JP7622386B2 - Time synchronization method, time synchronization method - Google Patents

Description

The present invention relates to time synchronization between nodes in the communications field.

The Precision Time Protocol (PTP) defined in IEEE 1588 basically aims to synchronize time between two nodes, a master and a slave (see Patent Document 1 and Non-Patent Document 1).

Time synchronization using PTP is achieved through one-to-one processing between the master and slave, so while the advantage is that there are no topology constraints, as the number of slaves increases, traffic concentration on the master becomes an issue.

In addition, time synchronization using PTP is based on the premise that the transmission delays on the outbound and inbound routes between the master and slave are identical, but in a typical Ethernet (registered trademark) network, this premise is not guaranteed to hold true, making the quality of time synchronization prone to instability.

(1) Basic concept of time synchronization The master clock and slave clock in Figure 1 refer to the clocks built into the master and slave, and the deviation between the two clocks is "Δt." Since the master clock is used as the reference clock, the deviation "Δt" can be considered an "error" on the slave clock side.

For example, if the time of the master clock is "T" and the time of the slave clock is "T'", then it can be expressed as "T' = T + Δt". Furthermore, if "Δt > 0", the time of the slave clock is advanced, and if "Δt < 0", the time of the slave clock is delayed.

Therefore, "time synchronization" means "Δt = 0 (as close as possible to 0)." If the slave can calculate the "Δt" of its own clock, it can synchronize with the master clock by correcting the time advance/delay accordingly.

(2) Conventional Time Synchronization In the conventional time synchronization method shown in Fig. 2, a synchronization message "SYNC message" is transmitted from the master to the slave. This message is stamped with the transmission time T1.

The slave that receives the "SYNC message" measures the reception time "T'" of the message using its own slave clock. If the propagation time of the "SYNC message" is sufficiently short compared to the required accuracy, then "Δt" is calculated as shown in formula (1).

However, to stamp the sending time of a "SYNC message," a fairly advanced hardware (HW) is required. Therefore, in the case of a node that does not have the above hardware, the sending time is notified by a following "FOLLOW-UP message," as shown in Figure 3.

(3) Problem of Transmission Delay The above-mentioned conventional time synchronization is based on the premise that the propagation time of the "SYNC message" is sufficiently small compared to the required accuracy. Figure 4 shows a case where such a prerequisite is not met.

If the transmission delay between the master and slave in Figure 3 is "D1 ≠ 0", then equation (1) contains an extra transmission delay component D1, so the equality is established by subtracting the extra component (D1), as shown in equation (2).

In this case, the value of "D1" is required to calculate "Δt". The value of "D1" is "T'2 - T'1", so we get equation (3). In equation (3), the variable "Δt" remains on the right-hand side.

As a result, the correct value of "D1" is needed to calculate "Δt", but "Δt" is needed to calculate the value of "D1". This can be called the "chicken and egg" paradox. In other words, from a mathematical point of view, two equations are needed to determine the values of two variables, but only one equation can be formulated, resulting in a problem that cannot be solved.

(4) Solution of PTP In order to solve the above problem, PTP assumes that "the transmission delay is the same on the outbound and inbound routes" and executes the process shown in Fig. 5. Here, the following process is added to the conventional time synchronization method.
The slave transmits a "DELAY REQ message" to the master and measures the transmission time of this message, "T'3", using its own slave clock.
The master that receives the "DELAY REQ message" replies to the slave with the reception time "T4" as a "DELAY RES" message. The slave that receives this reply obtains the time "T4" of the same message.

As a result, it is possible to set up a new equation: "Δt = T4 - T'3 - D2". This results in "2 unknown variables" and "2 equations", and the deviation "Δt" is found based on the transmission delay "D1 = D2 = D".

To explain in detail how the deviation "Δt" is calculated, it is clear that "D1" and "D2" are as shown in equations (4) and (5).

Next, equations (6) and (7) hold based on the definition of the deviation "Δt" between the master clock and slave clock (T' = T + Δt).

Then, by substituting equations (6) and (7) into equations (4) and (5), equations (8) and (9) are obtained.

Here, since the transmission delay "D1 = D2 = D" is assumed, equation (10) holds.

If the value of equation (10) is "D", equations (8) and (9) can be expressed as equations (11) and (12).

When "Equation (12) - Equation (11)" is evaluated, "D" on the left side cancels out, and when solving for "Δt", we obtain Equation (13). The slave synchronizes its time with the master by correcting its own clock based on this deviation "Δt".

Patent Publication 2018-174445

"Endrum IEEE 1588 PTP Grandmaster Clock", [online], searched on September 11, 2020, Internet <URL: http://www.shoshin.co.jp/c/endrum/1588ptp.html>

As mentioned above, PTP calculates the deviation "Δt" based on the assumption that the transmission delay is "D1 = D2 = D." However, in Ethernet, it is not guaranteed, nor can it be expected, that the transmission delays on the outbound and inbound paths between nodes will be the same.

Fluctuations in transmission delays in Ethernet are mainly caused by the forwarding process of Ethernet switches. In other words, when a PTP packet passes through an Ethernet switch, it fluctuates depending on whether or not there is a packet being sent to the destination port.

If there is no packet being sent to the destination port, the packet will be forwarded without waiting to be sent. On the other hand, if there is a packet being sent to the destination port, subsequent packets will be forced to wait until the packet is completely sent. This is true even if a priority is set for the packet.

Therefore, the transmission delay varies depending on whether or not a preceding packet is present at the destination port (or the length of the preceding packet), and it cannot be expected that the transmission delay will be the same on the outbound and return journeys. Since PTP essentially cannot avoid the inclusion of such sudden errors (the presence or absence of a preceding packet), it is necessary to implement a smoothing mechanism such as a moving average. This smoothing function can smooth out the suddenly introduced errors between the previous and following measured values, but the trade-off is that the time required for time synchronization may increase.

The fundamental reason why PTP must assume that the transmission delays on the outbound and inbound routes are the same is the paradox mentioned above. In other words, there is a contradiction inherent in the fact that the time difference between a master clock and a slave clock must be measured in order to calculate the error between the two clocks.

The present invention was made to solve such problems in the past, and aims to stabilize the quality of time synchronization by measuring the transmission delay only with the master clock, rather than with the master and slave clocks, in a situation where it is not guaranteed that the round-trip transmission delay between nodes is the same.

(1) One aspect of the present invention is a method for time-synchronizing a group of slaves on a ring network with a master, comprising:
The synchronization message transmitted by the master is transferred and relayed by each of the slaves in sequence, and then received by the master, thereby circulating the message on the network;
Calculating a transmission delay of the synchronization message based on a transmission time and a reception time of the synchronization message measured by the master clock;
The clock error between the slave and the master is calculated using the transmission delay, and the slave clock is corrected based on the calculated error.

(2) Another aspect of the present invention is a method for time-synchronizing a group of slaves on a ring network with a master, comprising the steps of:
a step of circulating the synchronization message transmitted by the master on the network by each of the slaves sequentially forwarding and relaying the synchronization message and then receiving the synchronization message by the master;
calculating a transmission delay of the synchronization message based on a transmission time and a reception time of the synchronization message measured by an internal clock of the master;
using the transmission delay to calculate an error between the built-in clock of the slave and the built-in clock of the master, and correcting the built-in clock of the slave based on the calculated error to synchronize with the built-in clock of the master.

According to the present invention, in a situation where it is not guaranteed that the round-trip transmission delay between nodes is the same, the transmission delay is measured using only the master clock, not the master and slave clocks, thereby stabilizing the quality of time synchronization.

Schematic diagram of the time deviation between the master and slave. FIG. 1 is a schematic diagram showing conventional time synchronization. FIG. 13 is a schematic diagram showing a follow-up notification of a “FOLLOW-UP message.” Schematic diagram of a case where "transmission delay = 0" cannot be assumed. FIG. 1 is a schematic diagram showing a time synchronization process of PTP. FIG. 2 is an explanatory diagram showing the transmission of a "SYNC message" in the time synchronization method according to the embodiment of the present invention. An explanatory diagram showing the transmission of the "FOLLOW-UP message" of the same embodiment. Master block diagram. Slave block diagram. (a) is a diagram of the Ethernet frame that makes up the "SYNC message", "FOLLOW-UP message", and "PROBE message", (b) is a diagram of the Ethernet frame payload of the "SYNC message", (c) is a diagram of the Ethernet frame payload of the "FOLLOW-UP message", (d) is an explanatory diagram of "FLAGS" in (c), and (e) is a diagram of the Ethernet payload of the "PROBE message". State transition diagram for each mode. 11 is a graph showing measurement results of ring circumnavigation time.

The following describes a time synchronization method (method) according to an embodiment of the present invention. This time synchronization method (method) aims to avoid the root cause of the conventional problem, that is, "measuring transmission delay with two clocks (master clock and slave clock) with errors," by limiting it to a ring-type connection form (topology).

This provides more efficient and reliable time synchronization than existing technology (PTP), making it particularly suitable for use in Time Sensitive Networking (TSN).

Specifically, the time synchronization method is used on the circular transmission path of a ring network (ring type LAN, ring topology, system: hereafter abbreviated as ring), and a master node on the network transmits a "SYNC message" from one of its ring ports (two ports connected to the ring) to an adjacent slave.

Then, the slave that receives the "SYNC message" forwards it to the adjacent slaves in turn via hardware forwarding (HW forwarding), causing the "SYNC message" to circulate around the ring, where it is received by the other ring port of the master. This means that the transmission delay it takes to go around the ring can be measured using only the grandmaster clock of the master node, leaving no room for clock errors with the slave clock.

In addition, the round trip delay when there is no transmission delay variation is measured in advance, and "SYNC messages" that are delayed beyond the allowable range during operation are discarded to avoid incorrect time synchronization and reduce deterioration of synchronization accuracy.

<Network configuration>
The network configuration of the embodiment of the time synchronization method will be described with reference to Figures 6 and 7. In Figures 6 and 7, 10 indicates a ring, in which a master node (hereinafter abbreviated as master) 1 and slave nodes (hereinafter abbreviated as slaves) 1a to 1i are communicatively connected to a circular transmission path.

As shown in Figure 1, in this ring 10, the master 1 periodically transmits a "SYNC message" 7, which circulates around the ring 10. After the "SYNC message" 7 has circulated, the master 1 transmits a "FOLLOW-UP message" 8, which circulates around the ring 10. The circulation of these two messages 7 and 8 constitutes one synchronization processing unit.

The transmission of the "SYNC message" 7 and "FOLLOW-UP message" 8 by Master 1 is preferably bidirectional, but may be unidirectional. Furthermore, the end point of the loop is Master 1, and Master 1 configures the switching operation so that general traffic does not circulate infinitely.

This corresponds to what is called a "blocking port" in "STP," but since putting the ring port into a blocking state would also block the aforementioned messages 7 and 8, this is achieved using a different method.

<Master 1 and slaves 1a to 1i configuration>
The configuration of the master 1 and slaves 1a to 1i will be described with reference to Figures 8 and 9. The master 1 and slaves 1a to 1i each have a similar structure to a general PTP-compatible switching hub (switch).

That is, it is equipped with a CPU 2 that executes the synchronous ring protocol of ring 10, a switch engine 3 that transfers packets, a clock (internal clock) 4 to be synchronized, time stamp registers 5a and 5b that detect the sending and receiving of "SYNC messages" and record the time, and ring ports 6a and 6b consisting of Ethernet ports connected to ring 10.

However, the switch engine 3 of the master 1 applies existing mechanisms such as VLAN (Virtual LAN) and ACL (Access Control List) to prevent infinite circulation of general packets.

Note that, in PTP, the timestamp registers 5a and 5b also have a function for directly stamping the "SYNC message" 7, but in this embodiment, the "FOLLOW-UP message" 8 must be sent, so this direct stamping is not necessary.

<Frame composition>
An example of the configuration of the "SYNC message" 7 and the "FOLLOW-UP message" 8 required for the time synchronization process of the embodiment will be described with reference to FIGS.

Specifically, both messages 7 and 8 are composed of the Ethernet frame shown in FIG. 10(a). FIGS. 10(b) and (c) show the fields of the Ethernet payload "PAYLOAD" of both messages 7 and 8. Table 1 shows the message format of the above fields.

Furthermore, the flags shown in FIG. 7(d) are set in the "FLAGS" of the "FOLLOW-UP message" 8. Table 2 shows the details of the flags in FIG. 7(d).

<<Operation Processing>>
As shown in FIG. 11, the operation of the entire ring in this embodiment includes three modes: (1) calibration mode, (2) synchronization mode, and (3) normal operation mode.

That is, the ring characteristics are measured in calibration mode, time synchronization is achieved in synchronous mode based on the measurement results, and then the system transitions to normal operation mode to allow general traffic to pass while maintaining the time synchronization state.

To maintain time synchronization, the time synchronization process that was performed in synchronous mode continues even in normal operation mode. The only difference between the two is whether or not general traffic is allowed to flow onto the ring.

In normal operation mode, if it becomes difficult to maintain the quality of time synchronization due to the influence of general traffic, it is possible to temporarily fall back to synchronous mode to re-establish time synchronization, and then transition back to normal operation mode to resume the passage of general traffic.

[Calibration Mode]
In the calibration mode, the ring characteristics are measured after setting the ring so that general traffic does not flow into the ring.

Methods for preventing the inflow of general traffic include:
A variety of implementations are possible, such as blocking in an STP blocking state, isolating the ring port using a VLAN, filtering out general packets using an ACL, shutting down the general port, and unplugging the LAN cable from the general port.

In the calibration mode, the following two items are measured:
Measurement item 1: Ring lap time Tr
Measurement item 2: Slave position H
(1) Measurement item 1: Ring rotation time Tr
In the calibration mode, general traffic is completely removed from the ring 10. Therefore, it is reasonably expected that the transfer delay time in the switch will be shortest and stable. The measurement process steps (S01 to S06) for the ring round time Tr will be described below.
S01: Sending the "SYNC message" 7 The master 1 sends the "SYNC message" 7 from the ring port 6a to the adjacent slave 1a. Table 3 shows the contents written in the Ethernet payload field of the "SYNC message" 7. Here, "F" stands for "FLAGS".

When the "SYNC message" 7 passes through the ring port 6a of the master 1, the time of transmission is recorded by the time stamp register 5a.
In PTP, a function for directly stamping the transmission time on the "SYNC message" 7 may be used, but as described above, in this embodiment, the transmission of the "FOLLOW-UP message" 8 is essential, so the direct stamping function is not necessary. In other words, the master 1 retrieves the transmission time of the "SYNC message" 7 from the register 5a and records it. The time value recorded here is assumed to be "T1".

S02: Relay of the "SYNC message" 7 Slave 1a receives the "SYNC message" 7 sent by master 1, and relays it to neighbor slave 1b by hardware forwarding using switch engine 3. Slaves 1b to 1i repeat this relay in turn, circulating the message 7 on ring 10.

In other words, each slave 1a to 1i transmits the "SYNC message" 7 received on one ring port 6a from the opposite ring port 6b without processing it. This is a flooding operation for multicast frames, and can be executed by any general switch engine.

At this time, in the ring port 6a, the timestamp register 5a records the time when the "SYNC message" 7 passes. The times recorded here are designated "Ta-Ti" for each of the slaves 1a-1i. These "Ta-Ti" values are referenced when a "FOLLOW-UP message" 8, which will be described later, is received.

The "SYNC message" 7 is mirrored (replicated) to the CPU 2 of each slave 1a to 1i. A typical switch engine has the functionality to catch the reservation multicast packet at the CPU 2, so it is easy to achieve reception of this at the CPU 2.

When the "SYNC message" 7 is caught by CPU 2, each slave 1a to 1i refers to the "QUIET bit" and "QUIET TIMER" described in the message 7, and configures itself accordingly so that no traffic occurs on ring ports 6a and 6b. Note that not only does it stop the inflow of general traffic, but it also stops transmissions from its own CPU.

S03: Return of the "SYNC message" 7: The "SYNC message" 7 goes around the ring by the relay in S02, and is finally received by the opposing ring port 6b of the master 1. At this time, the timestamp register 5b records the reception time. The value of the time recorded here is "T2".

As a result, master 1 can calculate the ring rotation time Tr based on the transmission time T1 recorded by time stamp register 5a in S01 and the reception time T2 recorded by time stamp register 5b, as shown in equation (14).

Both "T1" and "T2" in formula (14) are values measured by the clock 4 of the master 1. Therefore, there is no room for calculation error due to deviation between the clocks of the master 1 and the slaves 1a to 1i in the ring orbit time Tr calculated by formula (14).

S04: Transmission of the "FOLLOW-UP message" 8 After calculating the ring rotation time Tr in S03, the master 1 transmits a "FOLLOW-UP message" 8 from the ring port 6a to the adjacent slave 1a.

Table 4 shows the contents written in the Ethernet payload field of the "FOLLOW-UP message" 8. Here, "F" stands for "FLAGS." Note that in calibration mode, clock 4 synchronization is not performed, so the flag "SYNC_FAIL" is always set to "1."

S05: Relay and return of the "FOLLOW-UP message" 8 Slave 1a receives the "FOLLOW-UP message" 8 in the same way as the "SYNC message" 7, catches it with its own CPU 2, and forwards and relays it to neighbor slave 1b. This operation process is repeated in sequence by slaves 1b to 1i, causing the message 8 to circulate around ring 10, and after the message has circulated, it is received by master 1.

At this time, each slave 1a to 1i refers to the "SYNC_FAIL" flag in the "FOLLOW-UP message" 8, detects that this is "1", and discards the reception time of the previously received "SYNC message" 7. Note that since time synchronization is not started in calibration mode, the reception time of the "SYNC message" 7 may be discarded regardless of the value of the "SYNC_FAIL" flag.

Then, if we logically expand the discard condition for the reception time of the "SYNC message" 7, we get "(SYNC_FAIL) || (MODE = CALIBRATION)." Here, "||" represents a logical sum.

S06: Repeated measurements In this embodiment, the ring rotation time Tr is measured by steps S01 to S05. Since the inflow of general traffic is blocked at this time, the measured value is expected to be correct, but to further improve accuracy, the measurement is repeated a predetermined number of times as shown in FIG.

In Figure 12, the shortest time "Tmin" is at the left base, and a distribution peak with an extremely small width is obtained.

The width of this peak, "σ", must be sufficiently small compared to the required synchronization accuracy. If this is not the case with a poor switching engine, the method of this embodiment cannot be realized.

The X-coordinate value of the peak of this mountain is defined as the standard ring orbit time "Tstd." The standard ring orbit time "Tstd" can be said to be the time it takes for a "SYNC message" 7 to go around the ring 10 once without waiting for transmission when it is transferred.

The calibration mode continues until the standard ring orbit time "Tstd" is determined. The number of measurements can be determined based on the quality of the switch engine, the scale of the topology, the network administrator's thinking, and past experience. Note that if the network configuration of ring 10 changes, the calibration will be redone.

(2) Measurement item 2: Slave position H
In the calibration mode, in addition to measuring the standard ring orbit time "Tstd", the position of each of the slaves 1a to 1i on the ring 10 is also measured. Here, the position of each of the slaves 1a to 1i is the number of hops that elapse from the master 1 to each of the slaves 1a to 1i.

When the network configurations illustrated in FIGS. 6 and 7 are evaluated counterclockwise,
・Slave 1a = 1 hop (1st from Master 1)
・Slave 1b = 2 hops (second unit)
・Slave 1c = 3 hops (third unit)
・Slave 1d = 4 hops (4th unit)
・Slave 1e = 5 hops (5th unit)
・Slave 1f = 6 hops (sixth unit)
・Slave 1g = 7 hops (7th unit)
・Slave 1h = 8 hops (8th unit)
・Slave 1i = 9 hops (9th unit)
On the other hand, if you evaluate it clockwise,
・Slave 1i = 1 hop (from Master 1 to the 1st unit)
・Slave 1h = 2 hops (second unit)
・Slave 1g = 3 hops (3rd unit)
・Slave 1f = 4 hops (4th unit)
・Slave 1e = 5 hops (5th unit)
・Slave 1d = 6 hops (sixth unit)
・Slave 1c = 7 hops (7th unit)
・Slave 1b = 8 hops (8th unit)
・Slave 1a = 9 hops (9th unit)
It becomes.

Each slave 1a to 1i needs to know its own location (number of hops). Also, the master 1 does not need to know the locations of the slaves 1a to 1i, but in order to allow each slave 1a to 1i to detect its own location, it transmits a "PROBE message" 9 shown in Figure 10 (a) (e) from one of the ring ports 6a and 6b.

Table 5 shows the contents of the Ethernet payload fields of the "PROBE message" 9.

The "PROBE message" 9 differs from other messages in that the destination Ethernet address is "01:80:C2:00:00:00". This is a frame called a "BPDU", and is not flooded but is only received by CPU 2. In other words, it is transferred in a bucket relay by CPU 2 of each of slaves 1a to 1i on ring 10.

The slaves 1a to 1i that receive the "PROBE message" 9 record the value in the "NODES" field in Fig. 10(e) plus 1 as their own position (number of hops from the master 1). The value recorded here is assumed to be "H".
That is, the slaves 1a to 1i overwrite the "NODES" field with the hop count H and transfer the packet to the neighbors. As a result, the value of the "NODES" field is incremented by 1 each time the packet passes through each of the slaves 1a to 1i.
By repeating this bucket brigade, all slaves 1a to 1i can grasp their own positions. This measurement is performed for both the clockwise and counterclockwise directions of the ring 10. However, if the messages 7 and 8 are limited to one direction, it is sufficient to perform the measurement only in that direction.
The "PROBE message" 9 is ultimately received and returned to the other of the ring ports 6a, 6b of the master 1. At this time, the value of the "NODES" field written in the message 9 becomes the total number of slaves 1a to 1i on the ring 10, which is designated as "N."
The master 1 writes the value of "N" in the "NODES" field of each of the messages 7 and 8 and advertises it to the entire ring 10. After measuring "N", the master 1 and slaves 1a to 1i perform the following verification. Note that "N'" and "H'" shown below indicate the values of "N" and "H" counterclockwise, respectively.
(Master 1 checks the calculations)
Master 1 checks whether the number of nodes "N" and "N'" match in the clockwise and counterclockwise directions, that is, whether equation (15) holds.

(Verification of slaves 1a to 1i)
The slaves 1a to 1i check whether the sum of the clockwise and counterclockwise positions H, H' coincides with the number of nodes N, that is, whether the formula (16) holds.

If the above calculation results in an error, i.e., if equations (15) and (16) do not hold, the position measurement of slaves 1a to 1i will be redone.

[Synchronous mode]
(1) When the standard link circulation time "Tstd" and the number of nodes "N" are determined in the calibration mode, a transition to the synchronous mode is made (DONE transition in FIG. 11).

Only the master 1 has the authority to change the operation mode. That is, the master 1 writes its own decision in the “MODE” field of each of the messages 7 and 8 and announces it to the entire ring 10.
In the synchronous mode, similarly to the calibration mode, the processes using the two messages 7 and 8 are periodically executed. The frequency of execution is set according to the required accuracy, the scale of the ring, or the idea of the network administrator.
However, the synchronous mode differs from the calibration mode in the method of determining the "SYNC_FAIL" flags in both messages 7 and 8. That is, in the calibration mode, the "SYNC_FAIL" flag is always set to "1." In contrast, in the synchronous mode, the value of the "SYNC_FAIL" flag is adjusted according to the following evaluation.
(A) Periodically execute the processes using both messages 7 and 8, and calculate the ring circulation time "Tr" each time using equation (14).
(B) Calculate the excess delay "Trx" of each ring round time "Tr" from equation (17).

The value of the excess delay "Trx" is the total time that the "SYNC message" 7 is forced to wait due to the influence of a preceding frame being transmitted when it is forwarded by the switch.
(C) If the excess delay time "Trx" becomes a negative value, the standard ring rotation time "Tstd" obtained in the calibration mode is considered to be inaccurate, so fall back to the calibration mode ((BREAK transition in FIG. 11).
(5) When the excess delay time "Trx" becomes a positive value, it is compared with a cutoff line setting determined based on the required accuracy. As a result of this comparison, the "SYNC_FAIL" flag is determined as follows:
If the result is below the cutoff line: the "SYNC_FAIL" flag is set to "0".
If the result is above the cutoff line: the "SYNC_FAIL" flag is set to "1".

This makes it possible to reliably detect whether an unacceptable level of delay occurred anywhere on the ring 10 as the most recent "SYNC message" 7 circulated around the ring 10.

In addition, by advertising the detection results on the ring 10, it is possible to discard the reception time of the most recent "SYNC message" 7, thereby preventing deterioration of synchronization quality. Except for this adjustment of the "SYNC_FAIL" flag, the master 1 executes the same processing as in the calibration mode.

(2) If the "SYNC_FAIL" flag is "1", the slaves 1a to 1i execute the same operation process as in the calibration mode. On the other hand, if the "SYNC_FAIL" flag is "0", the slaves 1a to 1i execute the following operation process.

That is, when a slave 1a to 1i receives a "FOLLOW-UP message" 7 with a "SYNC_FAIL" flag of "0", it calculates the error (deviation) "Δt" of its own clock 4.

Here, the "SYNC_TIMESTAMP" written in the "FOLLOW-UP message" 7 is "T1", and the reception time of the most recent "SYNC message" 7 measured by each of the slaves 1a to 1i using its own clock 4 is "Ts". In this case, the error "Δt" can be expressed by equation (18).

"D _S " in formula (18) indicates the time it takes for the "SYNC message" 7 to propagate from the master 1 to the slaves 1a to 1i. The evaluation formula for formula (18) at this stage is the same as that for PTP, but the method for calculating "D _S " differs.

To explain the calculation method and basis of "D _S " in this embodiment, first, the round trip time "Tr" that it took for the most recent "SYNC message" 7 to go around the ring 10 once is recorded in the "LAP_DELAY" field of the "FOLLOW-UP message" 8. This round trip time "Tr" can be said to be a value obtained by adding the excess delay "Trx" to the standard ring round trip time "Tstd", as shown in equation (19).

Next, as mentioned above, the excess delay time "Trx" is measured with high accuracy by master 1. Also, when the "SYNC_FAIL" flag is "0", it is confirmed that the excess delay time "Trx" is sufficiently small compared to the required accuracy.

In this case, from the perspective of the required order of accuracy, the excess delay time "Trx" approximates "0" ("Trx ≒ 0"), and equation (20) holds.

Furthermore, the standard link round trip time "Tstd" is calculated as the time required when the delay of the switch engine 3 is zero and unchanged in the calibration mode. Therefore, the delay "Dhop" that occurs per hop can be estimated to be equal for each slave 1a to 1i, and is approximated as shown in equation (21).

In equation (21), "N" indicates the total number of slaves 1a to 1i on ring 10. The value of "N" is written in the "NODES" field of both messages 7 and 8, so it can always be read.

At this time, the slaves 1a to 1i know the number of hops H from the master 1 to themselves from the "PROBE message" 9. Using the value of this number of hops H, the time "D _S " that has elapsed until the "SYNC message" 7 arrives at the slaves 1a to 1i from the master 1 can be calculated as shown in equation (22).

Then, equation (23) can be obtained from equations (18) and (22).

From the above, the error "Δt" can be calculated using formula (24).

As a result, slaves 1a to 1i can synchronize the time with master 1 by subtracting the error "Δt" from their own clock 4, as shown in equation (25).

In PTP, in order to mitigate the effects of sudden errors, it is necessary to implement some sort of smoothing mechanism, such as a moving average. In contrast, in this embodiment, the error "Δt" that has been mixed in is accurately measured by Master 1, and a synchronization process in which a delay that exceeds the required accuracy has occurred can be reliably discarded when the "SYNC_FAIL" flag is "1". Therefore, according to this embodiment, a smoothing mechanism is not required, and a nearly perfect time synchronization state can be obtained in a single synchronization process.

However, in order to obtain a reliable time synchronization state, it is preferable to maintain the synchronous mode state until the conditions set by the network administrator are met. For example, examples of conditions for transitioning to the normal operation mode are as follows:
- The synchronization process is successful three times in a row. - It is preferable to set a safety standard such as repeating the synchronization process for about three minutes and making sure that the number of times the "SYNC_FAIL" flag is "1" is less than a preset value.

This safety standard can be determined based on the quality of the switch engine 3, the scale of the ring 10, and the network administrator's perspective. If this safety standard is met, the system transitions to normal operation mode (SYNCED transition in Figure 11).

<Normal operation mode>
When time synchronization is established in the synchronous mode, the above-mentioned SYNCED transition causes a transition to the normal operation mode.

Even in normal operation mode, the master 1 and slaves 1a to 1i periodically execute the processing using both messages 7 and 8. The frequency of execution can be determined according to the required accuracy, the scale of the ring, and the network administrator's ideas.

Master 1 sets an integer value representing the normal operation mode in the "MODE" field of both messages 7 and 8 that it sends, and announces to the entire ring 10 that it has switched to the normal operation mode.

At this time, the master 1 and the slaves 1a to 1i that have detected the transition to normal operation mode change the settings of the switch engine 3 to allow general traffic to pass through the ring 10. This establishes a system that maintains time synchronization while sharing the line with general traffic.

<<Fallback operation mode>>
The three operation modes (calibration mode, synchronization mode, normal operation mode) are the same in that the processing using both messages 7 and 8 is performed continuously, but differ in the implementation of the synchronization process/permission of general traffic.

Normally, the state transition is one-way in the order of "calibration mode → synchronization mode → normal operation mode". However, a fallback transition to the calibration mode due to a calibration failure/time synchronization failure is also assumed. The transition conditions to return to this calibration (BREAK in FIG. 11) are as follows:
(1) When the network configuration has changed. (2) When the excess delay "Trx" of the ring rotation time Tr is found to be a negative value. The transition condition (RESYNC) from the normal operation mode back to the synchronization mode is assumed to be when a yellow light turns on to maintain the synchronization quality, as described below.

<Observation of time synchronization quality>
In this embodiment, we propose a measure to monitor the quality of time synchronization. It is preferable that the network administrator defines the fallback conditions to the synchronous mode based on the results of this monitoring.

(1) Consecutive Failures If the processes using the two messages 7 and 8, which are periodically performed, fail consecutively, the system is caused to fall back to the synchronous mode according to the number of failures.

For example, let us assume that the network administrator has set this to three times. Master 1 fails to process both messages 7 and 8, and sets "SYNC_FAIL" to "1." In this case, it adds "1" to failure counter C. On the other hand, if the processing of both messages 7 and 8 is successful, it resets failure counter C to "0."

Each time the process is performed using both messages 7 and 8, it checks whether "failure counter C ≥ 3" is true, and if so, transitions to synchronous mode. At this time, the "MODES" fields of both messages 7 and 8 sent by master 1 from the next time onwards are set to synchronous mode, and this is advertised to the entire ring 10.

(2) Continuous Failure Time If the elapsed time from the last successful processing of both the messages 7 and 8 to the present time exceeds a preset threshold, the mode is caused to fall back to the synchronous mode.

(3) Error Correction Amount As described above, the slaves 1a to 1i calculate and correct the error "Δt" each time by processing the two messages 7 and 8. However, if the error "Δt" reaches a value that suggests a synchronization failure, the slaves 1a to 1i request the master 1 to transition to the synchronous mode.

The criteria for determining poor synchronization are as follows:
- When the error "Δt" exceeds a preset threshold value - When the value of the error "Δt" has changed by a percentage (%) that is greater than or equal to the threshold value compared to the previous time - When the error "Δt" continues to have a value of "0" etc. are possible cases.

Things like these are possible.

Network administrators can configure fallback conditions by combining these criteria, either individually or in combination, to stabilize the quality of time synchronization.

<Action and effect>
According to this embodiment, the following advantageous effects can be obtained.

(1) Reduction of the load on the CPU 2 and the network bandwidth by simplifying the procedure In the existing technology (PTP), two round trips of messages were required between the master 1 and the slaves 1a to 1i. In contrast, in this embodiment, a one-way message transmission from the master 1 is sufficient, which can reduce the load on the CPU 2 of the master 1 and the network bandwidth.

For example, in the network configurations of FIGS. 6 and 7, there are nine slaves 1a to 1i, so according to the prior art, the processing units of the master 1 that occur in one synchronization process are as follows (one message is one unit):
Sending one "SYNC message" 7 (flooding the entire network via multicast)
Receives 9 "DELAY_REQ messages" from slaves 1a to 1i. Returns 9 "DELAY_RES messages" to slaves 1a to 1i. A total of 19 units (20 units if "FOLLOW-UP messages" 8 are sent).
is required.

In contrast, in this embodiment,
Sending one "SYNC message" 7 (flooding the entire ring 10 by multicast)
Sending one "FOLLOW-UP message" 8 (flooding the entire ring 10 by multicast)
・Only 2 units in total are required.

In particular, in conventional technology, the processing unit of master 1 increases in proportion to the number of slaves. In contrast, in this embodiment, two units are always sufficient, regardless of the number of slaves.

(2) Stabilization of the quality of time synchronization In the existing technology, the synchronization accuracy is easily affected by fluctuations in each synchronization process. In contrast, according to the present embodiment, the ring orbit time Tr can be measured with high accuracy, so that a synchronization process with a large error "Δt" can be reliably discarded, and the quality of time synchronization can be easily maintained. In other words, since the processing structure is simple, there is no large fluctuation in the synchronization accuracy, and in this respect, the accuracy of time synchronization can be stabilized.

(3) Synchronization in a short time In the existing technology, the introduction of a smoothing mechanism increases the time required to achieve a synchronized state. In contrast, in the present embodiment, in theory, synchronization can be successfully achieved in one try, making it possible to minimize the time required to achieve a synchronized state.

1…Masternode (Master)
1a to 1i: slave nodes (slave)
2. CPU
3... Switch engine 4... Clock (built-in clock)
5a, 5b...Time stamp register 6a, 6b...Ring port 7...SYNC message (synchronization message)
8...FOLLOW-UP message (follow-up notification)
9... PROBE message 10... Ring network

Claims (7)

By executing a calibration mode for measuring the characteristics of the ring network and a synchronization mode based on the results of the measurement,
A method for time-synchronizing a group of slaves on the ring network with a master, comprising:
In the calibration mode,
Under a setting where general traffic does not flow in, a synchronization message transmitted by the master is sequentially forwarded and relayed by each of the slaves, and then the message is received by the master . This ring circulates multiple times,
the master measures a ring orbit time for each of the rings, and calculates a standard ring orbit time based on the measured ring orbit times and a minimum time according to a required accuracy;
In the synchronous mode,
The master periodically circulates the synchronization message in the same manner as in the calibration mode,
Calculating an excess delay time as a difference between a ring rotation time of the periodically circulated synchronization message and the standard ring rotation time;
If the excess delay time exceeds a preset tolerance level, the excess delay time is notified to each of the slaves.
Each of the slaves calculates a transmission delay of the synchronization message based on a transmission time and a reception time of the synchronization message measured by a clock of the master in response to the advertisement ;
a clock error between the slave and the master clock is calculated using the transmission delay, and the slave clock is corrected based on the calculated error. In the calibration mode, the master calculates the number of hops from the master to each of the slaves,
In the synchronous mode, a transit delay time per slave is estimated using the ring circulation time;
2. The time synchronization method according to claim 1 , wherein the transmission delay to each slave is calculated by multiplying the result of the estimation by the number of hops. In the synchronous mode, the master
By the advertisement,
3. The time synchronization system according to claim 1, wherein the reception time of the synchronization message in which the excessive delay time occurs is discarded. In the synchronization mode, the master notifies the slave of the circulation time of the synchronization message in a trailing manner;
3. The time synchronization system according to claim 2 , wherein a message is transmitted to inform the slave of the number of hops. 5. The time synchronization method according to claim 1, further comprising: transitioning to a normal mode in which general traffic can flow into the ring network after completion of time synchronization in the synchronous mode. 6. The time synchronization method according to claim 5 , wherein a fallback transition to the calibration mode or the synchronization mode is possible after the transition to the normal mode. By executing a calibration mode for measuring the characteristics of the ring network and a synchronization mode based on the results of the measurement,
A method for time-synchronizing a group of slaves on the ring network with a master, comprising the steps of:
The calibration mode includes:
A step of performing a ring rotation a plurality of times in which the synchronization message transmitted by the master is sequentially forwarded and relayed by each of the slaves under a setting that does not allow general traffic to flow, and the synchronization message is then received by the master;
a step of the master measuring a ring orbit time for each of the rings and calculating a shortest time according to a required accuracy from a group of the measured ring orbit times as a standard ring orbit time;
having
The synchronous mode is
the master periodically circulates the synchronization message in the same manner as in the calibration mode;
calculating a difference between a ring rotation time of the periodically circulated synchronization message and the standard ring rotation time as an excess delay time;
notifying each of the slaves if the excess delay time exceeds a preset tolerance level;
each of the slaves calculates a transmission delay of the synchronization message based on a transmission time and a reception time of the synchronization message measured by a clock of the master in response to the advertisement ;
calculating an error between the clock of the slave and the clock of the master using the transmission delay, and correcting the clock of the slave based on the calculated error;
A time synchronization method comprising:

Images