A time-sensitive network incremental routing scheduling method, system, device and storage medium

Abstract

The present invention discloses a time-sensitive network incremental routing scheduling method, system, device and storage medium. The method comprises: constructing a time-sensitive network (TSN) multi-routing transmission model using cyclic queue forwarding (CQF) as a transmission mode; the model includes network topology and flow characteristics; setting constraints for the model, and converting a traffic routing scheduling problem into an optimization problem based on the constraints; using a greedy strategy to process the priority of traffic processing, using the Dijkstra algorithm to obtain the optimal path for the selected traffic, and determining the scheduling process of time slot offset selection based on a load balancing method to complete the scheduling. The present invention has a higher scheduling success rate under the conditions of complex network models and traffic compositions.

Description

Technical Field

[0001] The present invention relates to an incremental routing scheduling method, system, device and storage medium, and in particular to a time-sensitive network incremental routing scheduling method, system, device and storage medium, belonging to the field of electronic information technology. Background Art

[0002] In today's industrial landscape, the demand for real-time and reliability is increasing dramatically across many scenarios. For example, applications such as automotive control systems, avionics, industrial control, and smart grids all require the transmission of time-sensitive services that require both real-time and reliability. Traditional networking methods can only reduce the usage of non-time-sensitive services by increasing bandwidth, thereby enabling the transmission of time-sensitive services at a high cost. However, increasing bandwidth does not address the challenges posed by transmitting time-sensitive services and wastes valuable bandwidth resources.

[0003] To solve these problems, the IEEE 802.1 Time-Sensitive Networking (TSN) working group has developed a set of standards for TSN. The TSN standard is based on traditional Ethernet, acts on the data link layer, and inserts 4 bytes into the standard Ethernet data packet to define its characteristics. It can support real-time applications with limited end-to-end latency and zero packet loss capabilities. In TSN, messages are exchanged as data packet streams, transmitted from one source to one or more destinations, with specific paths, sizes, and priorities. TSN uses service levels to classify different types of flows while ensuring that time-sensitive flows and BE flows are transmitted on the same network. Cyclic Queuing Forwarding (CQF) is a scheduling mechanism in time-sensitive networks. In CQF networks, a periodic timer and two queues are maintained on each switch output port, and the two queues are converted according to timing. The specific operations are as follows: Figure 2 As shown in the figure: In odd-numbered time slots, the receive gate of queue Q1 is open and the transmit gate is closed. At this time, Q1 is in the data receiving state. At the same time, the receive gate of queue Q2 is closed and the transmit gate is open, and Q2 is in the data forwarding state. In even-numbered time slots, the receive gate of queue Q1 is closed and the transmit gate is open, and Q1 enters the data forwarding state. The receive gate of queue Q2 is open and the transmit gate is closed, and Q2 enters the data receiving state. Within the same time slot, the states of queues Q1 and Q2 are different and transition according to the time slot cycle. This is the basic method of CQF data transmission.

[0004] Due to TSN's characteristics, such as time synchronization and periodicity, global control of network transmission can be achieved through a single terminal, requiring relevant algorithms to handle this. Current algorithms lack the ability to leverage the correlations and characteristics of flows, and may not be sufficient to support the processing of large amounts of data in complex traffic scheduling environments. Summary of the Invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a time-sensitive network incremental routing scheduling method, system, device and storage medium that can improve the scheduling success rate under complex network models and traffic composition.

[0006] Technical solution: The present invention provides a time-sensitive network incremental routing scheduling method, comprising:

[0007] (1) Constructing a multi-routing transmission model for a time-sensitive network (TSN) using cyclic queue forwarding (CQF) as the transmission mode; the model includes network topology and flow characteristics;

[0008] (2) Set the constraints of the model and transform the traffic routing scheduling problem into an optimization problem based on the constraints;

[0009] (3) A greedy strategy is used to prioritize traffic flow. For the selected traffic, the Dijkstra algorithm is used to obtain the optimal path for the traffic. The scheduling process of the time slot offset selection is determined based on the load balancing method to complete the scheduling.

[0010] Furthermore, the network topology is represented as a directed graph G = {V, E}, where V is a set of vertices S of hosts and vertices H of switches, and E represents directed connections between vertices connected to each other for transmission; the hosts are the source and destination of data flows, and are used to generate time-triggered data flows and control the sending time of the data flows;

[0011] The flow is characterized by the following equation,

[0012] f i ={src,dst,period,size,deadline,path,offset,frequency,score}

[0013] Among them, f i represents the i-th flow, where the source node src, destination node dst, period period, packet size size, deadline deadline, and path path are the different features observed during the flow generation phase. The offset offset represents the value used to adjust the scheduling process of each flow, and the frequency frequency represents the number of times a packet appears within the specified scheduling interval. The score score is calculated based on the known feature values.

[0014] The path is represented by a link, as shown in the following formula:

[0015] f i .path={L(S1,S2),L(S2,S3),...,L(Sn-1 ,S n )}

[0016] Among them, f i .path is stream f i The paths L(S1,S2) and L(S n-1 ,S n ) are the first and last links in the TSN network, S1 and S n They are f i The first and last vertices of .

[0017] Furthermore, the step (2) specifically includes:

[0018] (21) Set the scheduling period constraint. Specifically, define a time length that is the least common multiple of the period F.periods of the flow set as the scheduling period sched cycle , mapping all flows in a flow collection:

[0019] sched cycle =LCM(F.periods).

[0020] Where F is the set of flows, LCM represents the least common multiple; based on a given scheduling period, calculate f i Frequency f i .frequency, as shown below:

[0021]

[0022] (22) Set a time slot constraint. Specifically, the time slot is the granularity of the time offset, which is divisible by the period of all flows and is expressed by the greatest common factor of the period of all flows. The lower bound of the time slot is the time required for a queue to send and receive:

[0023]

[0024] Among them, slot cycle Indicates time slot, Represents a positive number, GCD represents the greatest common factor, Queue size is the measure of queue size, MTU is the maximum transmission unit, B is the bandwidth, d proc is the sum of the maximum delay caused by switch transmission, queuing and processing, d sync is the maximum delay for time synchronization;

[0025] (23) Set an offset constraint. Specifically, make the offset of the flow smaller than the period of the flow:

[0026]

[0027] Among them, f i .offset is f i The offset of , N is the total number of streams;

[0028] (24) Set the deadline constraint, f i The deadline f i .deadline, which indicates the latest time the flow needs to be sent to the target vertex. The specific settings are as follows:

[0029]

[0030] (f i .offset+|f i .path|+2)×slot cycle ≤f i .deadline.

[0031] Among them, (|f i .path|+2) indicates the number of time slots required during the stream transmission process;

[0032] (25) Set the queue resource constraints. Specifically, let the queue resource block Q represent the switch j, T (t) Time slot queue resource at port k, set mapping value Indicates whether the corresponding flow occupies the resource block. When the flow occupies the resource block, 1 if yes, 0 otherwise:

[0033]

[0034] Where hop represents the path length, mod represents the modulo operation, and μ represents the μth packet of the flow in the scheduling period, which is used to calculate the slot position occupied by the flow;

[0035] For a set of flows that completes the scheduling step, the occupied resources satisfy the following relationship to prevent resource overflow:

[0036]

[0037] Where Φ(i) represents f i Whether it is successfully scheduled, when f i When the scheduling is successful, the value of Φ(i) is 1, otherwise it is 0;

[0038] (26) Let maximizing the number of successfully scheduled flows be the scheduling optimization objective:

[0039]

[0040] Among them, F success Indicates the number of successfully scheduled flows.

[0041] Furthermore, the step (3) specifically includes:

[0042] (31) Perform traffic processing selection based on a greedy strategy; specifically, in the traffic feature tuple, select size, period, and frequency to perform a greedy selection strategy;

[0043] Perform Min-Max normalization on the three sets of data, then perform weighted sum on the obtained data and define the result as a score:

[0044] f i .score = α × f i .size new +β×f i .frequency new +γ×f i .deadline new .

[0045] Where α, β, and γ are weighting factors of packet size, frequency, and deadline, respectively. They are all positive and sum to 1.

[0046] Sort the traffic subsets in ascending order according to the size of the scores, and maximize the number of successfully scheduled flows by adjusting the weighting factor values;

[0047] (32) Perform incremental routing; specifically, establish a resource graph G in the form of a bidirectional link based on the directed graph G. * , the resource legend G * It only contains switch endpoints and interconnected links; initialize the distance on each link to 1, use Dijkstra algorithm to solve the shortest path, and update the distance of the corresponding link in the directed graph according to the obtained optimal path. When f i The shortest path contains the link L(S a ,S b ), the distance of the graph is updated as follows:

[0048] Dis(L(S a ,S b ))+=f i .size new *f i .frequency new

[0049] For the traffic scheduled later, the route selection operation is performed according to the updated network resources;

[0050] (33) Select an offset; specifically, traverse all offsets, exclude offsets that exceed the remaining resources, calculate the maximum resource occupancy of all paths among the remaining feasible offsets, and finally select the offset with the smallest maximum resource occupancy for data mapping;

[0051] Specifically, let the resource block be represented as For observing the time slot T (t) The resource occupancy of the directed link from switch j to switch k is assuming that a certain number of flows are successfully mapped to the resource space and the capacity occupied in the resource block is Expressed as:

[0052]

[0053] f i The scheduling resource increment caused by the corresponding offset is added to the existing resource occupancy In [1], we substitute the resource constraints for judgment and filter out the offsets that meet the resource constraints. For each offset that meets the conditions, we traverse all path nodes in the path, find the resource occupancy of the node with the largest space in the path, and select the offset that minimizes this value to perform resource mapping and complete the scheduling step. The formula is as follows:

[0054]

[0055] Based on the same inventive concept, the present invention also provides a time-sensitive network incremental routing scheduling system, including:

[0056] Model building module; used to build a time-sensitive network TSN multi-routing transmission model using cyclic queue forwarding (CQF) as the transmission mode; the model includes network topology and flow characteristics;

[0057] The conversion module is used to set the constraints of the model and convert the traffic routing scheduling problem into an optimization problem based on the constraints;

[0058] The scheduling module is used to process the priority of traffic processing using a greedy strategy. For the selected traffic, the Dijkstra algorithm is used to obtain the optimal path of the traffic, and the scheduling processing of the time slot offset selection is determined according to the load balancing method to complete the scheduling.

[0059] Furthermore, in the model building module, the network topology is represented as a directed graph G = {V, E}, where V is a set of vertices S, which are hosts, and H, which are switches, and E represents directed connections between vertices that are interconnected for transmission. The hosts are the source and destination of data flows, and are used to generate time-triggered data flows and control the time at which data flows are sent.

[0060] The flow is characterized by the following equation,

[0061] f i ={src,dst,period,size,deadline,path,offset,frequency,score}

[0062] Among them, f i represents the i-th flow, where the source node src, destination node dst, period period, packet size size, deadline deadline, and path path are the different features observed during the flow generation phase. The offset offset represents the value used to adjust the scheduling process of each flow, and the frequency frequency represents the number of times a packet appears within the specified scheduling interval. The score score is calculated based on the known feature values.

[0063] The path is represented by a link, as shown in the following formula:

[0064] f i .path={L(S1,S2),L(S2,S3),...,L(S n-1 ,S n )}

[0065] Among them, f i .path is stream f i The paths L(S1,S2) and L(S n-1 ,S n ) are the first and last links in the TSN network, S1 and S n They are f i The first and last vertices of .

[0066] Furthermore, the conversion module is specifically implemented as follows:

[0067] Set the scheduling period constraint. Specifically, define a time length that is the least common multiple of the period F.periods of the flow set as the scheduling period sched cycle , mapping all flows in a flow collection:

[0068] sched cycle =LCM(F.periods).

[0069] Where F is the set of flows, LCM represents the least common multiple; based on a given scheduling period, calculate f i Frequency f i .frequency, as shown below:

[0070]

[0071] Set the time slot constraint. Specifically, the time slot is the granularity of the time offset, which is divisible by the period of all flows and is expressed as the greatest common factor of the period of all flows. The lower bound of the time slot is the time required for a queue to send and receive:

[0072]

[0073] Among them, slot cycle Indicates time slot, Represents a positive number, GCD represents the greatest common factor, Queue size is the measure of queue size, MTU is the maximum transmission unit, B is the bandwidth, d proc is the sum of the maximum delay caused by switch transmission, queuing and processing, d sync is the maximum delay for time synchronization;

[0074] Set the offset constraint. Specifically, make the offset of the stream smaller than the period of the stream:

[0075]

[0076] Among them, f i .offset is f i The offset of , N is the total number of streams;

[0077] Set the deadline constraint, f i The deadline f i .deadline, which indicates the latest time the flow needs to be sent to the target vertex. The specific settings are as follows:

[0078]

[0079] (f i .offset+|f i .path|+2)×slot cycle ≤f i .deadline.

[0080] Among them, (|f i .path|+2) indicates the number of time slots required during the stream transmission process;

[0081] Set the queue resource constraints. Specifically, let the queue resource block Q represent the switch j T (t) The time slot is the queue resource at port k, and the mapping value is set Indicates whether the corresponding flow occupies the resource block. When the flow occupies the resource block, 1 if yes, 0 otherwise:

[0082]

[0083] Where hop represents the path length, mod represents the modulo operation, and μ represents the μth packet of the flow in the scheduling period, which is used to calculate the slot position occupied by the flow;

[0084] For a set of flows that completes the scheduling step, the occupied resources satisfy the following relationship to prevent resource overflow:

[0085]

[0086] Where Φ(i) represents f i Whether it is successfully scheduled, when f i When the scheduling is successful, the value of Φ(i) is 1, otherwise it is 0;

[0087] Let maximizing the number of successfully scheduled flows be the scheduling optimization objective:

[0088]

[0089] Among them, F success Indicates the number of successfully scheduled flows.

[0090] Furthermore, the scheduling module is specifically implemented as follows:

[0091] Perform traffic processing selection based on a greedy strategy. Specifically, in the traffic feature tuple, select size, period, and frequency to perform a greedy selection strategy.

[0092] Perform Min-Max normalization on the three sets of data, then perform weighted sum on the obtained data and define the result as a score:

[0093] f i .score = α × f i .size new +β×f i .frequency new +γ×f i .deadline new .

[0094] Where α, β, and γ are weighting factors of packet size, frequency, and deadline, respectively. They are all positive and sum to 1.

[0095] Sort the traffic subsets in ascending order according to the size of the scores, and maximize the number of successfully scheduled flows by adjusting the weighting factor values;

[0096] Perform incremental routing; specifically, establish a resource graph G in the form of a bidirectional link based on the directed graph G * , the resource legend G* It only contains switch endpoints and interconnected links; initialize the distance on each link to 1, use Dijkstra algorithm to solve the shortest path, and update the distance of the corresponding link in the directed graph according to the obtained optimal path. When f i The shortest path contains the link L(S a ,S b ), the distance of the graph is updated as follows:

[0097] Dis(L(S a ,S b ))+=f i .size new *f i .frequency new

[0098] For the traffic scheduled later, the route selection operation is performed according to the updated network resources;

[0099] Select an offset. Specifically, traverse all offsets, exclude offsets that exceed the remaining resources, calculate the maximum resource usage of all paths among the remaining feasible offsets, and finally select the offset with the minimum maximum resource usage for data mapping.

[0100] Specifically, let the resource block be represented as For observing the time slot T (t) The resource occupancy of the directed link from switch j to switch k is assuming that a certain number of flows are successfully mapped to the resource space and the capacity occupied in the resource block is Expressed as:

[0101]

[0102] f i The scheduling resource increment caused by the corresponding offset is added to the existing resource occupancy In [1], we substitute the resource constraints for judgment and filter out the offsets that meet the resource constraints. For each offset that meets the conditions, we traverse all path nodes in the path, find the resource occupancy of the node with the largest space in the path, and select the offset that minimizes this value to perform resource mapping and complete the scheduling step. The formula is as follows:

[0103]

[0104] Based on the same inventive concept, the present invention also provides a computing device, comprising: one or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the time-sensitive network incremental routing scheduling method according to any one of the above items are implemented.

[0105] Based on the same inventive concept, the present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the time-sensitive network incremental routing scheduling method according to any one of the above items.

[0106] Beneficial effects: Compared with the existing technology, the present invention has the following significant advantages: 1. It improves the scheduling success rate under complex network models and traffic compositions; 2. Due to the incremental routing scheme, there is no significant difference in time consumption in the solution speed compared with the shortest path algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0108] Figure 2 Schematic diagram of CQF transmission model in the prior art;

[0109] Figure 3 This is a schematic diagram of a TSN routing scheduling topology scenario according to an embodiment of the present invention;

[0110] Figure 4 This is a comparison chart of scheduling success rates of different algorithms in the Net_A network topology according to an embodiment of the present invention;

[0111] Figure 5 A comparison chart of scheduling success rates of different algorithms in the embodiment of the present invention under the Net_B network topology;

[0112] Figure 6 A comparison chart of time consumption of different algorithms in the embodiment of the present invention under the Net_A network topology;

[0113] Figure 7 This is a comparison chart of the time consumption of different algorithms in the embodiment of the present invention under the Net_B network topology. DETAILED DESCRIPTION

[0114] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0115] Example 1: This example uses a time-sensitive network (TSN) incremental routing scheduling method based on the Dijkstra algorithm. A TSN transmission model with CQF as the transmission mode is constructed. The model includes network topology and traffic characteristics, and a series of constraints. It can well simulate the resource mapping in TSN and effectively reflect the traffic-routing scheduling problem in a complex traffic scheduling environment.

[0116] As attached Figure 1 As shown, the time-sensitive network incremental routing scheduling method of this embodiment is characterized by including:

[0117] (1) Constructing a multi-routing transmission model for a time-sensitive network (TSN) using cyclic queue forwarding (CQF) as the transmission mode; the model includes network topology and flow characteristics;

[0118] (2) Set the constraints of the model and transform the traffic routing scheduling problem into an optimization problem based on the constraints;

[0119] (3) A greedy strategy is used to prioritize traffic flow. For the selected traffic, the Dijkstra algorithm is used to obtain the optimal path for the traffic. The scheduling process of the time slot offset selection is determined based on the load balancing method to complete the scheduling.

[0120] Specifically, step (1): Here, the actual physical network topology is represented as a directed graph G = {V, E}, where V is the vertex set consisting of hosts and switches, denoted as S and H, respectively. E represents the directed connections between vertices that are interconnected for transmission. Hosts are the source and destination of data flows, generating time-triggered data flows and controlling when data flows are sent.

[0121] The transmission in CQF involves two ports of the TSN switch, but using the path occupied by the port is still not conducive to subsequent calculation and processing. Since the transmission in the actual link is single and linear, the link is used here to represent the path, which is more conducive to subsequent calculation and representation. As shown in the following formula, we define the path of the flow as f i .path, where L(S a ,S b ) and L(S j ,S k ) are the first and last links in a TSN network. S a and S k They are f i The first and last switching vertices.

[0122] f i .path={L(S a ,S b ),L(Sb ,S c ),...,L(S j ,S k )}.

[0123] Here is a simple TSN routing scheduling topology scenario for easy understanding, such as Figure 3 As shown in Figure 2, a large amount of traffic needs to be transmitted from host H1 to H2. If only the shortest path method is used, all traffic will be concentrated on the link L(S1, S4), significantly increasing the scheduling pressure on local links in the network. However, for some traffic with a margin in deadlines, routes such as L(S1, S2) or L(S2, S4) are feasible. Therefore, for traffic with tight deadlines, shorter paths should be selected to ensure successful scheduling. For traffic with longer deadlines, paths with more hops can be selected to reduce resource congestion.

[0124] Here, an 8-tuple is used to describe the characteristics of a time-triggered flow, as shown in the following formula. This tuple includes the source node, destination node, period, packet size, maximum delay limit, routing path, offset, occurrence frequency, and score. The first six elements of the tuple represent the different characteristics observed during the flow generation phase, while the offset represents the value used to adjust the scheduling process for each flow. The frequency refers to the number of times a packet occurs within a specified scheduling interval.

[0125] f i ={src,dst,period,size,deadline,path,offset,frequency,score}.

[0126] Step (2): To ensure the effectiveness and rationality of flow scheduling, we set the following definitions and constraints to ensure that the scheduling mapping here meets the transmission requirements of CQF. Based on these constraints, the flow scheduling problem is transformed into an optimization problem for processing.

[0127] (21) Scheduling period. Since the time-triggered flow of the scheduling object is periodic, we need to define a time length that can accurately map all flows in the flow set without wasting computing resources. This time length is called the scheduling period, which is defined as the lowest common multiple of all flows.

[0128] sched cycle =LCM(F.periods).

[0129] Given the scheduling period, we can calculate f i .frequency, as shown in the following formula:

[0130]

[0131] (22) Time slot constraint. Here, the time slot is defined as the granularity of time offset, that is, the unit describing the shortest time period in the resource space. It requires that the time slot is divisible by the period of all flows, so the upper bound of the time slot can be expressed as the greatest common factor of the period of all flows. Similarly, since CQF forwarding requires that the sending and receiving of a queue must be in the same time slot, we make the time slot long enough to satisfy packet forwarding. Therefore, the lower bound of the time slot is the time required for a queue to send and receive.

[0132]

[0133] Among them, d proc is the sum of the maximum delay caused by switch transmission, queuing and processing, d sync The maximum delay for time synchronization.

[0134] (23) Offset constraint. For each flow, the offset should be less than its period. That is, packets in a flow must be sent before the next packet arrives. Otherwise, data packets with multiple periods will occupy the storage resources of the host or switch, resulting in data loss or delayed delivery. Both of these outcomes are unacceptable for TSN, which requires deterministic transmission.

[0135]

[0136] (24) Deadline constraint. In the flow feature tuple, there is a deadline value, which means that the flow needs to be sent to the target vertex before this time. We set this deadline constraint to avoid the situation where the flow does not arrive on time.

[0137]

[0138] (f i .offset+|f i .path|+2)×slot cycle ≤f i .deadline.

[0139] Here (|f i .path|+2) represents the number of time slots required during the streaming process. i .path consists only of links consisting of switches, and the number of additional links between the source node and the receiving node is 2, then |f i .path|Used to get the time slots spent in transmission.

[0140] (25) Queue resource limitation. A resource block Q is designed to describe the queue resources, representing the time slot T of switch j. (t) The queue resource at port k and a mapping value is designed To indicate whether the corresponding flow occupies the resource block. When the flow occupies the resource block, the corresponding is 1, otherwise it is 0. The following is the definition of the resource block mapping value:

[0141]

[0142] Where μ represents the μ-th data packet of the flow in the scheduling period, which is used to calculate the slot position occupied by the flow.

[0143] For the set of flows that complete the scheduling steps, the resources they occupy should satisfy the following relationship to prevent resource overflow.

[0144]

[0145] where we use Φ(i) to represent f i Whether it is successfully scheduled. For example, when f i If the scheduling is successful, the value of Φ(i) is 1, otherwise it is 0.

[0146] (26) The optimization goal of flow scheduling is to maximize the number of successfully scheduled flows, that is, the scheduling success rate:

[0147]

[0148] Step (3): In using algorithms to solve the time-triggered flow scheduling problem, we propose a time-sensitive network incremental routing scheduling method based on the Dijkstra algorithm: First, a greedy flow processing selection step is performed. Second, for the selected flow, incremental routing selection is performed. Finally, the scheduling process of the time slot offset selection is determined based on the load balancing method. The above three steps constitute our proposed flow scheduling algorithm.

[0149] (31) Greedy-based traffic processing selection step. In the traffic feature tuple, I select three of them: size, period, and frequency to perform a greedy selection strategy.

[0150] In order to perform unified weighted processing, the three sets of data are first subjected to Min-Max normalization:

[0151]

[0152] Next, we take the weighted sum of the data in the above equation and use that result as the score for the tuple data:

[0153] f i .score = α × f i .size new +β×f i .frequency new +γ×f i .deadline new .

[0154] Where α, β and γ are weighting factors of data size, occurrence frequency and cutoff time respectively, all of which are positive numbers and sum to 1.

[0155] The traffic subsets are sorted in ascending order according to the size of the scores, and the number of successfully scheduled flows can be maximized by adjusting the weighting factor value.

[0156] (32) Incremental routing step. In this part, a resource graph G in the form of bidirectional links is established based on the network directed graph in the problem description. * , which only includes the switch endpoints and the links connecting them. Initialize the distance on each link to 1. Based on this, use the Dijkstra algorithm to solve the shortest path, and update the distance of the corresponding link in the directed graph according to the obtained optimal path. The following is an example: When solving f i The shortest path contains the link L(S a ,S b ), the distance of the graph is updated according to the following formula:

[0157] Dis(L(S a ,S b ))+=f i .size new *f i .frequency new .

[0158] For the traffic scheduled later, the route selection operation is performed according to the updated network resources.

[0159] (33) Offset selection step. In this step, we traverse all offsets and exclude those that exceed the remaining resources. Among the remaining feasible offsets, we calculate the maximum resource usage of all paths. Finally, we select the offset with the smallest maximum resource usage for data mapping.

[0160] As mentioned above, a resource block can be represented as In order to observe the (t) The resource occupancy of the directed link from switch j to switch k is shown in the figure. Now assume that a certain number of flows have been successfully mapped to the resource space, and the specific capacity occupied in the resource block is It can be expressed as:

[0161]

[0162] Then, the scheduling resource increment caused by its corresponding offset is added to the existing resource occupancy In [1], we substitute the resource constraints for judgment and filter out the offsets that meet the resource constraints. For each offset that meets the conditions, we traverse all path nodes in the path, find the resource usage of the node with the largest space usage in the path, and select the offset that minimizes this value for resource mapping, thus completing the scheduling step.

[0163]

[0164] The specific algorithm implementation process is as follows:

[0165]

[0166]

[0167] Based on the same inventive concept, this embodiment also provides a time-sensitive network incremental routing scheduling system, including:

[0168] Model building module; used to build a time-sensitive network TSN multi-routing transmission model using cyclic queue forwarding (CQF) as the transmission mode; the model includes network topology and flow characteristics;

[0169] The conversion module is used to set the constraints of the model and convert the traffic routing scheduling problem into an optimization problem based on the constraints;

[0170] The scheduling module is used to process the priority of traffic processing using a greedy strategy. For the selected traffic, the Dijkstra algorithm is used to obtain the optimal path of the traffic, and the scheduling processing of the time slot offset selection is determined according to the load balancing method to complete the scheduling.

[0171] Furthermore, in the model building module, the network topology is represented as a directed graph G = {V, E}, where V is a set of vertices S, which are hosts, and H, which are switches, and E represents directed connections between vertices that are interconnected for transmission. The hosts are the source and destination of data flows, and are used to generate time-triggered data flows and control the time at which data flows are sent.

[0172] The flow is characterized by the following equation,

[0173] f i ={src,dst,period,size,deadline,path,offset,frequency,score}

[0174] Among them, f i represents the i-th flow, where the source node src, destination node dst, period period, packet size size, deadline deadline, and path path are the different features observed during the flow generation phase. The offset offset represents the value used to adjust the scheduling process of each flow, and the frequency frequency represents the number of times a packet appears within the specified scheduling interval. The score score is calculated based on the known feature values.

[0175] The path is represented by a link, as shown in the following formula:

[0176] f i .path={L(S1,S2),L(S2,S3),...,L(S n-1 ,S n )}

[0177] Among them, f i .path is stream f i The paths L(S1,S2) and L(S n-1 ,S n ) are the first and last links in the TSN network, S1 and S n They are f i The first and last vertices of .

[0178] Furthermore, the conversion module is specifically implemented as follows:

[0179] Set the scheduling period constraint. Specifically, define a time length that is the least common multiple of the period F.periods of the flow set as the scheduling period sched cycle , mapping all flows in a flow collection:

[0180] sched cycle =LCM(F.periods).

[0181] Where F is the set of flows, LCM represents the least common multiple; based on a given scheduling period, calculate f i Frequency f i .frequency, as shown below:

[0182]

[0183] Set the time slot constraint. Specifically, the time slot is the granularity of the time offset, which is divisible by the period of all flows and is expressed as the greatest common factor of the period of all flows. The lower bound of the time slot is the time required for a queue to send and receive:

[0184]

[0185] Among them, slot cycle Indicates time slot, Represents a positive number, GCD represents the greatest common factor, Queue size is the measure of queue size, MTU is the maximum transmission unit, B is the bandwidth, d proc is the sum of the maximum delay caused by switch transmission, queuing and processing, d sync is the maximum delay for time synchronization;

[0186] Set the offset constraint. Specifically, make the offset of the stream smaller than the period of the stream:

[0187]

[0188] Among them, f i .offset is f i The offset of , N is the total number of streams;

[0189] Set the deadline constraint, f i The deadline f i .deadline, which indicates the latest time the flow needs to be sent to the target vertex. The specific settings are as follows:

[0190]

[0191] (f i .offset+|f i .path|+2)×slot cycle ≤f i .deadline.

[0192] Among them, (|f i .path|+2) indicates the number of time slots required during the stream transmission process;

[0193] Set the queue resource constraints. Specifically, let the queue resource block Q represent the switch j T (t) The time slot is the queue resource at port k, and the mapping value is set Indicates whether the corresponding flow occupies the resource block. When the flow occupies the resource block, 1 if yes, 0 otherwise:

[0194]

[0195] Where hop represents the path length, mod represents the modulo operation, and μ represents the μth packet of the flow in the scheduling period, which is used to calculate the slot position occupied by the flow;

[0196] For a set of flows that completes the scheduling step, the occupied resources satisfy the following relationship to prevent resource overflow:

[0197]

[0198] Where Φ(i) represents f i Whether it is successfully scheduled, when f i When the scheduling is successful, the value of Φ(i) is 1, otherwise it is 0;

[0199] Let maximizing the number of successfully scheduled flows be the scheduling optimization objective:

[0200]

[0201] Among them, F success Indicates the number of successfully scheduled flows.

[0202] Furthermore, the scheduling module is specifically implemented as follows:

[0203] Perform traffic processing selection based on a greedy strategy. Specifically, in the traffic feature tuple, select size, period, and frequency to perform a greedy selection strategy.

[0204] Perform Min-Max normalization on the three sets of data, then perform weighted sum on the obtained data and define the result as a score:

[0205] f i .score = α × f i .size new +β×f i .frequency new +γ×f i .deadline new .

[0206] Where α, β, and γ are weighting factors of packet size, frequency, and deadline, respectively. They are all positive and sum to 1.

[0207] Sort the traffic subsets in ascending order according to the size of the scores, and maximize the number of successfully scheduled flows by adjusting the weighting factor values;

[0208] Perform incremental routing; specifically, establish a resource graph G in the form of a bidirectional link based on the directed graph G * , the resource legend G * It only contains switch endpoints and interconnected links; initialize the distance on each link to 1, use Dijkstra algorithm to solve the shortest path, and update the distance of the corresponding link in the directed graph according to the obtained optimal path. When f iThe shortest path contains the link L(S a ,S b ), the distance of the graph is updated as follows:

[0209] Dis(L(S a ,S b ))+=f i .size new *f i .frequency new

[0210] For the traffic scheduled later, the route selection operation is performed according to the updated network resources;

[0211] Select an offset. Specifically, traverse all offsets, exclude offsets that exceed the remaining resources, calculate the maximum resource usage of all paths among the remaining feasible offsets, and finally select the offset with the minimum maximum resource usage for data mapping.

[0212] Specifically, let the resource block be represented as For observing the time slot T (t) The resource occupancy of the directed link from switch j to switch k is assuming that a certain number of flows are successfully mapped to the resource space and the capacity occupied in the resource block is Expressed as:

[0213]

[0214] f i The scheduling resource increment caused by the corresponding offset is added to the existing resource occupancy In [1], we substitute the resource constraints for judgment and filter out the offsets that meet the resource constraints. For each offset that meets the conditions, we traverse all path nodes in the path, find the resource occupancy of the node with the largest space in the path, and select the offset that minimizes this value to perform resource mapping and complete the scheduling step. The formula is as follows:

[0215]

[0216] Based on the same inventive concept, this embodiment also provides a computing device, including: one or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the time-sensitive network incremental routing scheduling method according to any one of the above items are implemented.

[0217] Based on the same inventive concept, this embodiment also provides a computer-readable storage medium, which stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the time-sensitive network incremental routing scheduling method according to any one of the above items.

[0218] Example 2: The embodiment of the present invention is described in detail as follows. The system simulation is constructed using Python. Based on the distribution type of the industrial control network, two sets of TSN network topologies for reference industrial automation smart grids are given here. As shown in Table 1:

[0219] Table 1 Network topology example

[0220]

[0221] The basic network topology consists of switch nodes and bidirectional links between them. Based on the aforementioned constraints, we set the queue size to 10,000 bits, the link bandwidth to 1000 Mbps, and the time slot to 125 μs. Furthermore, in the flow generation step, we randomly generated time-triggered data flows, referring to the data flow characteristics described in the IEC / IEEE 60802 standard for industrial automation networks. We set two sets of periods to be adjusted, randomly selected from integers between 3 and 10. The packet size was selected from {200, 400, 500, 1000, 1500}. The flow deadline and period are typically measured in milliseconds (ms), so here we set the flow deadline to its period minus 0.5 ms. All experimental results described below are the average of 100 simulations under the aforementioned network topology and random traffic generation conditions.

[0222] Figure 4 and Figure 5 This is a comparison chart of the scheduling success rates of different algorithms under different circumstances in the embodiments of the present invention. The greedy algorithm, taboo search algorithm and naive algorithm were selected for comparison with the algorithm in this article. Among them, the greedy algorithm is a scheme that arranges the sizes in the flow characteristics in ascending order and then schedules. The tabu search algorithm is a heuristic algorithm that optimizes CQF scheduling by adjusting and optimizing the traffic scheduling results in two modes: exchanging the successful scheduling-failed scheduling traffic set or adjusting the time slot offset position of the successfully scheduled traffic. The Navie algorithm is a scheme that does not perform any traffic processing and transmits in the order of the flow set input.

[0223] like Figure 4 As shown in the case of Net_A network topology, the scheduling success rate of the proposed algorithm is compared with greedy, tabu and There are advantages of 5.1426%, 4.9762% and 35.5361% respectively.

[0224] like Figure 5 As shown in the case of Net_B network topology, the proposed algorithm compares greedy, tabu and They are 5.2461%, 5.1343% and 46.1003% higher respectively.

[0225] It can be concluded that the incremental routing scheduling algorithm proposed in the present invention has obvious advantages over greedy and tabu in terms of scheduling success rate, and can significantly improve the scheduling success rate.

[0226] Figure 6 and Figure 7 This is a comparison chart of the time consumption of the six algorithms in the embodiment of the present invention under different circumstances, used to show the differences in time consumption of each algorithm.

[0227] like Figure 6 As shown in the figure, under the Net_A network topology, as the total traffic increases, the time consumption of the proposed algorithm decreases from 0.98307s to 2.9892s. The time consumption of the greedy, tabu, and navie algorithms decreases from 0.22248s, 31.44196s, and 0.12853s to 0.9156s, 41.04565s, and 0.25555s, respectively.

[0228] And as Figure 7 As shown, in Net_B, the overall growth trend of each algorithm is the same as in Net_A, but the data is of a relatively larger order of magnitude. This experiment shows that the proposed incremental routing scheduling algorithm has a slight increase in time consumption compared to the greedy algorithm, but it still remains at a relatively low order of magnitude, showing a significant advantage over the tabu algorithm in terms of time consumption.

[0229] In summary, the present invention proposes a time-sensitive network incremental routing scheduling method based on the Dijkstra algorithm, constructs a multi-route TSN network model based on the CQF mechanism, proposes an optimization problem with the goal of maximizing the traffic scheduling success rate, and solves it using incremental routing scheduling. Experimental results show that the scheduling success rate of the algorithm has obvious advantages over the greedy and tabu algorithms. In terms of time consumption, the algorithm we proposed does not increase significantly compared with a series of algorithms using the shortest path. In general, the algorithm we proposed shows excellent performance in network routing scheduling scenarios involving complex traffic compositions in the field of industrial automation. The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited to this. Any changes or replacements that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A time-sensitive network incremental routing scheduling method, characterized in that: include: (1) A multi-route transmission model for a time-sensitive network (TSN) using cyclic queue forwarding (CQF) as the transmission mode is constructed. The model includes network topology and flow characteristics. The network topology is represented as a directed graph G = {V, E}, where V is a set of vertices S, which are hosts, and H, which are switches. E represents the directed connections between vertices that are interconnected for transmission. The hosts are the source and destination of data flows, and are used to generate time-triggered data flows and control the sending time of data flows. The flow is characterized by the following equation, f i ={src,dst,period,size,deadline,path,offset,frequency,score} Among them, f i represents the i-th flow, where the source node src, destination node dst, period period, packet size size, deadline deadline, and path path are the different features observed during the flow generation phase. The offset offset represents the value used to adjust the scheduling process of each flow, and the frequency frequency represents the number of times a packet appears within the specified scheduling interval. The score score is calculated based on the known feature values. The path is represented by a link, as shown in the following formula: f i .path={L(S1,S2),L(S2,S3),...,L(S n-1 ,S n )} Among them, f i .path is stream f i The paths L(S1,S2) and L(S n-1 ,S n ) are the first and last links in the TSN network, S1 and S n They are f i The first and last vertices of ; (2) Set the constraints of the model and transform the traffic routing problem into an optimization problem based on the constraints; specifically, (21) Set the scheduling period constraint. Specifically, define a time length that is the least common multiple of the period F.periods of the flow set as the scheduling period sched cycle , mapping all flows in a flow collection: sched cycle =LCM(F.periods). Where F is the set of flows, LCM represents the least common multiple; based on a given scheduling period, calculate f i Frequency f i .frequency, as shown below: (22) Set a time slot constraint. Specifically, the time slot is the granularity of the time offset, which is divisible by the period of all flows and is expressed by the greatest common factor of the period of all flows. The lower bound of the time slot is the time required for a queue to send and receive: Among them, slot cycle Indicates time slot, Represents a positive number, GCD represents the greatest common factor, Queue sizef is the measure of queue size, MTU is the maximum transmission unit, B is the bandwidth, d proc is the sum of the maximum delay caused by switch transmission, queuing and processing, d sync is the maximum delay for time synchronization; (23) Set an offset constraint. Specifically, make the offset of the flow smaller than the period of the flow: Among them, f i .offset is f i The offset of , N is the total number of streams; (24) Set the deadline constraint, f i The deadline f i .deadline, which indicates the latest time the flow needs to be sent to the target vertex. The specific settings are as follows: (f i .offset+|f i .path|+2)×slot cycle ≤f i .deadline. Among them, (|f i .path|+2) indicates the number of time slots required during the stream transmission process; (25) Set the queue resource constraints. Specifically, let the queue resource block Q represent the switch j, T (t) The time slot is the queue resource at port k, and the mapping value is set Indicates whether the corresponding flow occupies the resource block. When the flow occupies the resource block, 1 if yes, 0 otherwise: Where hop represents the path length, mod represents the modulo operation, and μ represents the μth packet of the flow in the scheduling period, which is used to calculate the slot position occupied by the flow; For a set of flows that completes the scheduling step, the occupied resources satisfy the following relationship to prevent resource overflow: Where Φ(i) represents f i Whether it is successfully scheduled, when f i When the scheduling is successful, the value of Φ(i) is 1, otherwise it is 0; (26) Let maximizing the number of successfully scheduled flows be the scheduling optimization objective: Among them, F success Indicates the number of successfully scheduled flows; (3) A greedy strategy is used to prioritize traffic flow. For the selected traffic, the Dijkstra algorithm is used to obtain the optimal path for the traffic. The scheduling process of the time slot offset selection is determined based on the load balancing method to complete the scheduling. Specifically, the following steps are performed: (31) Perform traffic processing selection based on a greedy strategy; specifically, in the traffic feature tuple, select size, period, and frequency to perform a greedy selection strategy; Perform Min-Max normalization on the three sets of data, then perform weighted sum on the obtained data and define the result as a score: f i .score=α×f i .size new +β×f i .frequency new +γ×f i .deadline new . Where α, β, and γ are weighting factors of packet size, frequency, and deadline, respectively. They are all positive and sum to 1. Sort the traffic subsets in ascending order according to the size of the scores, and maximize the number of successfully scheduled flows by adjusting the weighting factor values; (32) Perform incremental routing; specifically, establish a resource graph G in the form of a bidirectional link based on the directed graph G. * , the resource legend G * It only contains switch endpoints and interconnected links; initialize the distance on each link to 1, use Dijkstra algorithm to solve the shortest path, and update the distance of the corresponding link in the directed graph according to the obtained optimal path. When f i The shortest path contains the link L(S a ,S b ), the distance of the graph is updated as follows: Dis(L(S a ,S b ))+=f i .size new *f i .frequency new For the traffic scheduled later, the route selection operation is performed according to the updated network resources; (33) Select an offset; specifically, traverse all offsets, exclude offsets that exceed the remaining resources, calculate the maximum resource occupancy of all paths among the remaining feasible offsets, and finally select the offset with the smallest maximum resource occupancy for data mapping; Specifically, let the resource block be represented as For observing the time slot T (t) The resource occupancy of the directed link from switch j to switch k is assuming that a certain number of flows are successfully mapped to the resource space and the capacity occupied in the resource block is Expressed as: f i The scheduling resource increment caused by the corresponding offset is added to the existing resource occupancy In [1], we substitute the resource constraints for judgment and filter out the offsets that meet the resource constraints. For each offset that meets the conditions, we traverse all path nodes in the path, find the resource occupancy of the node with the largest space in the path, and select the offset that minimizes this value to perform resource mapping and complete the scheduling step. The formula is as follows:

2. A time-sensitive network incremental routing scheduling system, characterized in that: include: Model building module; A multi-route transmission model for a time-sensitive network (TSN) using cyclic queue forwarding (CQF) as the transmission mode is constructed; the model includes network topology and flow characteristics; the network topology is represented as a directed graph G = {V, E}, where V is a set of vertices S, which are hosts, and H, which are switches; and E represents directed connections between vertices that are interconnected for transmission; the hosts are the source and destination of data flows, and are used to generate time-triggered data flows and control the timing of data flow transmission; The flow is characterized by the following equation, f i ={src,dst,period,size,deadline,path,offset,frequency,score} Among them, f i represents the i-th flow, where the source node src, destination node dst, period period, packet size size, deadline deadline, and path path are the different features observed during the flow generation phase. The offset offset represents the value used to adjust the scheduling process of each flow, and the frequency frequency represents the number of times a packet appears within the specified scheduling interval. The score score is calculated based on the known feature values. The path is represented by a link, as shown in the following formula: f i .path={L(S1,S2),L(S2,S3),...,L(S n-1 ,S n )} Among them, f i .path is stream f i The paths L(S1,S2) and L(S n-1 ,S n ) are the first and last links in the TSN network, S1 and S n They are f i The first and last vertices of ; The conversion module is used to set the constraints of the model and convert the traffic routing scheduling problem into an optimization problem based on the constraints. The specific implementation method is as follows: Set the scheduling period constraint. Specifically, define a time length that is the least common multiple of the period F.periods of the flow set as the scheduling period sched cycle , mapping all flows in a flow collection: sched cycle =LCM(F.periods). Where F is the set of flows, LCM represents the least common multiple; based on a given scheduling period, calculate f i Frequency f i .frequency, as shown below: Set the time slot constraint. Specifically, the time slot is the granularity of the time offset, which is divisible by the period of all flows and is expressed as the greatest common factor of the period of all flows. The lower bound of the time slot is the time required for a queue to send and receive: Among them, slot cycle Indicates time slot, Represents a positive number, GCD represents the greatest common factor, Queue size is the measure of queue size, MTU is the maximum transmission unit, B is the bandwidth, d proc is the sum of the maximum delay caused by switch transmission, queuing and processing, d sync is the maximum delay for time synchronization; Set the offset constraint. Specifically, make the offset of the stream smaller than the period of the stream: Among them, f i .offset is f i The offset of , N is the total number of streams; Set the deadline constraint, f i The deadline f i .deadline, which indicates the latest time the flow needs to be sent to the target vertex. The specific settings are as follows: Among them, (|f i .path|+2) indicates the number of time slots required during the stream transmission process; Set the queue resource constraints. Specifically, let the queue resource block Q represent the switch j T (t) The time slot is the queue resource at port k, and the mapping value is set Indicates whether the corresponding flow occupies the resource block. When the flow occupies the resource block, 1 if yes, 0 otherwise: Where hop represents the path length, mod represents the modulo operation, and μ represents the μth packet of the flow in the scheduling period, which is used to calculate the slot position occupied by the flow; For a set of flows that completes the scheduling step, the occupied resources satisfy the following relationship to prevent resource overflow: Where Φ(i) represents f i Whether it is successfully scheduled, when f i When the scheduling is successful, the value of Φ(i) is 1, otherwise it is 0; Let maximizing the number of successfully scheduled flows be the scheduling optimization objective: Among them, F success Indicates the number of successfully scheduled flows; The scheduling module is used to process the priority of traffic processing using a greedy strategy. For the selected traffic, the Dijkstra algorithm is used to obtain the optimal path for the traffic. The scheduling process is completed by determining the time slot offset selection based on the load balancing method. The specific implementation method is as follows: Perform traffic processing selection based on a greedy strategy. Specifically, in the traffic feature tuple, select size, period, and frequency to perform a greedy selection strategy. Perform Min-Max normalization on the three sets of data, then perform weighted sum on the obtained data and define the result as a score: f i .score=α×f i .size new +β×f i .frequency new +γ×f i .deadline new . Where α, β, and γ are weighting factors of packet size, frequency, and deadline, respectively. They are all positive and sum to 1. Sort the traffic subsets in ascending order according to the size of the scores, and maximize the number of successfully scheduled flows by adjusting the weighting factor values; Perform incremental routing; specifically, establish a resource graph G in the form of a bidirectional link based on the directed graph G * , the resource legend G * It only contains switch endpoints and interconnected links; initialize the distance on each link to 1, use Dijkstra algorithm to solve the shortest path, and update the distance of the corresponding link in the directed graph according to the obtained optimal path. When f i The shortest path contains the link L(S a ,S b ), the distance of the graph is updated as follows: Dis(L(S a ,S b ))+=f i .size new *f i .frequency new For the traffic scheduled later, the route selection operation is performed according to the updated network resources; Select an offset. Specifically, traverse all offsets, exclude offsets that exceed the remaining resources, calculate the maximum resource usage of all paths among the remaining feasible offsets, and finally select the offset with the minimum maximum resource usage for data mapping. Specifically, let the resource block be represented as For observing the time slot T (t) The resource occupancy of the directed link from switch j to switch k is assuming that a certain number of flows are successfully mapped to the resource space and the capacity occupied in the resource block is Expressed as: f i The scheduling resource increment caused by the corresponding offset is added to the existing resource occupancy In [1], we substitute the resource constraints for judgment and filter out the offsets that meet the resource constraints. For each offset that meets the conditions, we traverse all path nodes in the path, find the resource occupancy of the node with the largest space in the path, and select the offset that minimizes this value to perform resource mapping and complete the scheduling step. The formula is as follows:

3. A computing device, characterized in that include: One or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the time-sensitive network incremental routing scheduling method according to claim 1 are implemented.

4. A storage medium, characterized in that The storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the time-sensitive network incremental routing scheduling method according to claim 1.

Images