JP2001111623A - Distributed time synchronization method and distributed time synchronization system utilizing this method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a distributed time synchronization method that can realize more accurate distributed time synchronization by eliminating a delay part proportional to a length of a packet included as an error in a one-way delay time caused by different communication paths and using only the delay time component independently of the packet length and to provide the distributed time synchronization system employing this method. SOLUTION: In the case of conducting time synchronization between computers connected via a network, they make communication with each other by using a plurality of packets whose length are different and the times of the respective computers are synchronized on the basis of the difference between transfer delay times of a plurality of the packets whose length are different.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a distributed time synchronization method and a distributed time synchronization system using the method.
More specifically, a transmission delay compensation type dispersion that synchronizes the time between a time server having accurate time information on a network to which a plurality of computers such as the Internet are connected and a host computer at a remote place. The present invention relates to a time synchronization method and a distributed time synchronization system using the method.

That is, according to the present invention, in packet communication between two or more computers, communication is performed using packets having different packet lengths, and a transfer delay time is determined based on a difference in transfer delay time required for the communication. By calculating the delay time component that is proportional to the packet length, and using only the delay time that is removed, that is, only the transfer delay time that is independent of the packet length, in a network environment where the quality of bidirectional communication is heterogeneous And a method for more accurately synchronizing the times held by the two or more computers and a distributed time synchronization system using the method.

[0003]

2. Description of the Related Art As a conventional method for synchronizing time between computers connected via a network, NTP is used.
(Network Time Protocol) is the mainstream. This N
In the distributed time synchronization by the TP method, for example, communication is performed between two computers using fixed-length packets, and a half of the time required for a packet to reciprocate between the two computers is a one-way delay time. Then, the time of one computer is corrected using the numerical value, and the time is synchronized.

[0004] The one-way delay time used in this method includes a delay time component independent of the packet length and a delay component generated in proportion to the packet length. Hereinafter, a time synchronization method according to the related art will be described with reference to the drawings. A more detailed description of NTP can be found in, for example, Mills, “Improved Algorithms for Sync.
hr-onizing Computer Network Clocks ”(IEEE /
ACM TRANSACTIONS ON NETWOR
KING, VOL. 3, NO. 3, JUNE 1995).

FIG. 3 is a schematic explanatory diagram showing the basic configuration of a conventional distributed time synchronization system using the NTP method. In the figure, 2010 is a host computer, 2020 is a time server, 2030 is a fixed-length packet, 2040 and 2
070, 2080, 2110 are time entry sections, 2050,
Reference numeral 2090 denotes a packet transmitting unit, 2060 and 2100 denote packet receiving units, and 2130 denotes a host computer time correcting unit.

[0006] The fixed-length packet 2030 has the time held by the host computer 2010 entered by the time entry unit 2040 of the host computer 2010, and is transmitted to the time server 2020 by the packet transmission unit 2050. The fixed-length packet 2030 is received by the packet receiving unit 2060 of the time server 2020, and the time entry unit 2070 immediately writes the time held by the time server 2020.

The fixed-length packet 2030 in which the reception time is entered by the time of the time server 2020 is sent to the time entry unit 2080 by the time entry unit 2080 immediately before sending the packet.
The time of 020 is entered, and the packet is transmitted to the host computer 2010 by the packet transmitting unit 2090. The fixed-length packet 2030 is transmitted to the host computer 2.
010 received by the packet receiving unit 2100 and the time entry unit 2110 of the host computer 2010.
Thus, the time held by the host computer 2010 is entered.

At this point, four pieces of time information are written in the packet. Here, 2040 and 207 in FIG.
The times entered in the four time entry sections 0, 2080, and 2110 are T1, T2, T3, and T4, respectively.
Next, a method of calculating a time difference between two computers based on the four times obtained as described above will be described.

The difference between the time held by the host computer 2010 and the time held by the time server 2020 is represented by D
Considering T1, T2, T3, and T4 written in a packet having a packet length n that reciprocates between the host computer 2010 and the time server 2020 in the shortest time, T2-T1 = D + A1 + B1 * n (Equation 1) T4- T3 = -D + A2 + B2 * n (Expression 2) Here, A1, B1, A2, B2
Is a constant of the communication route in each direction.

In the conventional distributed time synchronization using NTP, the one-way delay time of packet transfer from the host computer 2010 to the time server 2020 and the time server 2
020 to the host computer 2010 under the assumption that the one-way delay time is equal.
The time difference between the clock of 010 and the clock of the time server 2020,
That is, the time is corrected by calculating D in Expressions 1 and 2 described above.

When expressed in mathematical formulas, the assumption that the one-way delay times in both directions are equal is A1 + B1 * n = A2 + B2 * n (Equation 3).
Difference D between the clock of the server and the clock of the time server 2020
Is expressed as D = ((T2-T1)-(T4-T3)) / 2 (Equation 4).

[0012]

By the way, in communication between arbitrary two points (A, B) in communication on a network such as the Internet, a communication path from A to B (outgoing) and a communication path from B to B There is no guarantee that the communication path in the direction (return) to A is the same, and the shortest delay time may differ greatly depending on the communication direction due to the difference in the bandwidth of the communication path between the forward and return paths. . Also, in the case of an Internet connection using a cable or a connection via a communication path whose bandwidth varies greatly depending on the direction even on the same route, the values of B1 and B2 in the above equation 3 differ greatly, and two computers are connected. An error in accurately synchronizing the time occurs.

That is, when the communication path changes, the performance, characteristics, and the like of the processing devices in the communication path change.
The delay time may change depending on the direction of communication. Thus, when the nature of the delay differs depending on the direction of communication,
In the conventional distributed time synchronization using NTP, a difference in a value in each direction of one-way communication, which is calculated as half of a round-trip time of a fixed-length packet, is included as an error from an actual value of a one-way delay time. There is.

The present invention has been made in view of the above circumstances, and an object of the present invention is to solve the problems in the prior art and to solve the problem.
When time synchronization is performed between two computers, the delay time component proportional to the packet length, which is included as an error in the calculation of one-way delay time due to the difference in the communication path, is removed, and the delay is independent of the packet length. It is an object of the present invention to provide a distributed time synchronization method capable of realizing more accurate distributed time synchronization using only a time component, and a distributed time synchronization system using this method.

[0015]

In order to achieve the above object, a distributed time synchronization method according to the present invention provides a method for synchronizing time between two or more computers connected via a network. The two or more computers communicate with each other using a plurality of packets having different packet lengths, and based on a difference in transfer delay time between the plurality of packets having the different packet lengths, a time held by each of the computers is determined. It is characterized by being synchronized.

In the calculation of the transfer delay time, a transfer delay time that occurs in proportion to the packet length is calculated using a difference between the transfer times of the plurality of packets having different packet lengths, and the delay time component is subtracted. And calculating the transfer delay time independent of the packet length.

Also, the distributed time synchronization method according to the present invention
A distributed time synchronization system using this method can be embodied as follows.

That is, the distributed time synchronization system according to the present invention is a system for performing time synchronization between two or more computers connected via a network, and at least one of the computers And a delay time calculating means for performing communication by a plurality of packets having different packet lengths between the computers, and calculating a transfer delay time proportional to the packet length based on a difference in the transfer delay time. The time held by the computer is synchronized based on the calculation result.

In the calculation of the transfer delay time by the delay time calculating means, a transfer delay time generated in proportion to the packet length is calculated by using a difference between the transfer times of the plurality of packets having different packet lengths. It is characterized in that a transfer delay time independent of the packet length is calculated by subtracting the delay time component.

In the distributed time synchronization method according to the present invention and the distributed time synchronization system using the method, in the time synchronization of two computers (host computer and time server) connected via a network, a plurality of different packets are synchronized. Exchange long packets. The packet is first sent from the host computer to the time server via the network, and the packet received by the time server is sent back to the host computer.

In this packet, the time at which the packet was sent from the computer and the time at which the packet arrived at the computer are written by the clocks of the computers. as a result,
Four pieces of time information are written in the packet.
These four times are respectively represented by T1, T2, T3, T4
And

That is, T1 is the time held by the host computer when the packet was sent from the host computer, T2 was the time held by the time server when the packet was received by the time server, and T2 was the time server. The time held by the time server when the packet is sent back to the computer is T3, and the time held by the host computer when the packet sent from the time server is received by the host computer is T4.

This operation is performed for a plurality of packets having different packet lengths, and the packet transfer delay from the host computer to the time server and from the time server to the host computer is calculated from the times written in each packet. Here, the one-way delay time calculated from each packet includes a time difference between the clocks of the host computer and the time server.

For this reason, it is not possible to directly know the exact one-way delay time. However, if the difference between the one-way delay times calculated for the packets having different packet lengths is taken, the time difference between the host computer and the time server is calculated. Alternatively, the delay time component independent of the packet length is removed, and only the delay time proportional to the packet length is obtained as a calculation result. As a result, a proportional constant that determines the relationship between the delay time and the packet length in proportion to the packet length is obtained.

In this manner, the delay time proportional to the packet length is calculated from the one-way delay time in each of the two communication paths from the host computer to the time server and from the time server to the host computer, and the packet time is removed. The error of the time synchronization generated when the delay time proportional to is different depending on the communication direction is removed, and more accurate time synchronization can be performed using only the delay time component independent of the packet length.

[0026]

Embodiments of the present invention will be described below in detail with reference to preferred embodiments shown in the accompanying drawings.

FIG. 1 is a schematic diagram showing the basic configuration of a distributed time synchronization system according to one embodiment of the present invention. FIG.
, 1010 is a host computer, 1020 is a time server, 1030 is a packet having a different packet length,
1040, 1070, 1080, 1110 are time entry units, 1050, 1090 are packet transmission units, 1060,
1100 is a packet receiving unit, 1120 is a delay time calculating unit proportional to the packet length, and 1130 is a host computer time correcting unit.

FIG. 1 shows a host computer 101.
0, it does not show the entire configuration of the time server 1020, but shows only those parts directly related to the present invention. Actually, processing for controlling the entire computer in addition to the illustrated components A device, a storage device such as a ROM and a RAM, and various other processing units are provided.

For a plurality of packets having different packet lengths, the time held by the host computer 1010 is entered by the time entry unit 1040 of the host computer 1010, and the time transmitted from the time server 1
020. Those packets are received by the packet receiving unit 1060 of the time server 1020, and immediately by the time entry unit 1070, the time server 1
020 is entered.

For the packet in which the reception time is written by the time of the time server 1020, the time of the time server 1020 is written by the time writing unit 1080 immediately before sending the packet, and then the packet is sent to the host computer 1010 by the packet sending unit 1090. Is done.

The packet is sent to the host computer 1
010 received by the packet receiving unit 1100, and the time entry unit 1110 of the host computer 1010.
Thus, the time held by the host computer 1010 is entered. At this point, four pieces of time information are written in the packet, and 1040, 1070, 1010 in FIG.
The times entered in the four time entry sections 80 and 1110 are referred to as T1, T2, T3 and T4, respectively.

That is, the host computer 101 when a packet is transmitted from the host computer 1010
T1 is the time held by the time server 1020 when the packet was received by the time server 1020, and T2 is the time held by the time server 1020 when the packet was received by the time server 1020.
The time held by the host computer 1010 when the packet sent from the time server 1020 is received by the host computer 1010 is assumed to be T4.

Based on these four pieces of time information, a delay time calculator 1120 that is proportional to the packet length calculates a delay time component proportional to the packet length, and calculates the value and T1, T2,
Based on T3 and T4, the host computer time correction unit 1130 in FIG. 1 synchronizes the time of the host computer 1010 and the time server 1020. FIG. 2 shows a calculation procedure performed by the delay time calculation unit 1120 proportional to the packet length. Based on this, a specific calculation method will be described below.

The time difference between the host computer 1010 and the time server 1020 is represented by D, and T1 and T1, which are written in a packet having a packet length n that reciprocates between the host computer 1010 and the time server 1020 in the shortest time.
Considering T2, T3, and T4, as described above, T2-T1 = D + A1 + B1 * n (Equation 1) T4-T3 = -D + A2 + B2 * n (Equation 2) Here, A1, B1, A2, B2
Is a constant of the communication path in each direction.

In the conventional distributed time synchronization using NTP, the computer A from the packet transfer
The time difference between the clocks of the computer A and the computer B, that is, D in Expressions 1 and 2, is calculated on the assumption that the one-way delay time to the computer and the one-way delay time from the computer B to the computer A are equal. The time is being corrected. Expressed by a mathematical expression, the assumption that the one-way delay time in both directions is equal is A1 + B1 * n = A2 + B2 * n (Equation 3) as described above, and based on this assumption, the relationship between the host computer and the time server is obtained. The time difference D of the clock is represented by D = ((T2-T1)-(T4-T3)) / 2 (Equation 4).

In the distributed time synchronization system according to the present invention,
The terms B1 * n and B2 * n expressed in the above equation 3 are removed, and the time difference D of the clock between the computers is calculated on the assumption that A1 = A2. The method of realizing this will be described below.

Communication similar to the above is performed between the host computer 1010 and the time server 1020 using packets having different packet lengths (for example, packet lengths m and n, where m> n), and writing in each packet. Time T1m, T2 for a packet of packet length m
m, T3m, and T4m, and T1n, T2n, T3n, and T4n for a packet having a packet length n, the following relationship is established between the times written in these packets.

For a packet of packet length m: T2m−T1m = D + A1 + B1 * m (Equation 5) T4m−T3m = −D + A2 + B2 * m (Equation 6) For a packet of packet length n: T4n-T3n = -D + A2 + B2 * n (Equation 8)

Here, from Equations (5) and (7), B1 = ((T2m−T1m) − (T2n−T1n)) / (mn) (Equation 9) Also, from Equations (6) and (8), B2 = ( (T4m-T3m)-(T4n-T3n)) / (mn) The values of B1 and B2 can be calculated as follows.

Then, under the assumption that A1 = A2 (Equation 11), from Equations 7, 8, 9, and 10,
Time difference D between clock between host computer and time server
Can be obtained by the following calculation (processing 1121 to
1124).

D = (((T2n−T1n) − (T4n−T3n)) − B1 * n + B2 * n) / 2 = ((((T2n−T1n) − (T4n−T3n)) − ((T2m−T1m) − (T2n−T1n) − (T4m−T3m) + (T4n−T3n)) * n / (mn)) / 2 (Equation 12)

As described above, in the conventional method, the time difference D between the clocks of the computers is calculated as in Expression 4 on the assumption that Expression 3 is satisfied. However, in the distributed time synchronization system according to the present invention, Assuming that Expression 11 is satisfied from the transfer times of packets having different packet lengths, the time difference D is obtained by the calculation of Expression 12.

Therefore, in the distributed time synchronization system according to the present invention, when B1 has a value different from B2, the error generated between the one-way transfer time obtained by the conventional method and the actual one-way delay time is reduced. It becomes possible to perform more accurate time synchronization.

According to the above embodiment, the delay time component proportional to the packet length is removed by calculation from the transfer delay time when a packet is transferred between the host computer 1010 and the time server 1020 connected via the network. And the host computer 101 based on the
0, since the time of the time server 1020 can be synchronized, it is possible to remove the time synchronization error caused by the difference in the delay time component proportional to the packet length in the forward and return communication paths, and to perform more accurate distributed time synchronization. Will be possible.

The above embodiment is an example of the present invention, and it goes without saying that the present invention is not limited to this.

[0046]

As described above in detail, according to the present invention, when a packet is transferred between two computers, such as a host computer and a time server, connected via a network. From the transfer delay time,
The delay time component proportional to the packet length is removed by calculation, and the time of the two computers can be adjusted based on the calculation. Therefore, the difference is caused by the difference in the delay time component proportional to the packet length in the forward and return communication paths. Time synchronization errors can be removed and more accurate distributed time synchronization can be performed.

That is, for a communication path, a delay independent of the packet length and a delay component proportional to the packet length can be calculated, so that a one-way delay of transfer experienced by a packet having an arbitrary packet length can be calculated. Can be estimated by calculation without actually transmitting and receiving packets.

[Brief description of the drawings]

FIG. 1 is a schematic diagram illustrating a basic configuration of a distributed time synchronization system according to an embodiment of the present invention.

FIG. 2 is a diagram showing a calculation procedure performed by a delay time calculation unit 1120 proportional to a packet length in FIG.

FIG. 3 is a schematic explanatory diagram showing a basic configuration of a conventional distributed time synchronization system using the NTP method.

[Explanation of symbols]

1010 Host computer 1020 Time server 1030 Packets of different packet lengths 1031 Packets of m length 1032 Packets of n length packets 1040, 1070, 1080, 1110 Time entry units 1050, 1090 Packet transmission units 1060, 1110 Packet reception units 1120 packet length Proportional delay time calculation unit 1130 Host computer time correction unit 1121 Time information reading 1122 One-way delay time difference calculation 1123 Calculation of one-way delay time component proportional to packet length 1124 Calculation of one-way delay time component independent of packet length

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G06F 15/177 680 H04L 11/20 102Z 9A001 H04L 7/00 G06F 1/04 351B F-term (Reference) 2F002 AA00 AD07 AF01 FA16 GA06 5B045 BB24 BB58 CC09 5B089 GA21 GB02 HA10 JA40 JB22 KA13 KB11 5K030 GA11 KA21 LA15 MB11 5K047 AA11 AA18 BB15 9A001 BB01 BB04 CC08 JJ05 JJ12

Claims (4)

[Claims] 1. A system connected via a network.
A time synchronization method between two or more computers,
The two or more computers communicate with each other using a plurality of packets having different packet lengths, and based on a difference in transfer delay time between the plurality of packets having the different packet lengths, a time held by each of the computers is determined. A distributed time synchronization method characterized by synchronizing. 2. When calculating the transfer delay time, a transfer delay time generated in proportion to a packet length is calculated using a difference in transfer time for a plurality of packets having different packet lengths, and the delay time component is calculated. 2. The distributed time synchronization method according to claim 1, wherein a transfer delay time independent of a packet length is calculated by subtracting the transfer delay time. 3. A connection 2 via a network.
A system for performing time synchronization between two or more computers, wherein at least one of the computers communicates with a plurality of packets having different packet lengths between the computers, and the difference in transfer delay time is determined. Distributed time calculating means for calculating a transfer delay time proportional to the packet length based on the calculated time, and synchronizing the time held by the computer based on the calculation result of the delay time calculating means. Synchronization system. In the calculation of the transfer delay time by the delay time calculation means, a transfer delay time generated in proportion to a packet length is calculated by using a difference in transfer time for a plurality of packets having different packet lengths, 4. The distributed time synchronization system according to claim 3, wherein a transfer delay time independent of a packet length is calculated by subtracting the delay time component.