Dynamic flow rate control method and device between heterogeneous network element devices

Abstract

The invention discloses a dynamic flow rate control method and device between heterogeneous network element devices, and relates to the technical field of heterogeneous network element device collaboration.The method comprises the steps that a control plane analyzes network flow duration and packet counting features through a flow feature perception model, and a flow state probability model is established to dynamically calculate the processing priority of each flow; in combination with the model and the real-time processing capacity and load of downstream equipment, a data plane dynamically adjusts the flow transmission rate by adopting a probability token bucket mechanism, and global statistical information is used for guiding a local decision; triggering statistical updating by a data packet arrival event, checking whether a preset period is exceeded or not, and synchronously updating traffic characteristics; and after reasoning analysis of the heterogeneous device coprocessing sub-model, a result is fed back to a control plane, token bucket parameters and model weights are dynamically adjusted, and closed-loop feedback of flow distribution is realized. According to the method, the overall throughput performance and the load balancing capacity in a high-speed network environment are remarkably enhanced.

Description

A method and apparatus for dynamic flow rate control between heterogeneous network elements Technical Field

[0001] This invention relates to the field of heterogeneous network element collaboration technology, and in particular to a method and apparatus for dynamic flow rate control among heterogeneous network element devices. Background Technology

[0002] With the rapid development of applications such as cloud computing, big data analytics, and artificial intelligence, increasingly heterogeneous network elements are appearing in modern network architectures. These devices differ significantly in processing power, functional characteristics, and deployment locations. For example, in data center networks, high-performance switches can achieve Tbps-level forwarding capabilities, while dedicated network function devices such as FPGA accelerator cards, smart network interface cards (NICs), and edge computing nodes typically only achieve hundreds of Gbps-level processing capabilities. Simultaneously, different types of network applications have significantly different traffic transmission requirements. Storage and data mining applications require high network bandwidth to meet high throughput demands, while applications such as network content retrieval and machine learning require low communication latency. This significant difference in processing power and the diversification of application requirements present new challenges to network traffic management, and traditional fixed-rate traffic forwarding methods cannot effectively address the performance gap between heterogeneous devices.

[0003] In traditional network architectures, traffic transmission between devices typically employs static configuration or simple traffic shaping mechanisms. When the forwarding capacity of an upstream device far exceeds the processing capacity of a downstream device, congestion can easily occur at the downstream device, thereby affecting the performance of the entire network. Existing traffic control methods mainly include the following: 1) Backpressure-based traffic control, which adjusts the traffic rate by having the downstream device send a congestion signal to the upstream device. However, this method has a large response delay and is difficult to adapt to dynamic traffic changes in high-speed network environments; 2) Token bucket-based static traffic shaping, which controls traffic transmission through preset rate limits, but cannot make real-time dynamic adjustments based on network conditions and device load; 3) Priority queue-based traffic scheduling, which achieves differentiated services by setting different priorities for different types of traffic, but lacks precise consideration and modeling of the differences in processing capabilities among heterogeneous devices.

[0004] These traditional methods have several limitations: 1) They lack accurate modeling and perception of the differences in processing capabilities among heterogeneous devices, making it impossible to intelligently allocate traffic based on the actual processing capacity and current load status of the devices; 2) The traffic control strategies are relatively static, making it difficult to adaptively adjust according to the spatiotemporal changes in network traffic characteristics and the dynamic evolution of device load status; 3) They lack a global perspective on traffic optimization strategies, with each device controlling traffic independently, failing to achieve coordinated optimization of overall network performance. Especially in scenarios with complex and variable network traffic patterns and mixed deployment of heterogeneous devices, existing methods struggle to maximize the utilization of heterogeneous device processing resources while ensuring the quality of service for critical traffic, easily leading to resource waste on high-performance devices or overload and congestion on low-performance devices.

[0005] In recent years, the development of programmable network technology has provided a new technical approach to solving the flow control problem between heterogeneous devices. Programmable data plane technology allows for flexible configuration and programming of the data plane functions in network devices, enabling networks to quickly adapt to different application scenarios and needs. By introducing high-speed programmable switching chips into network devices, high-speed processing and forwarding of data packets have been achieved, providing network devices with higher performance and throughput. Simultaneously, software-defined networking technology can implement global flow scheduling and optimization strategies, providing a technical foundation for collaborative work between heterogeneous devices. However, current programmable switching chip hardware resources are limited, and achieving intelligent dynamic control of network traffic solely through programmable switches on the data plane still faces challenges. On the one hand, the number of static random access memory and register resources in programmable switching chips is limited, making it difficult to store complex flow control algorithms and state information; on the other hand, the types of operations supported by programmable switching chips are limited, making high-precision calculations and modeling difficult. Summary of the Invention

[0006] The main objective of this invention is to provide a method for dynamic flow rate control between heterogeneous network elements.

[0007] Another objective of this invention is to provide a dynamic flow rate control device for heterogeneous network element devices.

[0008] The third objective of this invention is to provide an electronic device.

[0009] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above objectives, a first aspect of the present invention proposes a dynamic traffic rate control method among heterogeneous network elements, comprising: S1, analyzing the duration and packet count characteristics of network flows through a traffic characteristic perception model in the control plane, and establishing a probabilistic model based on flow state to dynamically calculate the processing priority of each flow; S2, based on the probabilistic model and the real-time processing capacity and load status of downstream devices, dynamically adjusting the traffic transmission rate in the data plane using a probabilistic token bucket mechanism, and guiding local rate decisions through global traffic statistics; S3, triggering the update of traffic statistics information using packet arrival events, checking whether the current time exceeds a preset statistical period, and updating traffic characteristic information; S4, receiving characteristic data and performing inference analysis through a collaborative processing sub-model of heterogeneous devices, and feeding back the processing results to the control plane to dynamically adjust the token bucket parameters and probabilistic model weight coefficients, thereby achieving closed-loop feedback of traffic allocation.

[0011] Optionally, the step of analyzing the duration and packet count characteristics of network flows through a control plane traffic characteristic perception model, and establishing a probabilistic model based on flow state to dynamically calculate the processing priority of each flow, further includes: using a truncated hash value of a 5-tuple as a unique flow identifier, recording the flow hash value, backlog packet count, backlog timestamp, classification result, and buffer index to distinguish different flows and maintain a flow information table; based on the elapsed time of flow i... and number of data packets The processing priority is calculated using a piecewise probability model, and the probability model formula is as follows: Where N is the global flow count, Q is the global packet rate, and V is the token generation rate. This represents the average packet rate of stream i.

[0012] Optionally, the step of dynamically adjusting the traffic transmission rate in the data plane using a probabilistic token bucket mechanism based on the probability model and the real-time processing capacity and load status of downstream devices, and guiding local rate decisions through global traffic statistics, further includes: calculating the token generation rate using the following formula: Where V represents the token bucket generation rate, B represents the communication bandwidth between heterogeneous devices, F represents the device processing frequency, and W represents the feature vector width; the token bucket generation rate is determined by pre-computing a lookup table. and The range is evenly partitioned to achieve efficient projection of probability values ​​(prob), avoiding the direct calculation of complex probability expressions in the data plane.

[0013] Optionally, the step of triggering the update of traffic statistics information using the arrival of data packets, checking whether the current time exceeds a preset statistical period, and updating the traffic characteristic information further includes: when it is detected that the current time exceeds the preset statistical period... When necessary, update the traffic statistics registers that need to be reset; otherwise, update the corresponding feature information based on whether the data packet belongs to a new flow. The statistics update is triggered by the arrival of the data packet, avoiding the resource overhead problem caused by updating all registers at the same time.

[0014] Optionally, the step of receiving feature data and performing inference analysis through the collaborative processing sub-model of heterogeneous devices, and feeding back the processing results to the control plane to dynamically adjust the token bucket parameters and probability model weight coefficients to achieve closed-loop feedback of traffic allocation, further includes: the system periodically monitoring the performance indicators of heterogeneous devices, including processing latency, queue length, and throughput, and dynamically adjusting the token bucket parameters and weight coefficients in the probability model based on the monitoring results; when a decrease in the processing capacity of a heterogeneous device is detected, the traffic allocated to that device is reduced, and part of the load is redistributed to other available devices.

[0015] Optionally, it also includes: converting data packets into feature vector form through feature packet header processing technology, the device extracting the feature packet header part and inputting it into the processing model for analysis, writing the result into the result field of the feature packet header after processing, and forwarding the complete data packet to the programmable switch to update the classification information and processing strategy.

[0016] To achieve the above objectives, a second aspect of the present invention proposes a dynamic traffic rate control device among heterogeneous network elements, comprising: a traffic feature sensing module, used to analyze the duration and packet count characteristics of network flows through a traffic feature sensing model in the control plane, and establish a probability model based on flow state to dynamically calculate the processing priority of each flow; a dynamic rate adjustment module, used to dynamically adjust the traffic transmission rate in the data plane using a probabilistic token bucket mechanism based on the probability model and the real-time processing capacity and load status of downstream devices, and guide local rate decisions through global traffic statistics; an event-triggered statistics update module, used to trigger the update of traffic statistics information using data packet arrival events, check whether the current time exceeds a preset statistical period and update the traffic feature information, avoiding resource overhead caused by timer triggering; and a closed-loop feedback parameter adjustment module, which receives feature data and performs inference analysis through a collaborative processing sub-model of heterogeneous devices, and feeds back the processing results to the control plane to dynamically adjust the token bucket parameters and probability model weight coefficients, thereby realizing closed-loop feedback of traffic allocation.

[0017] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0018] To achieve the above objectives, a third aspect of this application provides an electronic device, including a processor and a memory; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, for implementing a dynamic flow rate control method between heterogeneous network element devices as described in the first aspect embodiment.

[0019] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a dynamic flow rate control method between heterogeneous network elements as described in the first aspect embodiment.

[0020] The embodiments of this invention offer the following advantages: By designing a traffic characteristic perception model and an adaptive rate adjustment model, intelligent dynamic control of traffic transmission between heterogeneous devices is achieved. The traffic characteristic perception model operates in the control plane, establishing a probabilistic model based on flow state by analyzing characteristic parameters such as network flow duration and packet count, dynamically identifying the importance and processing requirements of different flows. The adaptive rate adjustment model combines the data plane and the control plane, dynamically adjusting the traffic transmission rate using a probabilistic token bucket mechanism based on the processing capacity of downstream devices and the current load state. It guides local decision-making with global statistical information, achieving intelligent traffic allocation between heterogeneous devices. Through collaborative operation, this invention can automatically identify network traffic patterns, dynamically match the processing capabilities of heterogeneous devices, maximize the utilization of network resources, and significantly improve overall performance in heterogeneous network environments. It is highly suitable for application in complex network scenarios where heterogeneous devices are deployed in combination, such as data center networks and edge computing networks. Attached Figure Description

[0021] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: FIG1 is a flowchart of a dynamic flow rate control method between heterogeneous network element devices provided by an embodiment of the present invention; FIG2 is an overall working principle diagram provided by an embodiment of the present invention; FIG3 is a structural diagram of a dynamic flow rate control device between heterogeneous network element devices provided by an embodiment of the present invention. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] The following description, with reference to the accompanying drawings, describes a method and apparatus for dynamic flow rate control between heterogeneous network element devices according to an embodiment of the present invention.

[0025] Example 1 This embodiment of the invention provides a dynamic traffic rate control method between heterogeneous network element devices. Figure 1 is a flowchart illustrating the dynamic traffic rate control method between heterogeneous network element devices provided in this embodiment of the invention. As shown in Figures 1 and 2, the method includes the following steps: Step S1, by analyzing the duration and packet count characteristics of network flows through a traffic characteristic perception model of the control plane, a probabilistic model based on flow state is established to dynamically calculate the processing priority of each flow.

[0026] In one embodiment of the present invention, the core objective of this step is to analyze the key characteristics of network flows and establish a probabilistic model through a traffic feature perception model that works collaboratively between the control plane and the data plane. This allows for the dynamic calculation of the processing priority of each flow, providing a basis for subsequent rate adjustment decisions. In heterogeneous network environments, the processing capabilities of different devices vary significantly. The traffic feature perception model designed in this embodiment can dynamically adapt to the processing capabilities of heterogeneous devices. Its included flow tracing sub-model, probabilistic modeling sub-model, and feature caching sub-model work collaboratively in this step.

[0027] Specifically, the flow tracing sub-model is deployed on the data plane of the programmable switch and is responsible for uniquely identifying network flows and maintaining flow information in real time. In this embodiment, different flows are distinguished by a "five-tuple" (source IP address, source port, destination IP address, destination port, protocol type), and packets with the same five-tuple are determined to be the same flow. To improve identification efficiency and resource utilization, the sub-model uses the truncated hash value of the five-tuple as the unique flow identifier in the flow information table. The information recorded in the table includes: flow hash value (used to identify new flow arrivals and handle table conflicts), backlog packet count bklog_n (the number of packets within the tracking feature transmission interval), backlog timestamp bklog_t (recording the feature transmission interval), classification result (used for custom processing), and buffer index buff_idx (pointing to the feature cache location). Each time a preset statistical period passes... The relevant statistical information in the flow information table will be updated synchronously, providing real-time data support for feature analysis.

[0028] In this embodiment of the invention, the probabilistic modeling sub-model is implemented collaboratively by the data plane and control plane of the programmable switch, and a probability function P( is constructed based on the real-time global traffic statistics collected by the flow tracing sub-model.) , ),in This represents the elapsed time since the last transmission feature of stream i. This represents the amount of data packets transmitted by flow i during this period. The probabilistic model in this application is based on two main criteria to establish a segmented model: Criterion 1 is that when all flow rates are equal, each flow receives tokens at an average interval of N / V (N is the global flow count, and V is the token generation rate); Criterion 2 is that in heterogeneous flow rate scenarios, token allocation is proportional to the flow rate. = / Let Q represent the average packet rate of flow i, and Q be the global packet rate. The specific probability model formula is as follows: .

[0029] It's clear that the feature caching sub-model is also deployed on the data plane, allocating a dedicated circular buffer for each network flow to temporarily store feature vectors until they are processed by the model engine. The buffer stores the 8-dimensional features of preceding data packets. - (Including packet interval, length, etc.) and current packet characteristics (Stored in metadata), when the buffer reaches its capacity, new entries are used to overwrite old features in a cyclic first-in-first-out manner to ensure that the recent state of the flow is reflected. In this embodiment, the flow feature perception model of the control plane is based on the feature data and probability model provided by the above sub-model to dynamically calculate the processing priority of each flow.

[0030] Step S2: Based on the probability model and the real-time processing capabilities and load status of downstream devices, a probabilistic token bucket mechanism is used in the data plane to dynamically adjust the traffic transmission rate, and global traffic statistics are used to guide local rate decisions.

[0031] In one embodiment of the present invention, this step, based on the probabilistic model obtained in step S1 and combined with the real-time status of downstream devices, dynamically adjusts the traffic transmission rate through an adaptive rate adjustment model of the data plane. The token bucket management sub-model and the probabilistic rate control sub-model included in this model play a core role in this step, while the device collaborative processing sub-model prepares for bottleneck handling. In this embodiment, rate adjustment is not a local, independent decision, but rather guides local operations through global traffic statistics to ensure overall load balancing.

[0032] The token bucket management sub-model first makes token allocation decisions based on a probability model, including token generation, probability judgment, and feature transmission control. Among them, the token generation rate V is a key parameter, which needs to be dynamically set according to the heterogeneous device communication bandwidth B, processing frequency F, and feature vector width W. The specific formula is , and this setting ensures that token generation does not exceed the upper limit of device processing and link transmission capabilities.

[0033] Furthermore, the sub-model maintains the token bucket status variables in real time, including the time of the last packet arrival , the current bucket capacity, etc.; when a packet arrives, first calculate the time interval to supplement tokens, and then generate a random number rand and compare it with the probability value prob in the pre-computed lookup table - since the data plane cannot directly calculate complex probability expressions, the embodiments of this application will and range evenly partitioned, and build a mapping table of all value pairs to the probability of [0,1] in advance to achieve efficient retrieval.

[0034] In the embodiments of the present invention, the probabilistic rate control sub-model implements probabilistic token allocation based on the above token bucket mechanism to complete precise control of the traffic rate. Its core logic is: if a packet is probabilistically selected (rand < prob) and the token bucket capacity is sufficient, deduct the corresponding token cost and transmit the feature vector; if the selection fails or the tokens are insufficient, only update the token count according to gap. This mechanism probabilistically rejects high-speed flow token requests, reserves transmission opportunities for low-speed flows, and realizes fairness among heterogeneous traffic. At the same time, in the embodiments of this application, the sub-model will collect the processing capabilities and load status of downstream devices in real time (such as the task queue length, computing load, etc.). If it is found that the device has a processing bottleneck, it will immediately trigger the work preparation of the device collaborative processing sub-model to lay a foundation for subsequent closed-loop regulation.

[0035] Step S3, trigger the update of traffic statistical information using the packet arrival event, check whether the current time exceeds the preset statistical period, and update the traffic feature information.

[0036] In an embodiment of the present invention, step S3 solves the resource constraint problem of programmable switch traffic statistical information update and联动 step S1's flow tracking sub-model to complete information synchronization. Due to the hardware characteristics of programmable switches, only one register can be operated in each processing stage, and all registers cannot be reset simultaneously within the statistical period; and the reset delay of the control plane for data plane registers exceeds 1ms, which cannot meet the real-time requirement. Therefore, the embodiments of this application propose a mechanism to trigger the update of traffic statistical information using the packet arrival event.

[0037] Specifically, whenever a data packet arrives, the system first checks whether the current time exceeds the preset statistical period. If the time limit has been exceeded, the corresponding flow statistics register (including the backlog packet count bklog_n, backlog timestamp bklog_t, etc. in the flow information table of the flow tracing submodel) is updated and reset to ensure that the statistical data reflects the flow status within the new period. If the time limit has not been exceeded, the flow tracing submodel's five-tuple identification mechanism determines whether the flow to which the data packet belongs is a new flow. If it is a new flow, the flow information table is initialized, a dedicated circular buffer (associated feature cache submodel) is allocated, and relevant identification information is recorded. If it is an existing flow, the packet count, timestamp, and other feature information in the flow information table are updated, and the features of the current data packet are recorded. The data is written to the corresponding buffer, and the buffer data of the feature cache sub-model is updated synchronously. This embodiment of the application, through this triggered update mechanism, avoids register operation conflicts and ensures the real-time accuracy of traffic feature information, providing reliable data support for the probability model optimization in step S1 and the rate adjustment in step S2.

[0038] Step S4: The feature data is received and inference analysis is performed through the collaborative processing sub-model of heterogeneous devices. The processing results are fed back to the control plane to dynamically adjust the token bucket parameters and probability model weight coefficients, thereby realizing closed-loop feedback of traffic allocation.

[0039] In one embodiment of the present invention, this step establishes a closed-loop feedback mechanism by linking the device collaborative processing sub-model in the adaptive rate adjustment model with the traffic feature perception model in step S1, thereby achieving dynamic optimization of traffic scheduling parameters. The device collaborative processing sub-model is deployed on heterogeneous devices such as FPGAs and is responsible for receiving feature data, inference analysis, and result feedback; it is the core execution unit for closed-loop adjustment.

[0040] Specifically, firstly, the programmable switch data plane stores the feature vectors (in the feature cache sub-model of step S1) - The data packet is transmitted to the collaborative processing sub-model of heterogeneous devices via a dedicated communication link. After receiving the data packet, the heterogeneous devices first convert it from a data packet format to a feature vector form usable for inference, and extract the feature packet header portion to input into the processing model for analysis—the analysis includes the current traffic pattern matching degree, downstream device load change trend, and the rationality of the current rate adjustment strategy. After processing, the sub-model writes the results (including token bucket parameter adjustment suggestions and probability model weight coefficient correction values) into the result field of the feature packet header, and then forwards the complete data packet back to the programmable switch.

[0041] In this embodiment of the invention, after receiving the processing result, the programmable switch immediately feeds it back to the control plane, triggering parameter optimization throughout the entire process: the control plane updates the probability model weight coefficients of the probability modeling sub-model in step S1 based on the result (e.g., adjusting the weight ratio of global flow count N and global packet rate Q), and optimizes the priority calculation logic; simultaneously, it updates parameters such as the token generation rate V and token bucket capacity of the token bucket management sub-model in step S2, and corrects the rate adjustment strategy. Furthermore, the control plane also adjusts the statistical period in step S3 based on the result. This allows the information update frequency to adapt to the current traffic characteristics. The embodiments of this application achieve dynamic adaptive traffic allocation through this closed-loop mechanism of "feature perception - rate adjustment - status feedback - parameter optimization," ensuring that the system maintains optimal operating status in scenarios with performance fluctuations of heterogeneous devices and changes in traffic patterns.

[0042] Furthermore, based on step S4, this embodiment of the application constructs a heterogeneous device performance adaptive mechanism, which works in conjunction with the device collaborative processing sub-model to achieve dynamic load balancing. The system periodically (e.g., every 100ms) collects performance metrics such as processing latency, queue length, and throughput of heterogeneous devices through the device collaborative processing sub-model. When a device's processing capacity is detected to be reduced (e.g., latency exceeds a threshold, queue is full), the control plane reduces the flow into that device by adjusting the token bucket generation rate V and distributes some of the load to other low-load devices. Once the device's performance recovers, normal load distribution is gradually restored. This mechanism ensures that the system as a whole maintains load balancing and efficient operation even when device performance changes dynamically.

[0043] Example 2: This embodiment of the invention provides a dynamic traffic rate control device between heterogeneous network element devices. Figure 3 is a flowchart illustrating the dynamic traffic rate control device between heterogeneous network element devices provided in this embodiment. As shown in Figure 3, the device includes: a traffic feature sensing module 100, used to analyze the duration and packet count characteristics of network flows through a traffic feature sensing model in the control plane, and establish a probability model based on flow state to dynamically calculate the processing priority of each flow; a dynamic rate adjustment module 200, used to dynamically adjust the traffic transmission rate in the data plane using a probabilistic token bucket mechanism based on the probability model and the real-time processing capacity and load status of downstream devices, and guide local rate decisions through global traffic statistics; an event-triggered statistics update module 300, used to trigger the update of traffic statistics information using data packet arrival events, check whether the current time exceeds a preset statistical period and update traffic feature information to avoid resource overhead caused by timer triggering; and a closed-loop feedback parameter adjustment module 400, which receives feature data and performs inference analysis through a collaborative processing sub-model of heterogeneous devices, and feeds back the processing results to the control plane to dynamically adjust the token bucket parameters and probability model weight coefficients to achieve closed-loop feedback of traffic allocation.

[0044] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0045] Example 3 To implement the method of the above embodiments, the present invention also provides an electronic device, which includes a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the various steps of the method described above.

[0046] Example 4 In order to implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0048] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0049] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for dynamic flow rate control among heterogeneous network element devices, characterized in that, include: S1. Analyze the duration and packet count characteristics of network flows using the traffic characteristic perception model in the control plane, and establish a probabilistic model based on flow state to dynamically calculate the processing priority of each flow; S2. Based on the probabilistic model and the real-time processing capacity and load status of downstream devices, dynamically adjust the traffic transmission rate in the data plane using a probabilistic token bucket mechanism, and guide local rate decisions through global traffic statistics; S3. Use packet arrival events to trigger updates to traffic statistics, check whether the current time exceeds the preset statistical period, and update the traffic characteristic information; S4. Receive characteristic data and perform inference analysis through the collaborative processing sub-model of heterogeneous devices, and feed the processing results back to the control plane to dynamically adjust the token bucket parameters and probabilistic model weight coefficients, achieving closed-loop feedback for traffic allocation.

2. The method according to claim 1, characterized in that, The method of analyzing network flow duration and packet count characteristics through a control plane traffic characteristic perception model, and establishing a probabilistic model based on flow state to dynamically calculate the processing priority of each flow, also includes: using a 5-tuple truncated hash value as a unique flow identifier, recording the flow hash value, backlog packet count, backlog timestamp, classification result, and buffer index to distinguish different flows and maintain a flow information table; based on the elapsed time of flow i... and number of data packets The processing priority is calculated using a segmented probability model.

3. The method according to claim 1, characterized in that, Based on the aforementioned probability model and the real-time processing capabilities and load status of downstream devices, the method of dynamically adjusting the traffic transmission rate in the data plane using a probabilistic token bucket mechanism, and guiding local rate decisions with global traffic statistics, further includes: calculating the token generation rate using the following formula: Where V represents the token bucket generation rate, B represents the communication bandwidth between heterogeneous devices, F represents the device processing frequency, and W represents the feature vector width; the pre-computed lookup table will then... and The range is evenly partitioned to achieve efficient projection of probability values ​​(prob), avoiding the direct calculation of complex probability expressions in the data plane.

4. The method according to claim 1, characterized in that, The method of triggering traffic statistics updates using data packet arrival events, checking whether the current time exceeds a preset statistical period, and updating traffic characteristic information further includes: when it is detected that the current time exceeds a preset statistical period... When necessary, update the traffic statistics registers that need to be reset; otherwise, update the corresponding feature information based on whether the data packet belongs to a new flow. The statistics update is triggered by the arrival of the data packet, avoiding the resource overhead problem caused by updating all registers at the same time.

5. The method according to claim 1, characterized in that, The process of receiving feature data and performing inference analysis through the collaborative processing sub-model of heterogeneous devices, and feeding the processing results back to the control plane to dynamically adjust the token bucket parameters and probability model weight coefficients to achieve closed-loop feedback of traffic allocation, also includes: the system periodically monitoring the performance indicators of heterogeneous devices, including processing latency, queue length and throughput, and dynamically adjusting the token bucket parameters and weight coefficients in the probability model according to the monitoring results; when a decrease in the processing capacity of a heterogeneous device is detected, the traffic allocated to that device is reduced and some of the load is redistributed to other available devices.

6. The method according to claim 1, characterized in that, Also includes: The data packet is converted into a feature vector form through feature packet header processing technology. The device extracts the feature packet header part and inputs it into the processing model for analysis. After processing, the result is written into the result field of the feature packet header, and the complete data packet is forwarded to the programmable switch to update the classification information and processing strategy.

7. A dynamic flow rate control device for heterogeneous network element devices, characterized in that, include: The traffic characteristic perception module is used to analyze the duration and packet count characteristics of network flows through the traffic characteristic perception model of the control plane, and establish a probability model based on flow state to dynamically calculate the processing priority of each flow; the dynamic rate adjustment module is used to dynamically adjust the traffic transmission rate in the data plane using a probabilistic token bucket mechanism based on the probability model and the real-time processing capacity and load status of downstream devices, and guide local rate decisions through global traffic statistics. The event-triggered statistics update module is used to trigger the update of traffic statistics information by utilizing the arrival of data packets. It checks whether the current time exceeds the preset statistical period and updates the traffic characteristic information to avoid resource overhead caused by timer triggering. The closed-loop feedback parameter adjustment module receives feature data and performs inference analysis through the collaborative processing sub-model of heterogeneous devices. It feeds back the processing results to the control plane to dynamically adjust the token bucket parameters and probability model weight coefficients to achieve closed-loop feedback of traffic allocation.

8. The apparatus according to claim 7, characterized in that, The traffic feature acquisition and priority modeling module is also used to: use the truncated hash value of the five-tuple as a unique flow identifier, record the flow hash value, backlog packet count, backlog timestamp, classification result and buffer index, so as to distinguish different traffic and maintain the flow information table; Based on the elapsed time of stream i and number of data packets The processing priority is calculated using a segmented probability model.

9. An electronic device, characterized in that, It includes a processor and a memory; wherein the processor runs a program corresponding to the executable program code stored in the memory to implement the method as described in any one of claims 1-6.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images