finally, in both PLL and FLL modes set the phase to x=Vsubs and
   frequency y=y+ysubPLL+ysubFLL.

   Once each second the adjustment process computes a phase increment
   z=x/(16*Tsubc) and new phase adjustment x=x-z.  The phase increment z
   is passed to the kernel time adjustment function.  This continues
   until the next update which recomputes x and y.

   A key factor in the performance of the PLL/FLL hybrid algorithm are
   the weight factors for the ysubPLL and ysubFLL adjustments, which
   depend on the poll exponent tau which in turn determines the time
   constant Tsubc = 2^(tau), in seconds.  PLL contributions should
   dominate at the lower values of tau, while FLL contributions should
   dominate at the higher values.  The clock discipline algorithm
   response times to several PPM deviation examples is presented in
   [MIL05].

7.1  Poll Interval Control

   The NTPv4 algorithm aims to set the averaging time somewhere near the
   Allan intercept. A key to this strategy is the measured clock jitter

Kasch, et al.               Expires April, 2006                [Page 12]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   and oscillator wander statistics. The clock jitter is estimated from
   phase differences psisubc=<?x^2>^(1/2), where the brackets indicate
   an exponential average.  The oscillator wander is estimated from
   frequency differences psisubf = Tsubc*<?y^2>^(1/2).  As the poll
   exponent tau increases, it is expected that psisubc will decrease and
   psisubf will increase, depending on the relative contributions of
   phase noise and frequency noise.

   In the NTPv4 algorithm at each update a counter is incremented by one
   if x is within the bound |x|< 4psisubc, where the constant 4 is
   determined by experiment, and decremented by one otherwise.

   In order to avoid needless hunting, a degree of hysteresis is built
   into the scheme. If the counter reaches an upper limit 30, tau is
   increased by one; if it reaches a lower limit -30, tau is reduced by
   two. In either case the counter is reset to zero. Under normal
   conditions tau increases in stages from a default lower limit of 6
   (64 s) to a default upper limit of 10 (1024 s). However, if the
   wander increases because the oscillator frequency is deviating too
   fast, tau is quickly reduced. Once the oscillator wander subsides,
   tau is slowly increased again. Under typical operating conditions,
   tau hovers close to the maximum; but, on occasions of a heat spike
   when the oscillator wanders more than about 1 PPM, it quickly drops
   to lower values until the wander subsides.

7.2  State Machine

   The clock discipline must operate over an extremely wide range of
   network jitter and oscillator wander conditions without manual
   intervention or prior configuration. As determined by past
   experience and experiment, the data grooming algorithms work well to
   sift good data from bad, especially under conditions of light to
   moderate network and server loads. The PLL/FLL hybrid algorithm may
   perform poorly and even become unstable under heavy network loading.
   The state machine functions to bypass some discipline functions under
   conditions of hardware or software failure, severe time or frequency
   transients and especially when the poll interval is relatively large.

   Under normal conditions the NTP discipline algorithm writes the
   current frequency offset to a file at hourly intervals. Once the file
   is written and the daemon is restarted after reboot, for example, it
   initializes the frequency offset from the file, which avoids the
   training time, possibly several hours, to determine the intrinsic
   frequency offset when the daemon is started for the first time.
   When toll charges accrue for every NTP message, as in a telephone
   modem service, it is important to determine the presence of a a
   possibly large intrinsic frequency offset, especially if the interval
   between telephone calls must be 15 minutes or more. For instance,
   without the state machine it might take many calls spaced at 15
   minutes until the frequency offset is determined and the call
   spacing can be increased. With the state machine it usually takes

Kasch, et al.               Expires April, 2006                [Page 13]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   only two calls to complete the process.

   The clock state machine transition function is shown in Table 1. It
   determines the action and next state when an update with specified
   offset occurs in a given state shown in the first column.  The second
   column shows what happens if the offset is less than the step
   threshold, the third when the step threshold is exceeded but not the
   stepout threshold and the third when both thresholds are exceeded.
   The state machine responds to the current state and event to cause
   the action shown.

          Table 1. Clock State Machine Transition Function
======================================================================
| State |   abs(T) < STEP   |   abs(T) > STEP   |    Comments        |
----------------------------------------------------------------------
| NSET  | > FREQ; adjust    | > FREQ; step      | no frequency       |
|       | time              | time              | file               |
----------------------------------------------------------------------
| FSET  | > SYNC; adjust    | > SYNC; step      | frequency file     |
|       | time              | time              |                    |
----------------------------------------------------------------------
| SPIK  | > SYNC; adjust    | if (<900 s)>SPIK  | outlier detected   |
|       | freq, adjust time | else SYNC; step   |                    |
|       |                   | freq; step time   |                    |
----------------------------------------------------------------------
| FREQ  | if (<900 s)> FREQ | if (<900 s)>FREQ  | initial frequency  |
|       | else >SYNC; step  | else >SYNC; step  |                    |
|       | freq, adjust time | freq, adjust time |                    |
----------------------------------------------------------------------
| SYNC  | >SYNC; adjust freq| if (<900 s)>SPIK  | normal operation   |
|       | adjust time       | else >SYNC; step  |                    |
|       |                   | freq; step time   |                    |
----------------------------------------------------------------------

   The actions listed in the state diagram include adjust-frequency,
   step-frequency, adjust-time and step-time actions. The normal action
   in the SYNC state is to adjust both frequency and time. The step-time
   action is to set the system clock, while the step-frequency action is
   to calculate the frequency offset directly. This has to be done
   carefully to avoid contamination of the frequency estimate by the
   phase adjustment since the last update.

   The machine can be initialized in two states, FSET if the frequency
   file is present or NSET if it has not yet been created. If the file
   is not present, this may be the first time the discipline has ever
   been activated, so it may have to quickly determine the oscillator
   intrinsic frequency offset. It is important to realize that a number
   of NTP messages may be exchanged before the mitigation algorithms
   determine a reliable time offset and call the clock discipline
   algorithm.


Kasch, et al.               Expires April, 2006                [Page 14]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


   When the first valid offset arrives in the NSET state, (1) the time
   is stepped to that offset, if necessary, (2) the watchdog counter is
   started and (3) the machine exits to the FREQ state. Subsequently,
   updates will be ignored until the stepout threshold has been reached,
   at which time the frequency is stepped, the time is stepped if
   necessary, and the machine exits to SYNC state. When the first valid
   offset arrives in the FSET state, the frequency has already been
   initialized, so the machine does the same things as in the NSET
   state, but exits to the SYNC state.

   In the SYNC state the machine watches for outliers above the step
   threshold. If one is found, the machine exits to SPIK state and
   starts the watchdog timer. If another offset less than the step
   threshold is found, the counter is stopped and the machine exits to
   the SYNC state. If the watchdog timer reaches the stepout threshold,
   the time and frequency are both stepped as required and the machine
   exits to the SYNC state.


8.  Security Considerations

   There are no security considerations.


9. IANA Considerations

   There are no IANA considerations.


10.  Acknowledgements

11.  References

11.1  Normative References

11.2   Informative References

[NTP05]  Burbank, J., Martin, J., and Mills, D., "Network Time Protocol
         Version 4 Protocol Specification,"
         <draft-ietf-ntp-ntpv4-proto-00.txt>, July 2005, work in
         progress.

[MIL92]  Mills, D., "Network Time Protocol (Version 3) Specification,
         Implementation", RFC 1305, March 1992.

[MIL96]  Mills, D., "Simple Network Time Protocol (SNTP) Version 4 for
         IPv4, IPv6, and OSI", RFC 2030, October 1996.

[DOL95]  Dolev, D., J. Halpern, B. Simons, and R. Strong, "Dynamic
         Fault-Tolerant Clock Synchronization," JACM 42, January 1995,
         pp. 143-185.

Kasch, et al.               Expires April, 2006                [Page 15]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


[BER00]  Berthaud, J. M., "Time Synchronization over Networks using
         Convex Closures," IEEE/ACM Transactions on Networking, April
         2000, pp. 265-277.


12.  Authors' Addresses

   William T. Kasch (Editor)
   The Johns Hopkins University Applied Physics Laboratory (JHU/APL)
   11100 Johns Hopkins Road
   Laurel, MD  20723

   Phone: +1 443-778-7463
   EMail: william.kasch@jhuapl.edu


   Jack L. Burbank (Editor)
   JHU/APL
   11100 Johns Hopkins Road
   Laurel, MD 20723

   Phone: +1 443-778-7127
   EMail: jack.burbank@jhuapl.edu


   Dr. David L. Mills
   The University of Delaware
   Electrical Engineering Department
   University of Delaware
   Newark, DE 19716

   Phone: (302) 831-8247
   EMail: mills@udel.edu



















Kasch, et al.               Expires April, 2006                [Page 16]


Internet-Draft         NTPv4 Algorithms Specification       October 2005


Intellectual Property Statement

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at
   http://www.ietf.org/ipr.

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at
   ietf-ipr@ietf.org.


Disclaimer of Validity

   This document and the information contained herein are provided on an
   "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
   OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
   ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
   INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
   INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
   WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.


Copyright Statement

   Copyright (C) The Internet Society (2005).  This document is subject
   to the rights, licenses and restrictions contained in BCP 78, and
   except as set forth therein, the authors retain all their rights.


Acknowledgment

   Funding for the RFC Editor function is currently provided by the
   Internet Society.





Kasch, et al.                 Expires April, 2006              [Page 17]

Html markup produced by rfcmarkup 1.129c, available from https://tools.ietf.org/tools/rfcmarkup/