JP2007158425A - Time synchronization method, time client, time server, application apparatus, and time synchronization system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To achieve precise time synchronization even if a periodical application packet is multiplexed on the communication path between a time client and a time server in the time synchronization by NTP (Network time protocol). <P>SOLUTION: The time client 3 comprises a timing generator 11 for generating a pulse I(t) of a basic period I in the transmission of a time request packet, a delay generator 12 for giving delay time (d) changing arbitrarily to the pulse I(t), and a time request packet generator 13 that generates a time request packet when a delay pulse O(t) is outputted from the delay generator 12 for transmitting to a time server. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a method and system for synchronizing time between two or more devices connected by communication means, and in particular, a time synchronization method and system for performing time synchronization with high accuracy via communication means such as the Internet. About.

  As a conventional technique for performing time synchronization between two or more devices connected via communication means, there is an NTP (Network Time Protocol) defined in RFC1305 which is an Internet standard (Non-patent Document 1). NTP uses an IP (Internet protocol) network such as the Internet as a communication means, and a server (also referred to as a time server) that provides a time reference for such an IP network and a client that matches the time with this server (time client) Is also a protocol that functions between the server and the client. In NTP, the local time of the client is calculated by using the round-trip transmission delay time 2d of the packet when the packet is reciprocated between the server and the client, and the difference Δt between the time indicated by the server and the local time of the client. Adjust the clock to synchronize the client local time with the server time. Hereinafter, a specific procedure of time synchronization by NTP will be described.

  In NTP, in order to calculate the relative difference Δt between the round-trip transmission delay time 2d and the client local clock server, the client first sends a message to the server about the local time of the client. In response to this message, the server returns a response with a start timestamp, a receive timestamp, and a send timestamp. At the time of this response, the server copies the time sent from the client to the start time stamp. Then, the reception time stamp and the transmission time stamp are set to the date and time of the server clock.

The client manages the variables t ₁ , t ₂ , and t ₃ for recording the start time stamp, the reception time stamp, and the transmission time stamp included in the response from the server, respectively, and further, the response measured by the local clock of the client manages the variable t ₄ to record the end time stamp is the reception time, when receiving the response of the server, the arrival time as measured by the client of the clock, set to end timestamp variable. The table below summarizes the four time stamps.

Using these variables, the round trip transmission delay time 2d is 2d = {(t ₄ −t ₁ ) − (t ₃ −t ₂ )}.
It is represented by

When the delay amount backward from the forward and the server from the client to the server to the client is assumed to be equal, the time that the client has received the response, the value t ₃ when the server is added to the delay amount of the one-way time which has transmitted the response + d Therefore, the error Δt of the client clock with respect to the server clock can be obtained as follows.

Δt = t ₀ − (t ₃ + d) = {(t ₁ −t ₂ ) − (t ₃ −t ₄ )} / 2
Therefore, in NTP, the client's local clock can be synchronized with the server clock by correcting the client clock so that Δt = 0.
Network Time Protocol (Version3) Specification, Implementation and Analysis, David L. Mills, IETF RFC-1305

  As described above, in the NTP, when calculating the error Δt of the local clock in the client, the assumption that the transmission delay amount between the server and the client is equal in the round trip is used. Therefore, when there is a delay difference between the forward path and the backward path, this assumption is not satisfied, so that the accuracy of NTP deteriorates or an offset occurs.

  Consider a case where a time request packet by NTP and a packet of another application are packet-multiplexed on a communication path between a client and a server. When such application packets have periodicity and the period is highly stable or synchronized with the server time, the period of the time request packet sent by the time client and the period of the application packet match or are simple. May be a ratio. At that time, depending on the phase relationship between the two, it is possible that the forward time request packet collides with the application packet in the router or switch that multiplexes the packets and always receives a certain delay. In such a case, the assumption that the transmission delay amount is the same in both round trips does not hold.

  Even on the time server side, the time request packet from the time client is affected by the periodicity of the application packet going from the client side to the server side (outbound), so the cycle of the application packet going from the server side to the client side (return) In the same manner, the time packet on the return path may always be subjected to a certain delay.

  As described above, when a time request packet and a time response packet are transmitted via a communication path in which application packets are multiplexed, in NTP, depending on the combination of the period and phase of the time request packet, the forward path and the return path application packet, There is a possibility that an offset of ± (application packet time length) may occur, which is an obstacle to time synchronization with high accuracy.

  Standards such as IEEE 802.1q and Diffserver define priority control that prioritizes the transmission of specific types of packets, but in store-and-forward transfers that are generally used in IP routers / switches, output is in progress. Since the transfer of the next packet cannot be started unless the current packet ends, even if priority control of the time synchronization packet (time request packet and time response packet) is performed according to these standards, the above-described offset problem is solved. It is not possible. This is because even if the time synchronization packet arrives while the application packet is output, the time synchronization packet is output after waiting for the end of the application packet.

  The present invention has been made in view of the above circumstances, and it is an object of the present invention to provide a method and system capable of time synchronization with high accuracy.

  The time synchronization method of the present invention in a time client is a time synchronization method for connecting to a time server via a communication means and synchronizing the time of the local clock with the time held by the time server. Generating a time request packet by giving a delay time arbitrarily changing with respect to the basic period for each basic period, and sending the generated time request packet to a time server. And

  The time synchronization method of the present invention in a time server is a time synchronization method for receiving a time request packet from a time client, generating a time response packet based on the received time request packet, and sending the generated time response packet to the time client. In claim 1, when a time request packet is received, a time response packet corresponding to the received time request packet is transmitted after giving a delay time which changes arbitrarily.

  The time synchronization method of the present invention in the application server is a time synchronization method in a communication means to which a time client can be connected. In the application apparatus connected to the communication means, a step of generating a basic cycle of sending application packets; Each of the basic periods includes a step of generating the application packet by giving a delay time arbitrarily changing with respect to the basic period, and sending the generated application packet to the communication means.

  The time client of the present invention is characterized in that in the time client in a system that synchronizes time between two or more devices, the time request packet is sent out of a certain period.

  Another time client of the present invention includes a local clock, is connected to a time server via a communication means, and a time client that synchronizes the time of the local clock with the time held by the time server is the basic of sending a time request packet. Timing generating means for generating a period; and a packet generating means for generating a time request packet by giving a delay time d arbitrarily changing with respect to the basic period, and sending the generated time request packet to a time server.

In the time client of the present invention, the arbitrarily varying delay time d is such that L is the maximum packet time length of the application packet or transmission path in the communication means, L _tp is the packet length of the time request packet, G is the minimum interval between packets, I As a fundamental period, for example,
0 ≦ d ≦ d _max (where L ≦ d _max ≦ I−L _tp −G)
It is a delay time that is randomly determined so as to satisfy. Here, it is preferable to set d _max = L.

  The time server of the present invention receives the time request packet from the time client via the communication means, and sends the time response packet to the time client in response to the time request packet from the time when the time response packet is received. And a means for sending a time response packet after giving a delay time d which changes arbitrarily.

In the time server of the present invention, the delay time d that is arbitrarily changed may be set such that L is an application packet in the communication means or the maximum packet time length of the transmission path,
0 ≦ d ≦ d _max (where L ≦ d _max )
It is a delay time that is randomly determined so as to satisfy. Here, it is preferable to set d _max = L.

  Specifically, such a time server includes, for example, a plurality of buffers each capable of storing a time request packet, and one of the empty buffers among the plurality of buffers when the time request packet is received. Input switch means for storing a time request packet at a time, and scheduler means for selecting one of a plurality of buffers at random, and selecting the next buffer after a lapse of random time if the selected buffer is empty, And time response packet generating means for generating and sending a time response packet based on the time request packet stored in the buffer when the buffer selected by the scheduler means is not empty. Alternatively, the time server, for example, stores a plurality of buffers each capable of storing a time request packet and, when receiving the time request packet, stores the time request packet in one of the plurality of buffers. When the time request packet is stored in the buffer, the generation reservation time of the time response packet corresponding to the time request packet is randomly set, and the buffer corresponding to the generation reservation time is set. Scheduler means for selecting, and time response packet generating means for generating and sending a time response packet based on the time request packet stored in the buffer selected by the scheduler means.

  An application apparatus of the present invention includes a local clock, and an application apparatus that transmits application packets to communication means. Timing generating means for generating a basic period for transmitting application packets, and an arbitrary change with respect to the basic period. A packet generation unit configured to generate an application packet by giving a delay time d, and to transmit the generated application packet to the communication unit.

In the application apparatus of the present invention, the delay time d that arbitrarily changes is as follows: L is the maximum packet time length of the application packet or transmission path in the communication means, L _tp is the packet length of the time request packet, G is the minimum interval between packets, I As the basic period of application packet transmission, for example,
0 ≦ d ≦ d _max (where L ≦ d _max ≦ I−L _tp −G)
It is a delay time that is randomly determined so as to satisfy. Here, it is preferable to set d _max = L.

  In the present invention, the time client receives the time response packet corresponding to the time request packet to determine the round trip transmission delay time, and corrects the local clock based on the minimum of the plurality of round trip transmission delay times. It is preferable to make it.

  The time synchronization system of the present invention includes the time client of the present invention and the time server of the present invention. Or the time synchronization system of this invention has the time client of this invention, the time server of this invention, and the application apparatus of this invention.

  In the present invention, the time request packet is generated and transmitted by giving a delay time arbitrarily changing with respect to the basic period of generation of the time request packet in the time client, and / or the time request packet is transmitted in the time server. When receiving, after giving an arbitrarily changing delay time, a time response packet corresponding to the time request packet is sent, so that even in an environment where there is periodic application traffic, a router on the path In the (switch), there are cases where the forward time request packet and the backward time response packet pass without colliding with the application packet, so that highly accurate time synchronization is possible.

  Similarly, in an application apparatus that transmits application packets having periodicity, the time periodicity of the time request packet and the time response packet is also reduced in the router (switch) on the route by reducing the periodicity of the traffic due to the application packet. Since there are cases where the packet passes without colliding with the application packet, highly accurate time synchronization becomes possible.

  By using the time server, time client, and application device of the present invention in combination, the rate at which time request packets and time response packets pass through routers (switches) without periodic collisions increases, and more accurate time Synchronization is possible.

  Next, a preferred embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 shows a basic configuration of the time synchronization system. A time server 2 holding time with high accuracy is connected to a network 1 such as the Internet, and further has a local clock and a time to synchronize the local clock with the time server 2 Client 3 is connected. In the network 1, other application packets having periodicity are also transmitted. Therefore, in this embodiment, although NTP is basically used as a method for time synchronization, in order to reduce the influence of other application packets having periodicity on the accuracy of time synchronization, the conventional time by NTP is used. Several improvements will be made to the synchronization system as described below.

(Time client)
First, in the time client, the generation timing of the time request packet is improved so that the time request packet is transmitted at an interval that is randomly deviated from a certain period. FIG. 2 is a diagram showing the internal configuration of such a time client 3 and the transmission timing of the time request packet. FIG. 2 shows a portion related to transmission of the time request packet in the time client.

  The time client has a local clock (not shown in FIG. 2) of the time client, a trigger generator 11 that generates a pulse I (t) having a period I synchronized with the local clock, and a pulse I (t). A delay generator 12 that gives a delay to generate a delay pulse O (t) and a time request packet generator 13 that sends a time request packet based on the delay pulse O (t) are provided. Here, the period I is a constant period, and this is referred to as a time request packet transmission basic period I. The delay generator 12 receives the pulse I (t), and when the pulse I (t) arrives, gives a delay d (t, k) determined by a random number for each period, and outputs the delay pulse O (t). It is configured to output. The time request packet generator 13 is configured to send a time request packet using the client time as a start time stamp at the moment when the delay pulse O (t) arrives.

In such a time client 3, if the local time in the client is represented by _Tc ,
T _c = k · I + d (k) + T ₀
At this time, a time request packet is transmitted. Here, k is an arbitrary integer, T ₀ is a constant, and d (k) is 0 to d _max (where L ≦ d _max ≦ _ILTP− G; L is the maximum of the application packet or transmission path) The packet time length, L _TP is a time synchronization packet time length, and G is an arbitrary variation value that takes the value of the minimum interval between packets. It is preferable that d (k) is determined based on a random number.

In the time client 3 shown in FIG. 2, the improvement as described above is added, so that the transmission timing of the time request packet has a dispersion of one packet time length or more. This is shown in FIG. 2 by the fact that the transmission timing of each time request packet 14 varies as d (k), d (k + 1),... From the timing determined by the period I. Yes. As a result, there is a case in which an outgoing time request packet passes through an application packet having any period without colliding with an application packet in a router (or switch). Note that if there is a variance of one packet time length or more at the time request packet transmission timing, it may pass through the router (or switch) without colliding with the application packet, so the application packet or the maximum packet of the transmission path When the time length is L, L is sufficient as the maximum value d _max of d (k).

  FIG. 3 is a state transition diagram showing an example of state transition of the delay generator 12 in the time client shown in FIG. The default output value is O (t) = 0, and the process starts from state A. The state is in the state A until the input pulse I (t) arrives. When the pulse I (t) is input, a random number from 0 to L is stored in the variable D and the state transitions to the state B. In the state transition diagram, a function for generating a random number is represented by Rand (). In state B, 1 is subtracted per unit time until D becomes zero. When D = 0, the output value O (t) = 1 and the state A is entered.

(Burst control)
When it is desired to correct the time error at a substantially constant cycle, such as when controlling a VCO (voltage controlled oscillator), the following improvements are added to the conventional time synchronization system using NTP.

The time server 2 and the time client 3 are connected to the network 1 in the same manner as shown in FIG. 2 except that the time client 3 has a burst cycle of I and the basic transmission of time request packets within the burst. If the period is B and the local time at the client is T _c ,
T _c = k · I + h · B + d (k, h) + T ₀
The time request packet is transmitted at the time. Here, as described above, k is an arbitrary integer, T ₀ is a constant, and L is the maximum packet time length of an application packet or a transmission path. Further, here, assuming that the burst length is h _max , h is an integer from 0 to h _max , and B is a constant satisfying B ≧ L + L _TP + G. L _TP is the time synchronization packet time length, and G is the minimum interval between packets. d (k, h) is an arbitrary fluctuation value that takes any value of {0, 1, 2,..., L}.

When a time request packet is transmitted at such a time T _c , the probability that there is a time request packet that does not collide with an application packet in each burst is high, so control in the burst cycle becomes possible.

FIG. 4 shows the state transition of the delay generator 12 when such burst control is performed. The default output value is O (t) = 0, and the process starts from state A. Until input pulse I (t) arrives, it is in state A. When pulse I (t) is input, 0 is stored in variables g and h, and random numbers from 0 to L are stored in d. , Transition to state B. In state B, 1 is decreased until d becomes 0 every unit time, and g is increased by 1.
When d = 0, the output value O (t) = 1, and if h <h _max , transition is made to state C, and if h ≧ h _max , transition is made to state A.

  In state C, g is incremented by 1 for each unit time while g <B. When g ≧ B, 0 is stored in g, a random number from 0 to L is stored in d, h is incremented by 1, and a transition is made to state B.

(RTT filter)
As described above, in the method of shifting the transmission timing of the time request packet on the time client 3 side, the start time stamp is t ₁ , the reception time stamp is t ₂ , the transmission time stamp is t ₃ , and the end time stamp is t ₄ . The time information of the time response packet that minimizes the round-trip transmission delay time RTT = (t ₂ −t ₁ ) + (t ₄ −t ₃ ) among the time response packets from the time server 2 in each burst. Based on the update of the client time or the VCO control based on this, high-precision synchronization becomes possible. This is because there is a high probability that an accurate response packet of time information that is not subjected to excessive delay in a router (switch) on the communication path is selected.

  FIG. 5 is a block diagram illustrating an example of a configuration of a time client that performs time synchronization according to a time response packet having the smallest RTT among time response packets in a burst. FIG. 5 shows a configuration of a part related to reception of the time response packet in the time client.

The time client shown in FIG. 5 includes a local clock 41 of the time client, and is configured to synchronize the time of the local clock 41 with the time of the clock of the time server. The time client, based on the time signal from the local clock 41, generates a burst trigger B (t), a burst trigger generation circuit 42, a reception circuit 43 that receives a time response packet, and a transmission that calculates a round trip transmission delay time RTT. A delay time calculation circuit (RTT) 44, a client time error calculation circuit (Δ) 45 for calculating a time error Δ between the server clock and the local clock 41, and a minimum transmission delay in the burst based on the round trip transmission delay time RTT Intra-burst minimum transmission delay time calculation circuit 46 for calculating time RTT _burst , Intra-burst client time error calculation circuit 47 for calculating client time error Δ _{burst in burst} based on client time error Δ, burst trigger pulse, Based on the client time error Δburst and the minimum transmission delay in the burst Client time error calculating circuit 48 for calculating a client time error dt based on RTT _burst, provided with a time correction circuit 49 to correct the local clock time 41, the based on the client time error dt. Note that the time signal from the local clock 41 is also supplied to the receiving circuit 43 in order to give the end time stamp t ₄ .

  Next, the operation of this time client will be described.

When a time request packet is sent to a time server (not shown) by a time request packet transmission circuit (not shown) based on the local clock 41, the time server transmits a time response packet to this time client. When the time response packet is received by the receiving circuit 43, four time stamps included in the time response packet (start timestamp t _1, reception timestamp t _2, transmission time stamp t ₃ and end timestamp t ₄₎ 411 is input to the transmission delay time calculation circuit 44 and the client time error calculation circuit 45. The transmission delay time calculation circuit 44 calculates the round-trip transmission delay time RTT and supplies the calculation result to the intra-burst minimum transmission delay time calculation circuit 46. The client time error calculation circuit 45 calculates the client time error Δ and calculates it. The result is supplied to the intra-burst client time error calculation circuit 47.

The intra-burst minimum transmission delay time calculation circuit 46 is reset to RTT _burst = ∞ with a pulse from the burst trigger generation circuit 42. When the RTT calculation result 412 is input, the calculation result is used as the current value of the RTT _burst. Compare and place the smaller value in the RTT _burst . At this time, if the value of RTTb _burst has been updated, an update pulse 413 is output and supplied to the intra-burst client time error calculation circuit 47. The intra-burst client time error calculation circuit 47 holds the value of the client time error Δ when the update pulse 413 is output, and outputs the value as the intra-burst client time error Δ _burst .

The client time error calculation circuit 48 captures the in- _burst client time error Δ _burst in synchronization with the pulse of the burst trigger B (t) from the burst trigger generation circuit 42 and outputs it as client time information 417. The time correction circuit 49 corrects the time of the local clock 41 based on the information of the client time error 417.

In this way, the time client shown in FIG. 5 corrects the time of the local clock 41 based on the minimum transmission delay time RTT _burst in the _burst .

(Time server)
Next, the improvement performed on the time server 2 side for solving the problems in the conventional time synchronization by NTP will be described.

  When a periodic application packet and a time request packet are multiplexed in a router (or switch) in the network, the time request packet is placed between application packets, and thus is strongly affected by the periodicity of the application. It will be. Assuming that the processing delay in the time server is constant, the time response packet from the time server is multiplexed with a periodic (same or different) application packet in the router (or switch) as in the case of the time request packet. In such a case, the time response packet may always collide with the return path application packet when the periods of the forward path and return path application packets coincide or become a simple ratio.

  Therefore, as an improvement in the time server 2, the time server 2 receives the time request packet and transmits the time response packet in the same manner as the time client 3 has a fluctuation in the transmission cycle of the time request packet. It is conceivable to set fluctuations in the time interval. FIG. 6 shows the internal configuration of such a time server 2.

  The time server includes a time counter 51 as a highly accurate clock in the time server, a buffer 53 that stores the time request packet 510, a delay generator 52 that gives a delay to the time request packet 510, and a time response packet. And a time response generator 54 for generating and transmitting.

In this time server, when the time request packet 510 arrives, the current value 512 of the time counter 51 is input to the buffer 53 together with the time request packet 510 as the reception time stamp t ₂ . When the time request packet 510 arrives, the delay generator 52 outputs an output pulse 515 after a delay d (t, k) determined by a random number for each packet. The time response generator 54 starts reading the time request packet 514 from the buffer 53 using the output pulse 515 as a trigger, and the counter value of the time counter 51 at this time is included in the time request packet 514 as a transmission time stamp t _3. The time response packet 511 including the received information, the reception time stamp t ₂ and the transmission time stamp t ₃ is generated and transmitted.

In such a time server, when the reception time of the time request packet is T _RX and the time processing time in the server is S ₀ ,
T _S = T _RX + S ₀ + d (k)
At time T _S , a time response packet is transmitted. Here, k is an arbitrary integer, and d (k) is an arbitrary fluctuation value that takes a value from 0 to d _max (d _max ≧ L; L is the maximum packet time length of an application packet or a transmission path).

  By adding such improvements to the time server, the transmission timing of the time response packet from the time server has a dispersion of one packet time length or more. Therefore, any application packet of any period can be received at the router (or switch). There is a case where the return time response packet passes without colliding with the application packet.

(Buffer control)
By the way, in general, the time client only performs time synchronization with respect to one time server. However, in the time server, it is necessary to process time synchronization requests from a large number of time clients, and time requests that arrive frequently. Must correspond to a packet. Therefore, the time server shown in FIG. 7 includes a plurality of buffers for storing time request packets, and sets fluctuations in the time interval from reception of each time request packet to transmission of the corresponding time response packet. It is a thing.

  That is, the time server shown in FIG. 7 generates a time counter 61 as a highly accurate clock in the time server, N (where N ≧ 2) buffers 63, and a time response that generates and transmits a time response packet 611. A generator 64, an input switch 65 that distributes the time request packet 610 that has arrived to one of the buffers 63, an output switch 66 that selects the buffer 63 and supplies the time request packet 613 to the time response generator 64, and an output switch And a scheduler 67 for controlling 66. A signal 616 indicating the usage status of the buffer is supplied from each buffer 63 to the input switch 65. The input switch 65 outputs a signal 617 indicating the usage status of each buffer to the scheduler 67. The output switch 66 outputs an output pulse 615 for starting output of the time request packet to the selected buffer 63.

In this time server, when the time request packet 610 arrives, the input switch 65 selects one of the empty buffers based on the signal 616 indicating the usage status of each buffer 63, and sets the current value 612 of the time counter 61. The received time stamp t ₂ is stored in the buffer 63 together with the arrival time request packet 610. The scheduler 67 selects one of the N buffers 63 from the signal 617 indicating the buffer usage status in a predetermined procedure, and sends a selection signal 618 indicating the selection result to the output switch 66. The output switch 66 sends an output pulse 615 to the buffer indicated by the selection signal 618. As a result, reading from the buffer 63 is started. The time response generator 54 uses the counter value of the time counter 61 at the time of reading from the buffer as the transmission time stamp t ₃ , and uses the information included in the time request packet 614, the reception time stamp t ₂ and the transmission time stamp t ₃ . A time response packet 611 including this is generated and transmitted.

In such a time server, the information on the time request packet and the reception time T _RX is temporarily stored in one of the N buffers in the reception side. On the transmission side, one of N buffers is selected at random, and if the selected buffer is not empty, a time response packet is generated and transmitted, and then the process returns to the selection of the next buffer. If it is empty, d (k) (where d (k) is an arbitrary variation value ranging from 0 to d _max (d _max ≧ L)) and returns to the selection of the next buffer. Here, L is the maximum packet time length of the application packet or the transmission path. In this way, in this time server, it is possible to distribute the delay time in the server with respect to the arrival of a high frequency time request packet and transmit the time response packet.

  FIG. 8 shows an example of state transition of the scheduler 67 in the time server shown in FIG. The output value O (t) of the selection signal 618 by default is null, that is, nothing is selected, and the initial state is A. From this initial state A, a random number from 1 to N is assigned to the variable s, and a transition is made to state B. The signal buf [s] indicating the usage status of the s-th (where s ≦ N) buffer is referenced. When the s-th buffer is not empty, s is output to the selection signal 618 and the state C is transited. In the state C, s is output to the selection signal 618 until the s-th buffer becomes empty, and when it becomes empty, the next buffer is selected and the state B is changed. In this case, the next buffer is also selected based on the random number.

  On the other hand, when the s-th buffer is empty in state B, a random number from 0 to L is assigned to variable d and the state transitions to state D. In state D, d is decremented by 1 and when it becomes 0 or less, the next buffer is selected and transition is made to state B.

(Another example of buffer control)
In the case of a time server having a plurality of buffers, a random number is not used for selecting a buffer for reading a time request packet, but the timing for reading the time request packet at the time of accumulating the time request packet in the buffer is determined in advance based on the random number. You can also. In the case of performing such control, in the time server shown in FIG. 7, on the time request packet receiving side, the information on the time request packet and the reception time _TRX is temporarily stored in an empty buffer among the N buffers. Accumulate in one of the At the same time, the scheduler 67 sets the time response packet generation reservation time to T _RX + d (k) (where d (k) is an arbitrary fluctuation value that takes a value from 0 to d _max (d _max ≧ L)). The generation reservation time is set in this way, and when the reservation time is reached, the scheduler 67 selects the corresponding buffer. As a result, the time response generator 64 generates and transmits a time response packet in the same manner as described above. If another time response packet is being transmitted at the reserved time, after waiting for the end of the currently transmitted packet, the corresponding time response packet is generated again using the transmission time stamp t ₃ at that time. -Sending can be performed.

  FIG. 9 is a state transition diagram showing the operation of the scheduler 67 when it is configured to set the generation reservation time. In this case, the state transition (states C and D) related to the signal 617 indicating the usage status of the buffer 63 and the state transition (states A and B) related to the selection signal 618 are independently changed. The default selection signal 618 is null as in FIG.

  In the state transition related to the signal 617 indicating the use status of the buffer, in the initial state C, the input of the time request packet I (t) is waited. When the time request packet arrives, the time request packet is accumulated in the buffer, and The generation reservation time of the time response packet corresponding to the time request packet I (t) is set to delay {I (t)}, and the delay {I (t)} is later than the current time t by a random number from 0 to L. A time is set and the state D is entered. When the time request packet ends, the state D is returned to the state C.

  On the other hand, in the state transition related to the selection signal 618, when the function Select (t) that returns one of the buffers whose reservation time has passed and the initial state is A returns a valid value s, s is selected as the selection signal O (t). And transition to state B. In the state B, the completion of output from the corresponding buffer (s-th buffer) is waited, the generation reservation time delay [s] is reset, and the state A is returned.

(Server and client system)
As described above, the time client and the time server according to the present invention have been described. By combining these time client and time server, there is an environment where application traffic having periodicity exists in the communication path between the client and the server. In this case, highly accurate time synchronization is possible. FIG. 10 shows an example of a network configuration of a server-and-client system using the above-described time client and time server. The time client 91 is connected to the time server 92 via the server side router 94 via the client side router 93. Further, it is assumed that network devices 95 and 96 are connected to routers 93 and 94, respectively, and send traffic to each other.

  FIG. 11 schematically shows packet arrival and transmission timings in each router in the server-and-client system having such a network configuration.

  (1) shows the arrival status of the packet at the router 93 on the client side. In this example, the basic period of the time request packets A1 to A3 from the time client 91 and the basic period of the transmission packets B1 to B3 from the network device 95 are the same, but in the time client 91, the time request packets A1 to A3. Since the time request packets A1 and A2 collide with the transmission packets B1 and B2 from the network device 95, the network device 95 is similar to the time request packet A3. There is a time request packet that does not collide with any other packet. (2) shows a packet transmission situation from the client side router 93 under such a situation. When the time request packet and the transmission packet from the network device 95 collide, as shown in the time request packets A1a and A2a, the time request packet is transmitted after the packets B1a and B1b from the network device, and excessively. You will be delayed. This causes a time synchronization error. However, the time request packet A3a is sent from the router 93 without being delayed by the packet B3a. Therefore, if time synchronization is performed based on the time request packets A3 and A3a, accurate time synchronization can be achieved.

  The same can be said for the return trip from the time server 92. (3) shows the arrival status of packets in the router 94 on the server side, A1c, A2c, and A3c represent time response packets from the time server, and C1, C2, and C3 represent packets sent from the network device 96. It represents. Since the time server also sends a time response packet with an arbitrary fluctuating delay after receiving the time request packet, there is a case where the time server does not collide with the transmission packet of the network device 96 like the time response packets A1b and A3b. . (4) shows the packet transmission status from the router 94 on the server side. As shown in (4), the collided packet A2b is transmitted from the router 94 after the transmission packet C2b from the network device 96. And an excessive delay causes time synchronization errors. A case where the packet does not collide with other packets in both round trips, such as A3, A3a, A3b, and A3c, is detected by calculating the transmission delay time, and the client time is determined based on the packet time error information in such a case. By correcting the above, it becomes possible to establish synchronization between the time client 91 and the time server 92 with high accuracy.

(Application device)
Next, an improvement performed in an apparatus for sending application packets having periodicity for solving the problems in conventional time synchronization by NTP will be described. In the above-mentioned time client, the time request packet generation timing is improved, and the time request packet is sent out at an interval that is randomly deviated from a certain period, so that the time request packet, the time response packet, and other applications are transmitted. There is always a case where there is no collision with the application packet from. Here, even when the transmission timing of the application packet is shifted from a certain period instead of the time request packet, the case where the time request packet or the time response packet does not collide with other application packets can be surely generated.

  FIG. 12 is a block diagram showing an application apparatus with improved application packet transmission timing. The application device here means a packet other than the time request packet, which generates an application packet having periodicity, and sends the generated application packet to the network.

  The application apparatus 70 includes a trigger generator 71 that generates a pulse I (t) having a period I (this is referred to as a basic period) synchronized with the internal clock of the application apparatus, and a delay to the pulse I (t). A delay generator 72 for generating (t) and an application packet generator 73 for transmitting a time request packet based on the delay pulse O (t). The delay generator 72 receives the pulse I (t) and, when the pulse I (t) arrives, outputs a delay pulse O (t) after a delay d (k) determined by a random number for each period. It is configured. The application packet generator 73 is configured to send an application packet at the moment when the delay pulse O (t) arrives.

In such an application device, when the local time in the application device is represented by T _c ,
T _c = k · I + d (k) + T ₀
The application packet is transmitted at the time. Here, k is an arbitrary integer, T ₀ is a constant, and d (k) is 0 to d _max (where L ≦ d _max ≦ _ILTP− G; L is the maximum of the application packet or transmission path) The packet time length, L _TP is a time synchronization packet time length, and G is an arbitrary variation value that takes the value of the minimum interval between packets. It is preferable that d (k) is determined based on a random number.

In the application device 70, the improvement as described above is added, so that the position of the packet in FIG. 12 varies from the timing determined by the period I as d (k), d (k + 1),. In addition, the transmission timing of the application packet has a variance of one packet time length or more. As a result, there are cases where the application packet passes through the router (or switch) without colliding with the time request packet and the time response packet. Note that if there is a dispersion of one packet time length or longer at the application packet transmission timing, there is a case where the packet passes through the router (or switch) without colliding with the application packet. When the length is L, L is sufficient as the maximum value d _max of d (k). In other words, by operating with d _max = L, it becomes possible to secure the time adjustment accuracy at the time clients sharing the network while reducing the dispersion of the packet reception timing at the application packet receiving side.

(Server and client system in the presence of application devices)
By combining the time client, time server, and application device according to the present invention, even when there is application packet traffic from the application device in the communication path between the client and the server, time synchronization can be performed with higher accuracy. It becomes like this. FIG. 13 shows an example of a network configuration using the above-described time client, time server, and application device. The time client 81 is connected to the time server 82 via the router 93 on the server side via the router 93 on the client side. In addition, application devices 85 and 86 are connected to routers 83 and 84, respectively, and send traffic to each other. In particular, in this configuration, the application devices 85 and 86 transmit application packets having periodicity. Since each of the application devices 85 and 86 has a configuration as shown in FIG. The basic period is sent with fluctuations.

  In this network system, the packet output timings at the time client 81, the time server 82, and the application devices 85 and 86 are distributed, so that packets do not collide with each other at the routers 83 and 84, that is, an excessive delay. There is a probability that the packet will be transmitted without receiving. The time client 81 corrects the client time centering on the information of the time synchronization packet with a small round-trip delay time, thereby further reducing the influence of the traffic of the application packet having periodicity and highly accurately It is possible to establish synchronization between 81 and the time server 82.

It is a block diagram which shows the basic composition of a time synchronization system. It is a block diagram which shows the time client by which the fluctuation was set to the space | interval of a time request packet. FIG. 3 is a state transition diagram of a delay generator in the time client shown in FIG. 2. FIG. 4 is a state transition diagram of another example of the delay generator in the time client shown in FIG. 2. It is a block diagram which shows an example of a structure of a time client. It is a block diagram which shows the time server by which the fluctuation was set to the space | interval of a time response packet. It is a block diagram which shows the time server which gives dispersion | distribution to the delay time in a server with respect to arrival of a high frequency time request packet, and sends a time response packet. It is a state transition diagram of the scheduler in the time server shown in FIG. FIG. 8 is a state transition diagram of another example scheduler in the time server illustrated in FIG. 7. It is a figure which shows an example of the network structure of the time synchronization system which has a time server and a time client. It is a timing chart explaining the delay time in the system shown in FIG. It is a block diagram which shows the application apparatus with which the fluctuation | variation was set to the interval of an application packet. It is a figure which shows an example of the network structure of the time synchronization system which has a time server, a time client, and an application apparatus.

Explanation of symbols

1 Network 2, 82, 92 Time Server 3, 81, 91 Time Client 11, 71 Trigger Generator 12, 52, 72 Delay Generator 13 Time Request Packet Generator 41 Local Clock 42 Burst Trigger Generation Circuit 43 Reception Circuit 44 Transmission Delay Time calculation circuit 45, 48 Client time error calculation circuit 46 Intra-burst minimum transmission delay time calculation circuit 47 In-burst client time error calculation circuit 49 Time correction circuit 51, 61 Time counter 53, 63 Buffer 54, 64 Time response generator 65 Input Switch 66 Output switch 67 Scheduler 70, 85, 86 Application device 71 Application packet generator 83, 84, 93, 94 Router 95, 96 Network device

Claims (18)

In a time synchronization method of connecting to a time server via a communication means and synchronizing the time of a local clock with the time held by the time server,
Generating a basic period for sending a time request packet; and
Generating a time request packet shifted from the basic period, and sending the generated time request packet to the time server;
A time synchronization method characterized by comprising: In a time synchronization method of connecting to a time server via a communication means and synchronizing the time of a local clock with the time held by the time server,
Generating a basic period for sending a time request packet; and
Generating the time request packet by giving a delay time arbitrarily changing with respect to the basic period for each basic period, and sending the generated time request packet to the time server. The time synchronization method. In the time synchronization method of accepting a time request packet from a time client, generating a time response packet based on the accepted time request packet, and sending the generated time response packet to the time client,
When the time request packet is received, after giving a delay time which changes arbitrarily, the time response packet corresponding to the received time request packet is transmitted. Receiving a time response packet corresponding to the time request packet to determine a round trip transmission delay time;
Modifying the local clock based on a minimum of a plurality of the round trip transmission delay times;
The time synchronization method according to claim 2, further comprising: In the time synchronization method in the communication means to which the time client can connect,
In the application device connected to the communication means, generating a basic cycle of sending application packets;
In the application device, for each basic period, generating the application packet by giving a delay time arbitrarily changing with respect to the basic period, and sending the generated application packet to the communication unit;
A time synchronization method characterized by comprising: In a time client in a system that synchronizes time between two or more devices,
A time client, characterized in that the time request packet is sent out of a certain period. In a time client comprising a local clock, connected to a time server via a communication means, and synchronizing the time of the local clock with the time held by the time server,
Timing generating means for generating a basic period of transmission of a time request packet;
A packet generation means for generating the time request packet by giving a delay time d arbitrarily changing with respect to the basic period, and sending the generated time request packet to the time server;
A time client characterized by comprising: The arbitrarily changing delay time d is such that L is the maximum packet time length of the application packet or transmission path in the communication means, L _tp is the packet length of the time request packet, G is the minimum interval between packets, and I is the basic period. As
0 ≦ d ≦ d _max (where L ≦ d _max ≦ I−L _tp −G)
The time client according to claim 7, wherein the time client is a delay time randomly determined so as to satisfy. Means for receiving a time response packet corresponding to the time request packet from the time server;
Means for determining a round trip transmission delay time based on the received time response packet;
Updating means for updating the time indicated by the local clock based on time information included in a time response packet corresponding to a minimum one of the plurality of round-trip transmission delays;
The time client according to claim 7 or 8, comprising: Burst time setting means for setting a burst time with a length longer than the basic period of transmission of the time request packet;
The time client according to claim 9, wherein the time is updated based on time information of the time response packet that minimizes the round trip transmission delay time within the burst time. In a time server that receives a time request packet from a time client via a communication means and sends a time response packet to the time client in response to the time request packet;
A time server comprising: means for transmitting the time response packet after giving a delay time d that arbitrarily changes from the time when the time response packet is received. The arbitrarily changing delay time d is L, where L is the application packet in the communication means or the maximum packet time length of the transmission path,
0 ≦ d ≦ d _max (where L ≦ d _max )
The time server according to claim 11, wherein the time server is a delay time that is randomly determined so as to satisfy. In a time server that receives a time request packet from a time client via a communication means and sends a time response packet to the time client in response to the time request packet;
A plurality of buffers each capable of storing the time request packet;
Input switch means for storing the time request packet in one of the empty buffers of the plurality of buffers when the time request packet is received;
Scheduler means for randomly selecting one of the plurality of buffers and selecting a next buffer after a lapse of random time if the selected buffer is empty;
Time response packet generation means for generating and sending the time response packet based on the time request packet stored in the buffer when the buffer selected by the scheduler means is not empty;
A time server comprising: In a time server that receives a time request packet from a time client via a communication means and sends a time response packet to the time client in response to the time request packet;
A plurality of buffers each capable of storing the time request packet;
Input switch means for storing the time request packet in one of the empty buffers of the plurality of buffers when the time request packet is received;
When storing the time request packet in the buffer, scheduler means for randomly setting a generation reservation time of a time response packet corresponding to the time request packet and selecting a buffer corresponding to the generation reservation time When,
Time response packet generating means for generating and sending the time response packet based on the time request packet stored in the buffer selected by the scheduler means;
A time server comprising: In an application device comprising a local clock and sending application packets to the communication means,
Timing generating means for generating a basic period of transmission of the application packet;
Packet generating means for generating the application packet by giving a delay time d arbitrarily changing with respect to the basic period, and sending the generated application packet to the communication means;
An application device comprising: The arbitrarily changing delay time d is such that L is the maximum packet time length of the application packet or transmission path in the communication means, L _tp is the packet length of the time request packet, G is the minimum interval between packets, and I is the basic period. As
0 ≦ d ≦ d _max (where L ≦ d _max ≦ I−L _tp −G)
The application device according to claim 15, wherein the delay time is randomly determined so as to satisfy.   A time synchronization system comprising: the time client according to any one of claims 6 to 10; and the time server according to any one of claims 11 to 14.   A time synchronization comprising the time client according to any one of claims 6 to 10, the time server according to any one of claims 11 to 14, and the application device according to claim 15 or 16. system.

Images