A method for allocating resources for a TSN network

Abstract

This invention discloses a method for allocating resources in a TSN network, comprising the following steps executed sequentially: Step 1: Preprocessing TSN service flow resource allocation requests, classifying service requests, and dividing flow demands into services with strict latency requirements and services with general latency requirements according to priority and cost-effectiveness ratio calculation; calculating the transmission path of service flow fij and the scheduling period T of all flows; Step 2: Allocating resources for services with strict latency requirements and locking the time window for resource occupancy; Step 3: Allocating resources for services with general latency requirements; Step 4: Outputting results, including time window data and service deployment statistics. This invention significantly optimizes the traditional NP-hard problem solving method, improving the cost-effectiveness ratio and efficiency of resource allocation planning.

Description

Technical Field

[0001] This invention relates to computer network technology, and more particularly to a method for allocating resources in a TSN network. Background Technology

[0002] The processing of TSN time-sensitive data frames has the following characteristics; firstly, it is time-constrained, neither strictly slot-based in TTE nor cycle-by-cycle switching in CQF, such as... Figure 1 , Figure 2 As shown, 802.1Q features enhanced multi-priority scheduling. Furthermore, it employs strict multi-priority FIFO switching behavior.

[0003] In recent years, as Time-Sensitive Networking (TSN) technology has received wider and greater attention in various application areas, IEEE has also conducted research and developed standards for its application in vertical industries. The TSN standard primarily describes the implementation of TSN end devices (single-port switches) and TSN switching devices; it is a large and continuously expanding standard system. For example... Figure 3 As shown, it currently involves:

[0004] 1) 802.1AS is responsible for providing global time to Qch and Qbv;

[0005] 2) Qci, Qch, and Qcr all describe enqueue control;

[0006] 3) Qbv, Qch, and Qav all describe dequeue control;

[0007] 4) Qbu mainly introduces bandwidth preemption for high-priority frames;

[0008] 5) CB mainly introduces multi-link frame duplication and elimination;

[0009] Qcc primarily provides configuration models for users and networks.

[0010] However, when considering the above standards, most manufacturers often selectively implement only some standards to achieve the goal of upgrading TSN features on older equipment, based on application and cost considerations. The corresponding TSN network resource allocation calculation method needs to take into account support for such equipment. Therefore, most TSN network planning algorithms currently choose to be based on the TTE scheduling algorithm. However, the TTE scheduling algorithm has the following characteristics and problems in TSN adaptation.

[0011] 1) Based on SMT-solver (satisfiability modulo theories). SAT (Satisfiability), given a propositional logic formula, determines whether there exists a set of satisfiability assignments (NP-hard; if it exists, give the value; efficient solver DPLL--Davis-Putnam-Logemann-Loveland). The ability to describe this using the form "propositional logic formula == Boolean expression" (variables + operators) is insufficient.

[0012] 2) Formal rules / constraints, CNF (conjunctive normal form) are input into the SMT solver; there are efficiency issues (for example, experiments showed that with 10 switching nodes, 50 end nodes, and 50 streams [175 data frames], the solution time was as high as 100 minutes).

[0013] The algorithm suffers from NP-hard problems in resource allocation, lacks fine-grained mutual exclusion control, has a low degree of parallel computation, and is not well integrated with the existing Internet distributed computing architecture. Summary of the Invention

[0014] This invention aims to provide a method for allocating resources in TSN networks, which optimizes the traditional NP-hard problem-solving method and improves the cost-effectiveness and efficiency of resource allocation planning.

[0015] To achieve the above objectives, the present invention employs the following technical solution:

[0016] A method for allocating resources in a TSN network includes the following steps performed sequentially:

[0017] Step 1: Preprocess the TSN service flow resource allocation request, classify the service requests, and divide the flow demand into services with strict latency requirements and services with general latency requirements according to priority and cost-effectiveness ratio; calculate the transmission path of service flow fij and the scheduling period T of all flows;

[0018] Step 2: Allocate resources for services with strict latency requirements and lock in the time window for resource usage;

[0019] Step 3: Allocate resources for services with general latency requirements;

[0020] Step 4: Output the results, which include time window data and business deployment statistics.

[0021] Preferably, in step 2, after preprocessing the stream, a two-step heuristic search is used to allocate resources for services with strict latency requirements using offset; in step 3, resources are allocated for services with general latency requirements using offset + buffer + window merging.

[0022] Preferably, step 1 includes the following sub-steps:

[0023] Step 1.1: Based on fij.e2e, classify business requests according to latency requirements, where: Fstrict represents services with strict latency requirements, and Fnormal represents services with general latency requirements;

[0024] Step 1.2: Based on i, j and the network topology, calculate the shortest path SP of fij;

[0025] Step 1.3: Calculate the least common multiple based on fij.period of all service flows to obtain the scheduling period T under the current service request;

[0026] Step 1.4: Calculate the number of repetitions M for each fij within the scheduling period T; treat each repetition as a new business request and add it to the ST business flow set.

[0027] Preferably, step 2 includes the following sub-steps:

[0028] Step 2.1: Sort the business requests in the Fstrict collection in ascending order by fij.e2e; clear the GCL data structure;

[0029] Step 2.2: Calculate the time window value of the data frame on the output port / Queue for each business request in the Fstrict set as it passes through each node along the path direction.

[0030] Furthermore, step 2 also includes the following sub-steps:

[0031] Step 2.3: Calculate whether the total latency through multiple nodes and multiple links meets the requirements.

[0032] Furthermore, sub-step 2.3 also includes:

[0033] When the total latency meets the requirements, it is determined whether the time window occupied by the flow on each node Q conflicts with other flows. If there is no conflict, the time spent passing through the Queue is recorded; if there is a conflict, a conflict avoidance algorithm is used to avoid the conflict.

[0034] Preferably, the conflict avoidance algorithm includes:

[0035] Based on the time windows allocated to node Q within a scheduling cycle and the flow fi occupying those time windows, the process is as follows:

[0036] If the fi time window partially overlaps with an existing time window: update fi.offset = fi.offset + overlap length. After the offset is increased, the time window is lengthened. After the time window is lengthened, if it overlaps with the next time window, the resources of the next time window are reclaimed, and its corresponding flow fj is put back into Fstrict to participate in the next allocation calculation.

[0037] Recalculate the time window conflict situation of the flow fi on each Q in the path direction. If there is a conflict, restart the avoidance algorithm. If there is still a conflict when fi.offset reaches offsetMAX, put fj into the Fnormal set. If there is no conflict, end the current fi calculation.

[0038] Preferably, step 3 includes the following sub-steps:

[0039] Step 3.1: For services with general latency requirements, after calculating the service path SP, calculate the latency limit for each service flow and each hop.

[0040] Step 3.2: Sort each Q in descending order of the total length of the time windows it contains;

[0041] Step 3.3: Determine the relationship between the window time of the general latency requirement service appearing in a certain Q and the window of the deployed service, and update the window information of each Q according to the relationship.

[0042] Preferably, the time window data includes the time window data of node Q, and the time window data of node Q includes:

[0043] The Talker node transmit time offset fi.offset, the node receive time policy, the GCL of each switching node's output port, the forwarding MAC table of each node based on the listener MAC address, and the node flow classification table based on source and destination MAC addresses, VID and / or priority encoding.

[0044] Preferably, the business deployment statistics include:

[0045] Successful deployment ID, failure ID;

[0046] Node Q time window information statistics and GCL length information.

[0047] This invention provides a time-window-based TSN network resource allocation method for next-generation TSN networks.

[0048] This invention, based on SMT-solver's merge optimization, ensures deterministic forwarding of service flows. Critical traffic will not be lost due to congestion, latency has a definite upper limit, and packets appear at definite locations at definite times.

[0049] This invention significantly optimizes the traditional solution method for NP problems, improving the cost-effectiveness ratio and efficiency of resource allocation planning.

[0050] This invention uses a graph approach to support the description of network topology resources with different characteristics, supports the setting of global resource constraints and single-node resource constraints for nodes, and supports different flow forwarding time accuracies, thereby improving the adaptability of the algorithm to networks with different characteristics.

[0051] This invention performs well in networks with the characteristics of the Barabási-Albert model and has good compatibility with TSN-featured devices based on Ethernet 1.0, as referenced by TTE. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the existing 802.1Q multi-priority FIFO.

[0053] Figure 2 This is a schematic diagram of the existing 802.1Q standard.

[0054] Figure 3 This is a system block diagram for the existing 802.1 technology.

[0055] Figure 4 This is a flowchart of the present invention.

[0056] Figure 5 This is a schematic diagram of node time window merging in this invention.

[0057] Figure 6 This is a schematic diagram of step 2 of the present invention.

[0058] Figure 7 This is a schematic diagram of step 3 of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings.

[0060] Example 1

[0061] This embodiment discloses a method for allocating TSN network resources, such as... Figure 1 As shown, the processing flow is as follows:

[0062] 1. Preprocessing: TSN service flow classification, calculation of the least common multiple of service flow cycles, and shortest path routing;

[0063] 2. After sorting by Q, the "hot" queue is placed at the front, and the starting path service flow fi is searched from it (i.e., the switch where Q is located is the first hop node of service fi);

[0064] 2.1 If there is no starting path business flow in Q, then search in the next Q in sequence until all Qs are searched;

[0065] 2.2 Every business flow must have a first hop node, so searching all Q is equivalent to processing all business requests.

[0066] 3. Based on the time window information of the current frame (assuming the best-case scenario) and the currently deployed time window information, determine the relationship between the current frame and the deployed windows;

[0067] 3.1 Expand the current window, open the door earlier or close it later, and include the current data frame in the adjusted window;

[0068] 3.2 If the conditions are not met, add a new window.

[0069] 4. When the current frame increases by offset or buffer_delay changes, recalculate the time window information of other Qs in the path direction, and judge and adjust the time window one Q at a time;

[0070] 5. Until the listener is reached, update the deployed time window information;

[0071] 6. Perform the same operation on the next selected business flow;

[0072] 7. After processing all streams, output the time window data and business deployment statistics for each Q.

[0073] Specifically, such as Figure 5 As shown, it includes:

[0074] Step 1: First, preprocess the TSN service flow resource allocation requests, classify the service requests, and divide the flow requirements into services with strict latency requirements and services with general latency requirements according to priority and cost-effectiveness calculation. Calculate the transmission path of service flow fij and the scheduling period T of all flows;

[0075] Step 2: After preprocessing the stream, a two-step heuristic search is used. First, the offset is used to allocate resources for services with strict latency requirements, prioritizing the critical business flow and locking the time window for resource occupation.

[0076] Step 3: Merge offset, buffer, and window into resources for general latency-critical applications;

[0077] Step 4: Output the results, including time window data and business deployment statistics.

[0078] The above steps strictly meet e2e requirements, with no slot or ping-pong operations; compress GCL depth, reduce 1Q switch requirements, reduce Guard Band, and improve transmission bandwidth utilization.

[0079] Core elements of the algorithm:

[0080] 1) 1+1 Redundant Access

[0081] ① The path calculation part must guarantee the maximum number of non-intersecting paths;

[0082] ② Process as an independent business.

[0083] 2) Multi-priority queues (SP switch scheduling mechanism)

[0084] ① First, complete the business deployment of high-priority queues (window confirmation);

[0085] ②Lock the high-priority window, shift the offset and / or adjust the buffer_delay of the "best case" data frame window in the low-priority Q, and establish a GCL that is mutually exclusive with it.

[0086] 3) Parallelization

[0087] ① For non-intersecting business paths, the data frames carried by node Q do not affect each other;

[0088] ② After solving the business path for all business flows, classify them according to Q overlap: completely disjoint business flows are treated as independent subsets;

[0089] ③ These subsets can run the allocation algorithm independently;

[0090] ④ The degree of parallelization depends on the source and destination of the service request and the network topology. It is initially unknown, but during the calculation process, the calculation sequence is pre-allocated based on the locking priority or the same priority sorting method. During the calculation process, if a resource conflict occurs, the high sequence number service gives way to the low sequence number service; the granularity of serialization mutual exclusion is reduced to the competition for "port resources on the same node".

[0091] 4) Algorithm Latency Handling Content: End-to-end latency (e2e) is the maximum latency for a data frame to travel from node A to node B. Data frame latency is monotonically increasing in the network; the total e2e latency increases with each link segment and node along the path. This mainly involves:

[0092] ① Link propagation delay: related to the distance between nodes, such as 100ns for 20-meter twisted pair / fiber cable;

[0093] ② Node latency, including transmission processing latency (egress port scheduling, buffer reading, shaping, MAC_Tx, parallel-to-serial conversion, etc., related to processing structure, frame length, Ticks, etc.), reception processing latency (serial-to-parallel conversion, MAC_Rx, policy, counting statistics, classification search, fabric, etc., related to processing structure, frame length, Ticks, etc.), and node buffer latency (data frame buffer waiting latency, related to GCL, egress port scheduling, etc.).

[0094] It is specifically pointed out that when dealing with NP problems, due to the global nature of flow forwarding resources, flow planning often needs to be carried out serially. However, by adopting the time window merging method, the algorithm process reduces the mutual exclusion scale of the serialized flow planning operation to the time window conflict resolution of the same node and the same port by using priority hierarchy and incremental model.

[0095] In step 4,

[0096] 1) Time window data based on each Q

[0097] ① The time offset fi.offset sent by the Talker node;

[0098] ② Node reception time policy (including exchange nodes and Listener nodes);

[0099] ③ The GCL of the output ports of each switching node (appropriately relaxed considering GuardBand);

[0100] ④ MAC table for forwarding at each node based on the listener's MAC address;

[0101] ⑤ Node flow classification table based on source and destination MAC addresses, VID, and / or priority encoding.

[0102] 2) Business Deployment Statistics

[0103] ① Successful deployment service number, failure number;

[0104] Q time window information statistics, GCL length information.

[0105] Example 2

[0106] Based on Example 1, this example discloses the specific process of step 2, such as... Figure 6 As shown, step 2 specifically includes:

[0107] 1) Step 2.1: Sort the Fstrict set of services with strict latency requirements in ascending order by fij.e2e; clear the GCL data structure;

[0108] 2) Step 2.2: According to the "ideal case" or "best case", fij.offset = offset_min and buffer_delay = 0, calculate the time window value of the data frame passing through each node along the path and on the output port / Queue;

[0109] QSWn.Px.open=offset+(link_delay+proc_delay)*hop

[0110] QSWn.Px.close=offset+(link_delay+proc_delay)*hop+size

[0111] 3) Step 2.3: Verify whether the total latency through multiple nodes and multiple links meets the requirements:

[0112] a. Step 2.3.1, satisfying

[0113] ①Does the time window occupied by this flow on each node Q conflict with other flows?

[0114] ② No conflicts, record the time of passage through the Queue – Gate.open, Gate.close;

[0115] ③ If there is a conflict, invoke the conflict avoidance algorithm.

[0116] b. Step 2.3.2, not satisfied.

[0117] ①Since the business requirements (shortest path, buffer_delay=0) cannot be met even in the best-case scenario, there is no solution;

[0118] 4) Step 2.4: Conflict Resolution

[0119] c. Step 2.4.1: Input:

[0120] ① Any Q has been allocated a time window within a scheduling cycle (each Q of each node is maintained independently);

[0121] ②In the "best case" scenario, the streaming fi requires a time window.

[0122] d. Step 2.4.2: Processing flow:

[0123] ① Step 2.4.2.1: The fi time window partially overlaps with the existing time window: Update fi.offset = fi.offset + overlap length (1. No overlap means no conflict; 2. Increase the current data frame offset to avoid the current Q window from overlapping).

[0124] After the offset is increased, the time window is extended (gate.open remains unchanged while gate.close is delayed), no new overlap occurs, proceed to the next step.

[0125] When the time window is lengthened, it overlaps with the next time window: reclaim the resources of the next time window, put its corresponding stream fj back into Fstrict, and let it participate in the next allocation calculation.

[0126] ② Step 2.4.2.2: Based on fi.offset_update, recalculate the time window conflict situation of the flow fi on each Q in the path direction. If a conflict is found, restart the avoidance algorithm (further increase the offset on the new Q);

[0127] If a conflict still exists when offsetMAX is used, fj is put into the Fnormal set, and the next step is to try buffer not being equal to 0 - although this may cause e2e to be too large!

[0128] No conflict, end the current fi calculation and return.

[0129] Step 2.5: Repeat the next business process until Fstrict is empty.

[0130] Example 3

[0131] Based on Embodiment 1 or 2, this embodiment discloses the processing flow of step 3, such as... Figure 7 As shown, step 3 specifically includes:

[0132] 1) During the deployment of strict e2e services, the switch buffer is set to 0: data frames are only "regenerated" at the node, meaning they are sent immediately upon arrival—thus achieving the minimum end-to-end latency. However, offset adjustments are limited, and since offset adjustments need to be applied to high-priority strict e2e services, many services will inevitably fail to deploy. Therefore, for other general e2e services, the focus should be on how to utilize the node buffer.

[0133] 2) Step 3.1: For general latency requirements (including services where offset adjustment is ineffective in strict e2e service deployments), after calculating the service path SP, calculate the latency limit for each service flow and each hop, and uniformly calculate the time window required for each data frame to arrive at Q under the assumption of no buffering (best case).

[0134] 3) Step 3.2: Sort each Q in descending order of the total length of the time windows it contains;

[0135] Step 3.2.1: Calculate Qi weights

[0136] Qi_weight=Qi_sigma(windows) / sigma[Qi_sigma(windows)];

[0137] Sigma(windows) is the total length of all windows in Qi;

[0138] sigma[Qi_sigma(windows)] is the total length of all time windows in all Q.

[0139] Step 3.2.2: Calculate the upper limit of the single-hop delay for each fi through each Q.

[0140] Delay(hopi) = fi.e2e*Qk_weight / sigma(Qpath_weight); that is, the "busier" Q can be allocated a longer queuing time.

[0141] 4) Step 3.3: The relationship between the window time of a general service appearing in a certain Q according to the "best case" and the window of the deployed service (marked as locked window in the strict e2e service deployment section output);

[0142] ① Step 3.3.1: For the faraway data frame on the right, calculate the relationship between its arrival time and the current window's 'opening time', and whether it is less than the current Q fi's Delay(hop_j). If it is less than or equal to, it indicates that this frame is buffered and delayed in the current Q, which does not affect e2e. Calculate the waiting time and update the window information of each Q on the path. If it is greater than, the arrival time is too early, and buffering and waiting for the opening time may not meet the e2e requirements. Directly add a new GCL corresponding to this data frame on the Q.

[0143] ② Step 3.3.2: For the right-hand early data frame, calculate the waiting time (= current frame arrival time - original door opening time), and update the window information of each Q on the path;

[0144] ③ Step 3.3.3: In the above case—open the door in advance, Gate.open = Gate.open - current frame.size;

[0145] ④ Step 3.3.4: For the left Overlap_earlay and left Overlap_late data frames, increase the current frame offset until the frame's arrival time is later than the original closing time. Delayed closing time = gate.close + current frame.size. Based on the updated frame offset, recalculate the time window information for path Q.

[0146] Step 3.3.5: Left-side late and faraway data frames. Calculate the difference between the frame arrival time and the original closing time; if it is less than the threshold (e.g., twice the protection time), delay the closing time = gate.close + buffer_delay + current frame.size; if it is greater than the threshold, add a new GCL item. Of course, this invention can have many other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A method for allocating resources in a TSN network, characterized in that, This includes the following steps performed sequentially: Step 1: Preprocess the TSN service flow resource allocation request, classify the service flows, and divide the flow demand into services with strict latency requirements and services with general latency requirements according to priority and cost-effectiveness ratio; calculate the transmission path of service flow fij and the scheduling period T of all flows; Step 2: Allocate resources for services with strict latency requirements and lock in the time window for resource usage; Step 3: Allocate resources for services with general latency requirements; Step 4: Output the results, which include time window data and business deployment statistics. In step 2, after preprocessing the stream, a two-step heuristic search is used to allocate resources for services with strict latency requirements using offset; in step 3, resources are allocated for services with general latency requirements using offset + buffer + window merging. Step 2 includes the following sub-steps: Step 2.1: Sort the business flows in the Fstrict set in ascending order by fij.e2e; clear the GCL data structure; Step 2.2: Calculate the time window value of the data frame on the output port / Queue for each business flow passing through each node along the path direction in the Fstrict set; Step 2 also includes the following sub-steps: Step 2.3: Calculate whether the total latency through multiple nodes and links meets the requirements; Sub-step 2.3 also includes: When the total latency meets the requirements, it is determined whether the time window occupied by the business flow in the Fstrict set on each node Q conflicts with other flows. If there is no conflict, the time to pass through the Queue is recorded; if there is a conflict, a conflict avoidance algorithm is used to avoid the conflict. The conflict avoidance algorithm includes: Based on the time windows allocated to node Q within a scheduling cycle and the flow fi occupying those time windows, the process is as follows: If the fi time window partially overlaps with an existing time window: update fi.offset = fi.offset + overlap length. After the offset is increased, the time window is lengthened. After the time window is lengthened, if it overlaps with the next time window, the resources of the next time window are reclaimed, and its corresponding flow fj is put back into Fstrict to participate in the next allocation calculation. Recalculate the time window conflict situation of the flow fi in each Q along the path direction. If there is a conflict, restart the avoidance algorithm. If there is still a conflict when fi.offset reaches offsetMAX, add fj to the Fnormal set. If there is no conflict, end the current fi calculation. Step 3 includes the following sub-steps: Step 3.1: For services with general latency requirements, after calculating the service path SP, calculate the latency limit for each service flow and each hop. Step 3.2: Sort each Q in descending order of the total length of the time windows it contains; Step 3.3: Determine the relationship between the window time of the general latency requirement service appearing in a certain Q and the window of the deployed service, and update the window information of each Q according to the relationship.

2. The method for allocating TSN network resources according to claim 1, characterized in that, Step 1 includes the following sub-steps: Step 1.1: Based on fij.e2e, classify the service flow according to the latency requirements, where: Fstrict represents services with strict latency requirements, and Fnormal represents services with general latency requirements; Step 1.2: Based on i, j and the network topology, calculate the shortest path SP of fij; Step 1.3: Calculate the least common multiple based on fij.period of all service flows to obtain the scheduling period T under the current service flow; Step 1.4: Calculate the number of repetitions M of each fij within the scheduling period T; treat each repetition as a new service flow and add it to the ST service flow set.

3. The method for allocating TSN network resources according to claim 1, characterized in that, The time window data includes the time window data of node Q, which includes: The Talker node transmit time offset fi.offset, the node receive time policy, the GCL of each switching node's output port, the forwarding MAC table of each node based on the listener MAC address, and the node flow classification table based on source and destination MAC addresses, VID and / or priority encoding.

4. The method for allocating TSN network resources according to claim 1, characterized in that, The business deployment statistics include: Successful deployment ID, failure ID; Node Q time window information statistics and GCL length information.

Images