Non-central-node type distributed flow rate-limiting calculation method and non-central-node type distributed flow rate-limiting calculation device

Abstract

The invention discloses a non-central node type distributed flow rate limiting calculation method and device, and the method comprises the steps: defining a distributed flow rate limiting problem based on a local rate limiting value, an input rate and an output rate of each distributed network node, and the sum of global rate limiting values and output rates of all distributed network nodes, defining a target local speed limit value based on a fairness hypothesis; calculating a transfer matrix of node traffic distribution under a target local speed limit value which satisfies fairness hypothesis definition based on a distributed traffic speed limit problem, and updating a local speed limit value of a current distributed network node according to the transfer matrix and a local speed limit value of a neighbor node; and when the node input rate changes, the transfer matrix is recalculated, and the local speed limit value of each node is updated in real time. According to the invention, rapid convergence, high accuracy and wide universality of the distributed speed-limiting algorithm are realized, resource use optimization is facilitated, and service quality and user experience are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed speed limit calculation, and particularly relates to a non - central - node - type distributed traffic speed limit calculation method and device. Background Art

[0002] As cloud computing plays an increasingly important role in modern information technology infrastructure, cloud service providers (CSPs) face numerous challenges in managing the growing and diverse network traffic. This traffic originates from various applications and services hosted within data centers, and these resources are typically shared among multiple tenants. Different tenants have different bandwidth requirements. Therefore, in order to effectively allocate resources, ensure network stability, and meet service - level agreements (SLAs), cloud service providers need to implement traffic speed limit policies that can not only optimize resource usage but also enhance the overall service quality and ensure the user experience is guaranteed.

[0003] Traditional traffic speed limit mainly relies on a centralized architecture, where the speed limit table for each tenant is centrally stored on a speed limit node, and the central controller is responsible for forwarding tenant traffic to specific nodes. However, with the rapid development of network technology, especially the wide application of technologies such as high - speed links, multi - path transmission protocols, and network redundancy, cloud service providers have started to deploy network functions (such as NAT, IDS) in cluster configurations and process various traffic on distributed nodes. This transformation has forced them to adopt a distributed speed limit (DRL) strategy, in which speed - limit - related configurations are scattered across various distributed nodes, and traffic speed limit decisions are made locally on multiple nodes. This method not only improves processing efficiency but also enhances the flexibility and scalability of the system.

[0004] The research on distributed speed limit algorithms can be divided into central - node - type algorithms (CN) and non - central - node - type algorithms (NCN). The difference between them lies in whether the distributed nodes passively receive or actively estimate global parameters. Central - node - type algorithms have some form of central node (either physically or logically) responsible for collecting information from each node and configuring speed limit parameters. However, this method has problems such as large communication overhead, complex synchronization, and the central node may become a bottleneck.

[0005] In contrast, non - central - node - type algorithms are more flexible and efficient. Among them, any distributed node does not need to know or estimate global quantities, and only needs information exchange between neighbors to reasonably allocate speed limit values. This method can significantly reduce communication overhead because each node only needs to process information from neighbors instead of communicating with all nodes. In addition, each node only needs to process local information related to itself instead of global information, so non - central - node - type algorithms can also reduce the computational amount of each node.

[0006] However, despite the many advantages of non - central - node algorithms, there are still some problems with existing non - central - node distributed rate - limiting algorithms. For example, the C3P (Cloud Control with Constant Probabilities) algorithm, as a non - central - node distributed rate - limiting algorithm, can be applied to different granularities and can measure the packet loss rate. However, when applying the C3P algorithm under other granularities, the convergence of the algorithm cannot be guaranteed. For example, in the scenario of RDMA lossless network, there is no formula similar to TCP Square Root to measure the relationship between the packet loss rate and the round - trip time (RTT). In addition, although the C3P algorithm gives a proof of convergence under flow granularity, its convergence speed is not guaranteed and depends on the specific values of the algorithm hyperparameters. Since its value range is limited, it is difficult to obtain the fastest convergence speed.

[0007] In the non - central - node D2R2 (Distributed Deficit Round Robin) algorithm, a more widely used rate - limiting ratio is used instead of the packet loss rate, enabling it to be applied to different granularities. However, when the rate - limiting ratio is small, the deviation of the algorithm will increase sharply, resulting in violent fluctuations in the global rate - limiting value and making it difficult to achieve precise traffic control. This is because the D2R2 algorithm takes the fairness assumption determined in vector form between nodes as the primary goal rather than keeping the global rate - limiting value constant. In addition, most of the existing algorithms are protocol - related and not very general.

[0008] In summary, the existing non - central - node distributed rate - limiting algorithms still have deficiencies in terms of convergence, accuracy, and generality. Therefore, it is necessary to study a more efficient, accurate, and general non - central - node distributed traffic rate - limiting calculation method. Summary of the Invention

[0009] In view of the above, the object of the present invention is to provide a non - central - node distributed traffic rate - limiting calculation method and device, which can realize real - time update of the local rate - limiting values of non - centralized distributed network nodes under the distributed traffic rate - limiting target. It not only has better accuracy and convergence performance, but is also protocol - independent and has better applicability, which helps cloud service providers better manage network traffic, improve resource utilization rate and service quality, and promote the further development and application of cloud computing technology.

[0010] To achieve the above - mentioned invention object, the technical solutions provided by the present invention are as follows:

[0011] In the first aspect, a non - central - node distributed traffic rate - limiting calculation method provided by an embodiment of the present invention includes the following steps:

[0012] Define the distributed traffic rate limiting problem based on the local rate limiting values, input rates, and output rates of each distributed network node, as well as the global rate limiting value and the total output rate of all distributed network nodes, and define the target local rate limiting value based on the fairness assumption;

[0013] Based on the distributed traffic rate limiting problem, calculate the transition matrix of node traffic allocation through the distributed Metropolis-Hastings algorithm under the target local rate limiting value defined by the fairness assumption, and update the local rate limiting value of the current distributed network node according to the transition matrix and the local rate limiting values of neighboring nodes;

[0014] When the input rate of the distributed network node changes, recalculate the transition matrix and update the local rate limiting value of each distributed network node in real time.

[0015] Specifically, the definition of the distributed traffic rate limiting problem is as follows:

[0016] For n distributed network nodes, let a i , x i , y i be the local rate limiting value, input rate, and output rate of the i-th distributed network node respectively. The global rate limiting value of all distributed network nodes a = a 1 + a 2 + … + a i + … + a n , and the total output rate of all distributed network nodes y = y 1 + y 2 + … + y i + … + y n , then the distributed traffic rate limiting objective is expressed as:

[0017]

[0018] Among them, the first line means that a distributed rate limiter will be independently implemented on each distributed network node, and the second line means that in any case, the global rate limiting value does not exceed the preset rate limiting value size.

[0019] Specifically, take the limiting distribution of the Markov chain as the convergence target of the normalized local rate limiting value, and determine this limiting distribution through the preset fairness assumption. The fairness assumption is expressed as:

[0020]

[0021] Meet the requirements of different systems by setting optional weights and exponents, which is also called the proportional fairness assumption. Among them, π iDenote the convergence target of the normalized local speed limit value of the \(i\)-th distributed network node, i.e., the target local speed limit value, which is also the \(i\)-th component of the limiting distribution of the Markov chain, \(\theta\). i Denote the weight hyperparameter corresponding to the \(i\)-th distributed network node, \(x\). i Denote the input rate of the \(i\)-th distributed network node, where the superscript \(\alpha\) is a hyperparameter representing the power.

[0022] Specifically, calculating the transition matrix of node traffic allocation through the distributed Metropolis - Hastings algorithm under the target local speed limit value defined by the fairness assumption includes:

[0023] Obtain the input rate \(x\) of the current distributed network node \(k\). k And traverse the input rates \(x\) of all neighbor nodes \(j\) of \(k\). j As inputs, nodes \(j,k\in i\), and neighbor nodes refer to the nodes directly connected to \(k\) in the network logical topology.

[0024] After the fairness assumption is given, since the denominator in the fairness assumption is the same for all nodes, calculate the ratio of the limiting distributions according to the ratio of the input rates. And initialize the transition matrix.

[0025] Calculate the matrix element \(P_{jk}\) corresponding to the \(j\)-th row and \(k\)-th column in the transition matrix \(P\). jk :[[]]END]]

[0026]

[0027] Where,[[]]END]] Denote the matrix element corresponding to the \(j\)-th row and \(k\)-th column in the initialized transition matrix,[[]]END]] in,[[]]END]] Denote the matrix element corresponding to the \(k\)-th row and \(j\)-th column in the initialized transition matrix,[[]]END]] in.[[]]END]]

[0028] Calculate the matrix element \(P_{kk}\) corresponding to the \(k\)-th row and \(k\)-th column in the transition matrix \(P\). kk :[[]]END]]

[0029]

[0030] Where, the \(+\) sign means that after traversing a new neighbor element each time, add the calculation result on the right to the original value on the left. For the distributed network node \(k\), it only holds all the matrix elements in the \(k\)-th column of the transition matrix \(P\).

[0031] Specifically, updating the local speed limit value of the current distributed network node according to the transition matrix and the local speed limit values of neighbor nodes includes:[[]]END]]

[0032] For the current distributed network node k, its local speed limit value a k is updated as follows:

[0033]

[0034] where a j represents the local speed limit value of neighbor node j, (j,k)∈E represents that the edge connecting the current distributed network node k and any of its neighbor nodes j belongs to the set E of edges in the network logical topology, and P kj represents the matrix element corresponding to the j-th column and k-th row in the transition matrix P, and the superscript T represents transpose;

[0035] The entire time axis is evenly divided into several rounds. The duration of each round is determined according to the actual physical network configuration. The local speed limit value of the distributed network node is updated once in each round. After several rounds of updates, the deviation between the local speed limit value and the target local speed limit value of each distributed network node is less than the set threshold, which means that the algorithm converges to the target local speed limit value. The entire update process is a Markov chain.

[0036] Specifically, when the input rate of the distributed network node changes, recalculating the transition matrix and updating the local speed limit value of each distributed network node in real time includes:

[0037] If the input rate of the distributed network node changes in the previous round, then when updating in the current round, recalculate the matrix elements in the transition matrix P according to the new input rate that has changed, and update the local speed limit value of the distributed network node according to the matrix elements in the recalculated transition matrix P after the next round.

[0038] In a second aspect, an embodiment of the present invention also provides a non-central node type distributed traffic speed limit calculation device, which is implemented by using the above non-central node type distributed traffic speed limit calculation method, and includes: a speed limit problem definition module, a node speed limit calculation module, and a node speed limit update module;

[0039] The speed limit problem definition module is used to define the distributed traffic speed limit problem based on the local speed limit value, input rate and output rate of each distributed network node, and the global speed limit value and the total output rate of all distributed network nodes, and define the target local speed limit value based on the fairness assumption;

[0040] The node speed limit calculation module is used to calculate the transition matrix of node traffic allocation through the Metropolis-Hastings algorithm based on the distributed traffic speed limit problem under the target local speed limit value defined by the fairness assumption, and update the local speed limit value of the current distributed network node according to the transition matrix and the local speed limit value of the neighbor node;

[0041] The node speed limit update module is used to recalculate the transfer matrix and update the local speed limit value of each distributed network node in real time when the input rate of the distributed network node changes.

[0042] In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory and one or more processors. The memory is used to store a computer program, and the processor is used to implement the above non - central - node - based distributed traffic speed limit calculation method when executing the computer program.

[0043] In a fourth aspect, an embodiment of the present invention further provides a computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a computer, the above non - central - node - based distributed traffic speed limit calculation method is implemented.

[0044] In a fifth aspect, an embodiment of the present invention further provides a computer product, which includes a computer program. When the computer program is executed by a processor, the above non - central - node - based distributed traffic speed limit calculation method is implemented.

[0045] Compared with the prior art, the beneficial effects of the present invention at least include:

[0046] The non - central - node - based distributed traffic speed limit calculation method provided by the present invention has a fast calculation speed, does not rely on a protocol, and can update the node speed limit value in real time according to traffic changes, realizing the fast convergence, high accuracy, and wide generality of the distributed speed limit algorithm. It can effectively balance network traffic, prevent congestion, and improve the overall network performance and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 is a flowchart of the non - central - node - based distributed traffic speed limit calculation method provided by an embodiment of the present invention;

[0049] Figure 2 is a schematic diagram of the interaction process between distributed network nodes provided by an embodiment of the present invention;

[0050] Figure 3 is a schematic diagram of the adjustment process for keeping the constant global speed limit value unchanged provided by an embodiment of the present invention;

[0051] Figure 4 It is a schematic diagram of the evaluation result of the algorithm convergence time CDF provided by an embodiment of the present invention;

[0052] Figure 5 It is a schematic structural diagram of a non - central - node - type distributed traffic rate - limiting calculation device provided by an embodiment of the present invention. Detailed implementation manners

[0053] To make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific implementation manners described herein are only used to explain the present invention and do not limit the protection scope of the present invention.

[0054] The inventive concept of the present invention is as follows: Aiming at the problems that the non - central - node - type distributed rate - limiting algorithms in the prior art still have deficiencies in aspects such as convergence, accuracy, and generality, embodiments of the present invention provide a non - central - node - type distributed traffic rate - limiting calculation method and device. Under the condition of meeting the defined distributed traffic rate - limiting objectives, the transfer matrix is calculated through the distributed Metropolis - Hastings algorithm and used to update the local rate - limiting values of distributed network nodes, achieving fast convergence of traffic rate - limiting calculation under the fairness assumption, and being able to update the local rate - limiting values of nodes in real time when the input rate of nodes changes, realizing protocol - independent, accurate, and fast - converging distributed rate - limiting calculation.

[0055] Figure 1 It is a schematic flowchart of a non - central - node - type distributed traffic rate - limiting calculation method provided by an embodiment of the present invention. As Figure 1 shown, the embodiment provides a non - central - node - type distributed traffic rate - limiting calculation method, including the following steps:

[0056] S1. Define a distributed traffic rate - limiting problem based on the local rate - limiting values, input rates, and output rates of each distributed network node, as well as the global rate - limiting values and the total output rate of all distributed network nodes, and define the target local rate - limiting value based on the fairness assumption.,

[0057] S1.1. Define the distributed traffic rate - limiting problem.

[0058] In the embodiment, for n distributed network nodes, let a i , x i , y i be the local rate - limiting value, input rate, and output rate of the i - th distributed network node respectively. The global rate - limiting value a of all distributed network nodes is a = a 1 + a 2 + … + a i + … + a n, the total output rate y of all distributed network nodes is y = y 1 + y 2 + … + y i + … + y n , then the distributed traffic rate limiting problem is expressed as:

[0059]

[0060] Among them, the first line indicates that a distributed rate limiter will be independently implemented on each distributed network node, and the second line indicates that the global rate limiting value does not exceed the preset rate limiting value size under any circumstances.

[0061] S1.2. Define the target local rate limiting value based on the fairness assumption.

[0062] In the embodiment, the limiting distribution of the Markov chain is used as the convergence target of the normalized local rate limiting value, and this limiting distribution is determined through the preset fairness assumption. The limiting distribution of the node input rate satisfies the fairness assumption to achieve the convergence target of the traffic rate limiting calculation. For each node, what the final normalized rate limiting value should be should be determined by the following fairness assumption formula:

[0063]

[0064] By setting optional weights and exponents to meet the requirements of different systems, it is also called the proportional fairness assumption, where π i represents the convergence target of the normalized local rate limiting value of the i-th distributed network node, that is, the target local rate limiting value, and is also the i-th component of the limiting distribution of the Markov chain. θ i represents the weight hyperparameter corresponding to the i-th distributed network node, and x i represents the input rate of the i-th distributed network node, and the superscript α is a hyperparameter representing the power.

[0065] θ i and α can be adjusted according to the differences of specific services. The denominator in the fairness assumption formula is a global information, and the distributed network node cannot know all the parameters of x 1 to x n because a node can only communicate with its neighbors. Otherwise, multiple rounds of communication are required. Therefore, in the embodiment, the matrix elements in the transition matrix need to be calculated through the Metropolis-Hastings algorithm instead of directly calculating the matrix P of the i-th node i .

[0066] S2. Based on the problem of distributed traffic rate limiting, under the target local rate limit value defined by the fairness assumption, calculate the transition matrix of node traffic allocation through the Metropolis-Hastings algorithm, and update the local rate limit value of the current distributed network node according to the transition matrix and the local rate limit values of neighboring nodes.

[0067] S2.1. Calculate the transition matrix of node traffic allocation through the Metropolis-Hastings algorithm.

[0068] (1) Obtain the input rate x of the current distributed network node k k and traverse the input rates x of all neighboring nodes j of k j As inputs, nodes j, k ∈ i, and neighboring nodes represent the nodes directly connected to k in the network logical topology. The logical topology is different from the physical topology. The algorithm only has requirements for the logical topology of the network and has no requirements for the physical topology. Specifically, the algorithm requires that the logical topology of the network is strongly connected. Equivalently, it requires that all nodes in the network are connected and that two directly connected nodes can communicate bidirectionally. After the fairness assumption is given, since the denominator in the fairness assumption is the same for all nodes, the ratio of the limiting distributions is calculated according to the ratio of the input rates and initialize the transition matrix which is set in advance and does not need to be modified again later. Set the matrix element P corresponding to the k-th row and k-th column in the final transition matrix P to be kk initialized as in

[0069] (2) Calculate the matrix element P corresponding to the j-th row and k-th column in the transition matrix P jk :

[0070]

[0071] where represents the matrix element corresponding to the j-th row and k-th column in the initialized transition matrix , represents the matrix element corresponding to the k-th row and j-th column in the initialized transition matrix .

[0072] (3) Calculate the matrix element P corresponding to the k-th row and k-th column in the transition matrix P kk :

[0073]

[0074] Among them, the + = symbol means that after traversing a new neighbor element each time, the calculation result on the right is added to the original value on the left. Through the above steps, all the matrix elements in the k-th column of the transition matrix P can be calculated. For the distributed network node k, it only holds all the matrix elements in the k-th column of the transition matrix P.

[0075] S2.2. Update the local speed limit value of the current distributed network node according to the transition matrix and the local speed limit values of neighbor nodes.

[0076] Let represent the normalized local speed limit values of all distributed network nodes at time t, ‖a (t) ‖ 1 = 1, and the global speed limit value is a = a 1 + a 2 + … + a i + … + a n . In the embodiment, the distributed traffic speed limit calculation method updates the local speed limit value vector according to a (t+1) = P T a (t) . For example, if the total speed limit value is required to be 100 Gbps and there are 10 nodes, and the initial speed limit values at the beginning (time 0) are initialized to 10 Gbps for each node, then the normalized speed limit values are all 0.1.

[0077] For the current distributed network node k, the update method of its local speed limit value a k is as follows:

[0078]

[0079] Among them, a j represents the local speed limit value of the neighbor distributed network node j, (j, k) ∈ E means that the connection edge between the current distributed network node k and any of its neighbor distributed network nodes j belongs to the set E of edges in the entire network node distribution, and P kj represents the matrix element corresponding to the j-th column and k-th row in the transition matrix P, and the superscript T represents the transpose.

[0080] The entire time axis is evenly divided into several rounds, and the duration of each round is determined according to the actual physical network configuration. The local speed limit value of the distributed network node is updated once in each round. After several rounds of updates, the deviation between the local speed limit value of each distributed network node and the target local speed limit value is less than the set threshold, which means that the algorithm converges to the target local speed limit value, and the entire update process is a Markov chain.

[0081] Specifically, the update of the local speed limit value of the distributed network node performs the following steps in each round:

[0082] (1) Check whether the input rate fluctuation of the distributed network node k exceeds a pre-set threshold τ, that is, check whether |x node -x k |>τ holds, where x node represents the original input rate of the distributed network node k before the fluctuation, and x k represents the current input rate of the distributed network node k;

[0083] If it does not hold, it means that the input rate fluctuation can be ignored;

[0084] If it holds, then set the input rate x next of the distributed network node k in the next round to x k , and use the Metropolis-Hastings algorithm to update the parameters of the k-th column of the transition matrix P to obtain the updated transition matrix P next in the next round. The input of the Metropolis-Hastings algorithm is x next and the input rate x j of the previously cached neighbor distributed network node j.

[0085] (2) Initialize the local speed limit value a next of the distributed network node k in the next round to P kk a k .

[0086] (3) The distributed network node k announces its a k and x next to all neighbor nodes.

[0087] (4) Start receiving the a j and x j announced by the neighbor nodes, where a j must have a value, and x j may be a null value. For each message received from a neighbor node, execute the following steps:

[0088] Add the value of a next to P jk a j , that is, a next +=P jk a j ;

[0089] If x j is not null, then use the Metropolis-Hastings algorithm to update the values of P jk and P kk to P next , and the input of the Metropolis-Hastings algorithm is x next and the received xj ;

[0090] If x j is empty, then use the Metropolis-Hastings algorithm to update the values of P jk and P kk to P next . The input of the Metropolis-Hastings algorithm is x next and the cached x j ;

[0091] At the end of the round, update a k to a next , and update P to P next ;

[0092] Denormalize a next and apply it to the speed limiter.

[0093] In fact, the distributed network node k only holds the matrix elements of the k-th column of P. The above algorithm only needs to operate on this part of the matrix elements, and the other matrices are not used and do not need to be saved.

[0094] S3. When the input rate of the distributed network node changes, recalculate the transition matrix and update the local speed limit value of each distributed network node in real time.

[0095] In the embodiment, if the input rate of the distributed network node changes in the previous round, then when updating in the current round, recalculate the matrix elements in the transition matrix P according to the new input rate that has changed, and update the local speed limit value of the distributed network node according to the matrix elements in the recalculated transition matrix P after the next round.

[0096] As Figure 2 shown, it shows the time flow of the interaction between two neighbor nodes. Node 2 detected a fluctuation in the input rate in the previous round and needs to update the transition matrix in the current round and synchronize the information to neighbor node 1 at the same time. For node 1 and node 2, the communication between nodes is completed by remote procedure call (RPC). RPC is divided into Request and Response. In order for a node to obtain the information of a neighbor node, it first sends a Request and then receives a Response. Figure 2The black dashed line in [Figure 0] connects a pair of Request and Response. The left blue background area represents the previous round, the middle yellow background area represents the current round, and the right blue background area represents the next round. In the previous round, Node 2 updated the input rate, which does not affect the previous round but will affect the current round. Therefore, in the previous round, Node 2 still updated the speed limit value according to the formula in Step S2.3 as normal. However, in the current round, when Node 2 receives an RPC request from Node 1, it will send out the new input rate of Node 2. When Node 2 receives an RPC response from Node 1, it will recalculate the matrix elements in the transition matrix P according to the Metropolis-Hastings algorithm. P ·2 The · in [Figure 2] represents all the elements in this column. For multiple nodes, Node 2 may receive RPC responses sent by multiple neighbor nodes, and will recalculate the matrix elements in the transition matrix P according to the Metropolis-Hastings algorithm. Figure 2 The green dashed line in [Figure 4] represents the update of the matrix elements in the transition matrix P, but this matrix element will not be enabled in the current round. The original matrix elements in the transition matrix P are still used to update the speed limit value. It is not until the start of the next round that the matrix elements in the transition matrix P calculated in the current round are used for calculation.

[0097] As Figure 3 shown in [Figure 9], it shows an example of an adjustment process. The normalized local speed limit value converges to a certain convergence target, and during the adjustment process, the total global speed limit remains unchanged. 10 distributed network nodes are connected in a ring, and each node corresponds to 2 neighbor nodes. The normalized speed limit value of each node is adjusted in each round through the above-mentioned distributed traffic speed limit calculation method. Each color represents the proportion of the normalized speed limit value of the node in each round.

[0098] As Figure 4 shown in [Figure 14], it is a schematic diagram of the CDF evaluation result of the algorithm convergence time. The abscissa is the round number, and the ordinate is the CDF value (Cumulative Distribution Function, cumulative distribution function curve). The results of 10,000 random experiments are shown. In a 100G network, the duration of each round is set to 10ms. 10ms is sufficient for the calculation overhead and communication overhead, and even provides almost 10 times the redundant space. In this case, in 95% of the cases, the preset speed limit accuracy can still be achieved within 1s. The time to reach the speed limit accuracy is called Mixing Time, and it is considered that the algorithm converges after reaching Mixing Time.

[0099] In summary, the non - central - node - based distributed traffic rate - limiting calculation method provided by the embodiments of the present invention realizes protocol - independent, accurate, and fast - converging distributed rate - limiting calculation, thereby better optimizing resource utilization, improving the overall service quality, and ensuring the user experience.

[0100] Based on the same inventive concept, as Figure 5 shown, the embodiments of the present invention also provide a non - central - node - based distributed traffic rate - limiting calculation device 500, including: a rate - limiting problem definition module 510, a node rate - limiting calculation module 520, and a node rate - limiting update module 530.

[0101] The rate - limiting problem definition module 510 is used to define a distributed traffic rate - limiting problem based on the local rate - limiting values, input rates, and output rates of each distributed network node, as well as the global rate - limiting values and the total output rate of all distributed network nodes, and define the target local rate - limiting value based on the fairness assumption.

[0102] The node rate - limiting calculation module 520 is used to calculate the transition matrix of node traffic allocation through the Metropolis - Hastings algorithm under the target local rate - limiting value defined by the fairness assumption based on the distributed traffic rate - limiting problem, and update the local rate - limiting value of the current distributed network node according to the transition matrix and the local rate - limiting values of neighbor nodes.

[0103] The node rate - limiting update module 530 is used to recalculate the transition matrix and update the local rate - limiting value of each distributed network node in real - time when the input rate of the distributed network node changes.

[0104] Based on the same inventive concept, the embodiments of the present invention also provide an electronic device, including a memory and one or more processors. The memory is used to store a computer program, and the processor is used to implement the above - mentioned non - central - node - based distributed traffic rate - limiting calculation method when executing the computer program.

[0105] Based on the same inventive concept, the embodiments of the present invention also provide a computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a computer, the above - mentioned non - central - node - based distributed traffic rate - limiting calculation method is implemented.

[0106] Based on the same inventive concept, the embodiments of the present invention also provide a computer product, which includes a computer program. When the computer program is executed by a processor, the above - mentioned non - central - node - based distributed traffic rate - limiting calculation method is implemented.

[0107] It should be noted that the non - central - node - type distributed traffic rate - limiting calculation device, electronic device, computer - readable storage medium, and computer product provided in the above embodiments all belong to the same inventive concept as the non - central - node - type distributed traffic rate - limiting calculation method. For the specific implementation process, please refer to the embodiments of the non - central - node - type distributed traffic rate - limiting calculation method, which will not be elaborated here.

[0108] The above - described specific implementation manners have elaborated in detail the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, supplements, equivalent replacements, etc. made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-central node distributed flow rate limit calculation method, characterized in that: The following steps are involved: The distributed traffic speed limit problem is defined based on the local speed limit value, input rate and output rate of each distributed network node, and the global speed limit value and the sum of output rates of all distributed network nodes, and the target local speed limit value is defined based on the fairness assumption; Based on the distributed traffic speed limit problem, the distributed Metropolis-Hastings algorithm is used to calculate the transfer matrix of node traffic distribution under the target local speed limit value defined by the fairness assumption. The local speed limit value of the current distributed network node is updated according to the transfer matrix and the local speed limit value of the neighboring node. When the input rate of a distributed network node changes, the transfer matrix is ​​recalculated and the local rate limit value of each distributed network node is updated in real time.

2. The non-central node distributed flow rate limit calculation method according to claim 1 is characterized in that: The definition of the distributed traffic rate limit problem is: For n distributed network nodes, let a i 、x i ,y i The local speed limit value, input rate and output rate of the i-th distributed network node are respectively, and the global speed limit value of all distributed network nodes is a=a1+a2+…+a i +…+a n , the sum of the output rates of all distributed network nodes y = y1 + y2 + ... + y i +…+y n , then the distributed traffic speed limit target is expressed as: The first line indicates that a distributed speed limiter will be independently implemented on each distributed network node, and the second line indicates that the global speed limit value will not exceed the preset speed limit value in any case.

3. The non-central node distributed flow rate limit calculation method according to claim 1 or 2, characterized in that: The limit distribution of the Markov chain is used as the convergence target of the normalized local speed limit value, and the limit distribution is determined by the pre-set fairness assumption. The fairness assumption is expressed as: By setting optional weights and exponents to meet the needs of different systems, it is also called the proportional fairness assumption, where π i represents the convergence target of the normalized local speed limit value of the i-th distributed network node, that is, the target local speed limit value, which is also the i-th component of the Markov chain limit distribution, θ i represents the weight hyperparameter corresponding to the i-th distributed network node, x i represents the input rate of the i-th distributed network node, and the superscript α is a hyperparameter representing the power.

4. The non-central node distributed flow rate limit calculation method according to claim 3 is characterized in that: The transfer matrix of node traffic distribution is calculated by using a distributed Metropolis-Hastings algorithm under the target local speed limit value defined by the fairness assumption, including: Get the input rate x of the current distributed network node k k And the input rate x of traversing all neighbor nodes j of k j As input, for node j,k∈i, the neighbor nodes represent the nodes directly connected to k in the network logical topology; Given the fairness assumption, since the denominator in the fairness assumption is the same for all nodes, the ratio of the limiting distribution is calculated based on the ratio of the input rates And initialize the transfer matrix Calculate the matrix element P corresponding to the jth row and kth column in the transfer matrix P jk : in, Represents the initialization transfer matrix The matrix element corresponding to the j-th row and k-th column in is Represents the initialization transfer matrix The matrix element corresponding to the k-th row and j-th column in ; Calculate the matrix element P corresponding to the kth row and kth column in the transfer matrix P kk : The += symbol indicates that each time a new neighbor element is traversed, the calculation result on the right is added to the original value on the left. For a distributed network node k, it only holds all the matrix elements in the kth column of the transfer matrix P.

5. The non-central node distributed flow rate limit calculation method according to claim 4 is characterized in that: The updating of the local speed limit value of the current distributed network node according to the transfer matrix and the local speed limit value of the neighboring node includes: For the current distributed network node k, its local speed limit value a k The update method is: Among them, a j represents the local speed limit value of neighbor node j, (j,k)∈E represents the set of edges E of the network logical topology that connect the current distributed network node k and its neighbor node j, P kj represents the matrix element corresponding to the kth row and jth column in the transfer matrix P, and the superscript T represents the transpose; The entire timeline is evenly divided into several rounds. The duration of each round is determined according to the actual physical network configuration. The local speed limit value of the distributed network node is updated once in each round. After several rounds of updates, the deviation between the local speed limit value of each distributed network node and the target local speed limit value is less than the set threshold, which means that the algorithm converges to the target local speed limit value. The entire update process is a Markov chain.

6. The non-central node distributed flow rate limit calculation method according to claim 5 is characterized in that: When the input rate of the distributed network node changes, the transfer matrix is ​​recalculated and the local rate limit value of each distributed network node is updated in real time, including: If the input rate of the distributed network node changes in the previous round, the matrix elements in the transfer matrix P will be recalculated according to the changed new input rate when the current round is updated, and the local speed limit value of the distributed network node will be updated according to the recalculated matrix elements in the transfer matrix P after the next round.

7. A non-central node type distributed flow rate limit calculation device, implemented by using the non-central node type distributed flow rate limit calculation method according to any one of claims 1 to 6, characterized in that: include: Speed ​​limit problem definition module, node speed limit calculation module and node speed limit update module; The speed limit problem definition module is used to define the distributed traffic speed limit problem based on the local speed limit value, input rate and output rate of each distributed network node, and the global speed limit value and the sum of the output rates of all distributed network nodes, and define the target local speed limit value based on the fairness assumption; The node speed limit calculation module is used to calculate the transfer matrix of node traffic distribution by the Metropolis-Hastings algorithm based on the distributed traffic speed limit problem under the target local speed limit value defined by the fairness assumption, and update the local speed limit value of the current distributed network node according to the transfer matrix and the local speed limit value of the neighboring node; The node speed limit updating module is used to recalculate the transfer matrix and update the local speed limit value of each distributed network node in real time when the input rate of the distributed network node changes.

8. An electronic device comprising a memory and one or more processors, wherein the memory is used to store a computer program, characterized in that: The processor is used to implement the non-central node distributed traffic speed limit calculation method described in any one of claims 1-6 when executing the computer program.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a computer, the non-central node distributed traffic speed limit calculation method described in any one of claims 1 to 6 is implemented.

10. A computer product comprising a computer program, characterized in that When the computer program is executed by a processor, the non-central node distributed traffic speed limit calculation method described in any one of claims 1-6 is implemented.

Images