Traffic scheduling method and system based on power white-box network

Abstract

The invention discloses a traffic scheduling method and system based on a power white-box network, and the method comprises the steps: determining a current traffic threshold value through a white-box network controller, and transmitting the current traffic threshold value to an edge switch; the method comprises the following steps: acquiring to-be-scheduled traffic data of end-side equipment through an edge switch, determining a traffic type of the to-be-scheduled traffic data according to a current traffic threshold, and forwarding the to-be-scheduled traffic data to an inner-layer switch based on the traffic type of the to-be-scheduled traffic data; and forwarding the traffic data to be scheduled to the data center service cluster through the inner-layer switch. According to the technical scheme, the time delay problem caused by blocking of the short-flow data packets and the disorder problem of the long-flow data packets are avoided, the utilization rate and the processing efficiency of a switch forwarding network are improved, and better use experience is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of data transmission, and in particular to a flow scheduling method and system based on a power white box network. Background Art

[0002] There are many types of services running in the power system, but in general, the business network traffic in the power system can be divided into short flows that are sensitive to delay and long flows that do not require high delay. When long and short heterogeneous flows are mixed, a large number of long flow packets are scattered on all paths, which will block short flow packets; and the long flow packets are affected by the sudden short flow and will encounter packet disorder problems. At the same time, the long flow packets will consume a lot of switch cache space to obtain high throughput, which will cause the short flow packets to experience head-of-line blocking and cause serious queuing delays. Large queuing delays dramatically increase the completion time of short flows, causing these flows to miss the deadline defined by the Service Level Agreement (SLA).

[0003] At present, the industry has alleviated the problem of short flow congestion delay by reducing the switch cache occupancy rate. However, although reducing the switch cache occupancy rate improves the performance of short flows, it cannot guarantee that the link bandwidth is fully utilized, resulting in a loss of long flow throughput. Therefore, there is an urgent need for a traffic scheduling method based on a power white box network that is compatible with both long and short flows and has high efficiency and effect. Summary of the invention

[0004] The present invention provides a traffic scheduling method and system based on an electric power white box network, which avoids the delay problem caused by the congestion of short flow data packets and the disorder problem of long flow data packets, improves the utilization rate and processing efficiency of the switch forwarding network, and provides a better user experience.

[0005] In a first aspect, an embodiment of the present disclosure provides a traffic scheduling method based on a power white box network, which is applied to a traffic scheduling system based on a power white box network, wherein the system includes a white box network controller, at least one edge switch and at least one inner switch, wherein the white box network controller is connected to the edge switch and the inner switch respectively, wherein the edge switch is connected to the inner switch, wherein the system is connected to an end-side device through the edge switch, and is connected to a data center service cluster through the inner switch, wherein the method includes:

[0006] Determine a current traffic threshold through the white box network controller, and send the current traffic threshold to the edge switch;

[0007] Acquire the traffic data to be scheduled of the end-side device through the edge switch, determine the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forward the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled;

[0008] The traffic data to be scheduled is forwarded to the data center service cluster through the inner switch.

[0009] In a second aspect, an embodiment of the present disclosure provides a traffic scheduling system based on a power white box network, comprising: a white box network controller, at least one edge switch and at least one inner switch; the white box network controller is connected to the edge switch and the inner switch respectively, and the edge switch is connected to the inner switch; the system is connected to the end-side device through the edge switch, and is connected to the data center service cluster through the inner switch;

[0010] The white box network controller is used to determine a current traffic threshold and send the current traffic threshold to the edge switch;

[0011] The edge switch is used to obtain the traffic data to be scheduled of the end-side device, determine the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forward the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled;

[0012] The inner switch is used to forward the traffic data to be scheduled to the data center service cluster.

[0013] A traffic scheduling method and system based on a power white box network according to an embodiment of the present invention determines the current traffic threshold through a white box network controller, and sends the current traffic threshold to an edge switch; obtains the traffic data to be scheduled of the end-side device through the edge switch, and determines the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forwards the traffic data to be scheduled to an inner switch based on the traffic type of the traffic data to be scheduled; and forwards the traffic data to be scheduled to a data center service cluster through the inner switch. The above technical scheme realizes the classified collaborative scheduling of long and short flow traffic data packets, avoids the delay problem caused by the congestion of short flow data packets and the disorder problem of long flow data packets, improves the utilization rate and processing efficiency of the switch forwarding network, and provides a better user experience.

[0014] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 This is a flow chart of a traffic scheduling method based on a power white box network provided in Embodiment 1 of the present invention;

[0017] Figure 2 This is an example diagram of a traffic scheduling architecture provided by Embodiment 1 of the present invention;

[0018] Figure 3 This is a flow chart of a traffic scheduling method based on a power white box network provided in Embodiment 2 of the present invention;

[0019] Figure 4 It is a structural diagram of a flow scheduling system based on a power white box network provided in Embodiment 3 of the present invention;

[0020] Figure 5 It is a schematic diagram of the interaction structure between an edge switch and a white box network controller in a traffic scheduling system based on a power white box network provided by Embodiment 3 of the present invention;

[0021] Figure 6 It is a structural diagram of an inner switch in a traffic scheduling system based on a power white box network provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0023] It should be noted that the terms "first", "second" and "target" in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] With the construction of new power systems, the number of power Internet of Things information and communication access nodes has doubled, and the trend of cloud-based services has deepened. Some sensitive services have higher requirements for the link quality and service opening time limit of the cloud service bearer network. A large number of emerging computing tasks need to be delivered to the cloud for processing, which undoubtedly puts a severe test on network transmission capacity and cloud computing capacity. In recent years, artificial intelligence technology, especially deep learning, has developed rapidly, triggering a series of business and technology changes including the power industry. Supporting these emerging businesses also requires massive data, which means that while delivering computing tasks to the cloud for processing, data transmission also requires a lot of network resources.

[0025] There are many types of services running in the power system, but in general, traffic can be divided into two categories: long traffic and short traffic. Different types of traffic have different requirements for network performance. Long traffic is often generated by applications such as downloading data and disk backup. Such applications focus on network throughput, while network indicators such as latency and jitter have little impact on their application performance. Short traffic is often generated by real-time services such as substation real-time monitoring, distribution network differential protection, 10kV distributed energy storage regulation, and charging pile power balance. This traffic has a strong demand for immediacy, is more sensitive to network latency, and does not pay attention to indicators such as throughput. Since different types of network traffic have different requirements for network performance, in order to better utilize network resources, adopting different allocation and forwarding strategies for different network traffic has become an important way to achieve better load effects.

[0026] The above power business traffic is divided into long flows (delay-sensitive tasks) with throughput as the target and short flows (real-time processing tasks) with latency requirements as the main requirement according to latency requirements, data size, and business characteristics, as shown in the following table.

[0027] Classification table of long and short flow of power business traffic

[0028]

[0029] In order to cope with the exponential growth of data and high concurrent access, intermediate transmission networks and large data centers need to deploy load balancing modules to handle huge workloads and improve network link utilization. The primary goal of load balancing is to evenly distribute traffic to parallel paths to avoid high-load traffic causing network congestion while ensuring network utilization. To achieve this goal, the main issues that need to be considered are as follows:

[0030] (1) Network traffic can be divided into delay-sensitive short flows and delay-insensitive long flows. When heterogeneous flows are mixed, a large number of long flow packets are scattered on all paths, blocking short flow packets. Long flow packets are affected by bursty short flows and will encounter packet disorder problems.

[0031] (2) At the same time, long-flow packets consume a large amount of switch cache space to obtain high throughput, which will cause short-flow packets to experience head-of-line blocking and cause serious queuing delays. Large queuing delays dramatically increase the completion time of short flows, causing these flows to miss the deadline defined by the service level agreement (SLA). However, although reducing the switch cache occupancy rate improves the performance of short flows, it cannot ensure that the link bandwidth is fully utilized, resulting in a loss of throughput for long flows.

[0032] To solve the above problems, the present invention provides a method and system for coordinated scheduling of long and short-term easy-to-buy traffic based on an electric power white box network, wherein the traffic scheduling system based on the electric power white box network is constructed by cloud-edge-end system computing.

[0033] Cloud-edge-end collaborative computing is a computing model based on cloud computing and edge computing, built on a multi-level cloud-edge-end infrastructure. On the one hand, through unified management and control of the cloud-edge-end, the original cloud service can be sunk to the end and edge of the cloud-edge-end fusion system, forming a service architecture of cloud-edge-end collaboration, thereby supporting the local response of services; on the other hand, it can be done by sinking data preprocessing, caching, network forwarding and other tasks to the edge and end processing to shorten the response delay, reduce the pressure on the cloud, and reduce the transmission load. Expand some of the cloud's storage, computing, network, security and other capabilities to the edge infrastructure, and truly realize "ubiquitous computing" through multi-level collaboration, global scheduling and unified management and control of the cloud-edge-end.

[0034] Embodiment 1

[0035] Figure 1 It is a flow chart of a traffic scheduling method based on a power white box network provided in Example 1 of the present invention. This embodiment is applicable to the situation of classified and coordinated scheduling of long and short heterogeneous traffic data. The method can be executed by a traffic scheduling system based on a power white box network.

[0036] Based on the cloud-edge-end architecture, the traffic scheduling system based on the power white box network is connected to the end-side device and the data center service cluster respectively. Specifically, the traffic scheduling system based on the power white box network includes a white box network controller, at least one edge switch and at least one inner layer switch. The white box network controller is connected to the edge switch and the inner layer switch respectively, and the edge switch is connected to the inner layer switch. The system is connected to the end-side device through the edge switch and is connected to the data center service cluster through the inner layer switch.

[0037] like Figure 1 As shown, the method includes:

[0038] S101. Determine a current traffic threshold through a white box network controller, and send the current traffic threshold to an edge switch.

[0039] In this embodiment, an intelligent agent is provided in the white box network controller, and the intelligent agent is used to realize the real-time calculation of the current flow threshold. The current flow threshold can be understood as a threshold calculated in real time for judging the flow type. The edge switch can be understood as a white box switch provided at the edge of the switching network and connected to the end-side device, and the edge switch is connected to the end-side device and the inner layer switch respectively.

[0040] Specifically, the white box network controller obtains the load status information of the switching network composed of each edge switch and each inner layer switch in real time, calculates the dynamic current flow threshold in real time based on the status information, and sends the flow threshold to each associated edge switch.

[0041] S102: Obtain the traffic data to be scheduled of the end-side device through the edge switch, determine the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forward the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled.

[0042] In this embodiment, the end-side equipment includes substations, distribution rooms, charging piles and other equipment, and the data information generated by them is the traffic data to be scheduled in the embodiment of the invention. The traffic type can be understood as the length classification result of the traffic data to be scheduled, including long flows and short flows. The inner switch can be understood as a white box switch that is set in the inner layer of the switching network and is only used to perform the function of transferring the traffic data to be scheduled. It can be understood that some inner switches are also set at the edge of the switching network and connected to the data center management cluster, which is used to directly forward the traffic data to be scheduled to the data center management cluster.

[0043] Specifically, the edge switch obtains the forwarding strategy for different traffic types issued by the controller, obtains the current traffic threshold issued by the controller in real time, obtains the traffic data to be scheduled transmitted by the end-side device, and determines whether the traffic data to be scheduled is a long flow type or a short flow type based on the current traffic threshold. If it is a long flow type, the forwarding of the long flow type traffic data to be scheduled is executed based on the forwarding strategy corresponding to the long flow type. If it is a short flow type, the forwarding of the short flow type traffic data to be scheduled is executed based on the forwarding strategy corresponding to the short flow type, and the traffic data to be scheduled is forwarded to the inner layer switch connected to the edge switch, so that the inner layer switch continues to execute the forwarding operation of the traffic data to be scheduled.

[0044] S103. Forward the traffic data to be scheduled to the data center service cluster through the inner switch.

[0045] In this embodiment, the edge switch has processing and forwarding functions, can determine the traffic type of the traffic data to be scheduled and perform forwarding, and the inner switch only has forwarding function, which can realize the forwarding of the traffic to be scheduled.

[0046] Specifically, the inner switch obtains the forwarding strategy issued by the white box network controller, obtains the traffic data to be scheduled forwarded by the edge switch or the inner switch of the previous level, executes the corresponding forwarding strategy for the traffic data to be scheduled based on its traffic type, and forwards the traffic data to be scheduled to the inner switch of the next level or the data center service cluster.

[0047] For example, Figure 2 is an example diagram of a traffic scheduling architecture provided by the first embodiment of the present invention, such as Figure 2As shown, the traffic scheduling architecture adopts a cloud-edge-end architecture, combined with a traffic scheduling system. The traffic scheduling architecture includes end-side devices, edge clusters, a traffic scheduling system based on a power white box network, and a cloud data center. The end-side device sends the traffic data to be scheduled to the edge cluster. The edge cluster obtains the traffic data to be scheduled through an edge wireless access point (Wireless Access Point), and forwards the traffic data to be scheduled to the edge node, so that the traffic data to be scheduled of each end-side device is forwarded to the traffic scheduling system through the edge node. The traffic scheduling system obtains the traffic data to be scheduled of the end-side device forwarded by the edge node through the edge switch. The white box network controller obtains the load status information uploaded by the edge switch in the switching network, and calculates the current traffic threshold in real time based on the load status information, and sends the current traffic threshold to each edge switch in the switching network. The edge switch determines the traffic type of the traffic data to be scheduled based on the current traffic threshold obtained, and executes the corresponding forwarding strategy based on the traffic type, and forwards the traffic data to be scheduled to the inner switch. The inner switch also executes the corresponding forwarding strategy based on the traffic type of the traffic data to be scheduled, and forwards the traffic data to be scheduled to the inner switch of the next layer or directly to the data center service cluster in the cloud service center. The cloud data center obtains the traffic data to be scheduled through the edge wireless access point, and forwards the traffic data to be scheduled to the data center manager. The data center manager then forwards the traffic data to be scheduled to the data center service cluster, that is, the server cluster of each data center.

[0048] A traffic scheduling method based on an electric power white box network provided in an embodiment of the present invention includes determining a current traffic threshold through a white box network controller, and sending the current traffic threshold to an edge switch; obtaining traffic data to be scheduled of an end-side device through the edge switch, and determining a traffic type of the traffic data to be scheduled according to the current traffic threshold, and forwarding the traffic data to be scheduled to an inner layer switch based on the traffic type of the traffic data to be scheduled; and forwarding the traffic data to be scheduled to a data center service cluster through the inner layer switch. The above technical solution adopts the traffic scheduling architecture to solve the problem that when the edge side transmits a large amount of data to the cloud side through the intermediate switching network, the long and short flow transmissions affect each other due to different targets (that is, the long flow occupies the link for a long time due to the large amount of data, resulting in short flow congestion; or the short flow causes long flow starvation due to its high priority). A cloud-edge power business traffic scheduling architecture based on the division of long and short flows is designed, and long flows and short flows are isolated and transmitted on different links using different strategies. Long and short flows are identified on the edge white box switch to facilitate forwarding to different links according to the traffic type on the subsequent white box switch, realizing the classified and coordinated scheduling of long and short flow traffic data packets, avoiding the delay problem caused by the congestion of short flow data packets and the disorder problem of long flow data packets, improving the utilization and processing efficiency of the switch forwarding network, and providing a better user experience.

[0049] Optionally, as a first optional embodiment of this embodiment, the method further includes:

[0050] Through the white box network controller, the forwarding strategy corresponding to the traffic type is determined and sent to the edge switch and inner switch.

[0051] In this embodiment, the micro-burst growth of short flows in the cloud-edge collaborative network is an important cause of data packet loss. The micro-burst growth of short flows ends within a few us, while the global network information control cycle consumes a relatively long time. Therefore, short flows are suitable for making routing decisions based on local information. In addition, since the data volume of short flows is small and they are sensitive to latency indicators, they are mostly generated by emergency tasks. When scheduling short flows, a priority scheduling scheme is adopted to determine the forwarding order of short flows based on the processing priority of each flow. Long flows are sensitive to throughput indicators. In order to achieve a larger throughput for long flows and avoid the disorder of data packets in the traffic, a centralized queuing scheduling method based on global information is adopted for long flows, and priority is given to scheduling larger data flows to links with more remaining link capacity.

[0052] Specifically, the white box network controller determines the priority scheduling strategy corresponding to the short flow type and the centralized queuing scheduling strategy corresponding to the long flow; the priority scheduling strategy determines the forwarding order of short flows based on the processing priority of the traffic, with high priority being forwarded first and low priority being forwarded later; the centralized queuing scheduling strategy prioritizes scheduling larger data flows to links with more remaining link capacity to achieve better load balancing. The determined priority scheduling strategy for short flows and the centralized queuing scheduling strategy for long flows are sent to each edge switch and inner layer switch in the switching network, so that each white box switch can forward the corresponding type of traffic data to be scheduled based on the corresponding forwarding strategy, achieving efficient and high-utilization traffic collaborative scheduling.

[0053] Embodiment 2

[0054] Figure 3 It is a flow chart of a traffic scheduling method based on a power white box network provided in the second embodiment of the present invention. This embodiment is a further optimization of any of the above embodiments and can be applied to the situation of classified and coordinated scheduling of long and short heterogeneous traffic data. The method can be executed by a traffic scheduling system based on a power white box network.

[0055] like Figure 3 As shown, the method includes:

[0056] S201. Obtain current network status data of an edge switch through a white box network controller.

[0057] In this embodiment, the current network status data can be understood as data representing the network load status, including throughput data of the edge switch and long and short flow identification threshold data.

[0058] Specifically, the throughput data of each edge switch at the same time and the long and short flow identification threshold data of each edge switch are obtained. The long and short flow identification threshold data can be understood as the current flow threshold for identifying the flow type sent by the white box network controller to the edge switch for the most recent time, that is, the current flow threshold at the most recent historical moment.

[0059] S202: Update network parameters according to current network status data, and determine current traffic thresholds using the updated network parameters.

[0060] In this embodiment, the network parameter can be understood as an influencing factor that affects the generation of the current traffic threshold, and the network parameter is updated in real time based on the network load status of the switching network.

[0061] Specifically, since the threshold change of long and short flows is related to the load input of the entire network, the traffic in the network is injected into the network through the edge layer switch. Therefore, the throughput of the edge switch can reflect the load intensity of the entire network. Thus, the state space of the intelligent agent is determined The state space is represented as: Among them, S E(j) represents the throughput of the jth edge switch.

[0062] Since the main goal of the embodiment of the present invention is to achieve accurate identification of long and short flows under any load at any time, the action space here is set as the threshold for distinguishing long and short flows in the next time period. It can be expressed as: Among them, TH E(j) It represents the long and short flow identification threshold of the j-th edge switch.

[0063] The decision results of the Deep Deterministic Policy Gradient (DDPG) algorithm affect the overall network transmission quality. Long flows have large amounts of data and occupy links for a long time, so we hope to maximize the throughput of long flows. Short flows need to be transmitted quickly, and latency is their key performance indicator. Therefore, it is necessary to penalize the latency of short flows to ensure that short flows are not excessively blocked by long flows. Therefore, the mathematical expression of the reward function used to achieve dynamic adjustment of network parameters is: Where x k represents the throughput of the kth long flow, z mrepresents the delay of the mth short flow. α and β are the weight factors of the two sides, which need to be adjusted according to the actual scenario to balance the requirements of long flow throughput and short flow delay.

[0064] The embodiment of the present invention combines the deterministic policy gradient of the DDPG algorithm with a structure similar to the Deep Q-Network (DQN). The actor-critic framework is used to implement model parameter updates. The model is divided into a main network and a target grid. The main network has a parameter ω μ The actor network with parameters is The target network has parameters ω μ′ The actor network with parameters is The critic network of the proposed algorithm is introduced. Experience replay technology is added to the model to reduce the correlation of data. The model uses policy gradient to update the main network and updates the target through soft update, providing an adaptive threshold calculation algorithm based on DDPG.

[0065] The input of the algorithm is the hyperparameters in the neural network, the training batch I and the learning rate σ, and the input is the long and short flow identification threshold for the next time period, that is, the current flow threshold. The state space, action space and reward function are determined through the current network state data, and the network parameters are updated based on the state space, action space and reward function. Then, the input data is calculated based on the updated network parameters to obtain the new current flow threshold.

[0066] Specifically, for the algorithm (model) structure, first initialize the parameters: set the dimensions of the agent state space and action space; initialize the actor network parameters and the critic network parameters in the main network; initialize the actor network parameters and the critic network parameters in the target network; initialize the experience replay buffer and set the training batch I; set the learning rate σ of the target network; then initialize the network state and reset the network collection data for the nth training; set the loop structure; then collect the current network state data of each edge switch in the switching network, determine the agent state space, action space and reward function based on the current network state data, and based on the current state space S t Select an action a from the action space t , deploy the action to the edge switch and obtain the current reward r based on the reward function t , get the state S from the neural network t+1 , the data unit g t ={s t ,a t ,r t ,s t+1}Put it into the experience buffer and get I g from the experience buffer t Then calculate the target value y for each data unit i =r i+Q ′ (s i+1 ,μ ′ (s i+1 ;ω μ′ );ω Q′ ). The target value of the data unit is the evaluation of the current action a t The result, μ ′ (s i+1 ;ω μ′ ) means that in a given state s i+1 Next, according to the parameters ω of the target actor network μ′ The action selected. ′ () indicates that in a given state s i+1 and action μ ′ (s i+1 ;ω μ′ ), according to the parameter ω of the target critic network Q′ The expected value of the predicted future cumulative reward. Update the main network parameters through policy gradient: Update the main network critic network parameters through goal-oriented or policy gradient Use soft update method to update network target parameter ω Q′ ←σω Q +(1-σ)ω Q′ ;ω μ′ ←σω u +(1-σ)ω u′ Based on the updated network parameters, the input data is calculated to obtain the current flow threshold that matches the current network load status.

[0067] S203: Send the current traffic threshold to each edge switch.

[0068] Specifically, after the white box network controller calculates the current traffic threshold corresponding to the network load state at the current moment, it sends the current traffic threshold to each edge switch, so that the edge switch can distinguish the traffic type based on the current traffic threshold.

[0069] S204: Obtain the traffic data to be scheduled of the end-side device through the edge switch.

[0070] In this embodiment, the edge switch is connected to the end-side device through the edge cluster, and obtains the traffic data to be scheduled transmitted by the end-side device through the edge cluster.

[0071] S205: Determine a link bandwidth ratio according to the data stream bytes of the traffic data to be scheduled, and compare the link bandwidth ratio with a current traffic threshold.

[0072] In this embodiment, the link bandwidth ratio can be understood as the ratio of data stream bytes to the current maximum link bandwidth.

[0073] Specifically, determine the total data flow bytes byte1 of the traffic data to be scheduled received at time t1, the total data flow bytes byte2 of the traffic data to be scheduled received at time t2, and the current maximum link bandwidth BW max , based on the total data flow bytes byte1, total data flow bytes byte2 and the current link maximum bandwidth BW max ,pass Determine the link bandwidth ratio flow. Compare the link bandwidth ratio flow with the current traffic threshold to determine the relationship between the two.

[0074] S206: If the link bandwidth ratio is greater than the current traffic threshold, determine that the traffic type of the traffic data to be scheduled is a long flow.

[0075] S207: If the link bandwidth ratio is less than the current traffic threshold, determine that the traffic type of the traffic data to be scheduled is a short flow.

[0076] In this embodiment, the identification of traffic types is different from the traditional identification method. The traditional solution directly uses the data flow accumulation threshold as the only way to distinguish long and short flows. The threshold is often set by researchers based on their own experience, and the experience of researchers directly affects the effect of network transmission quality. Secondly, the existing major cases are often only optimized for solving the short flow congestion problem and solving the long flow hash collision problem, which improves the performance of a certain flow while damaging the performance of another flow.

[0077] The dynamically adaptive flow cluster threshold provided in this embodiment can better match the characteristics of long and short flows compared to the static flow threshold, thereby achieving accurate distinction of flow types.

[0078] S208: Add a label corresponding to the traffic type to the traffic data to be scheduled.

[0079] In this embodiment, since all traffic passes through the edge layer switch, the current traffic threshold calculated by the intelligent agent is deployed to the edge switch. The edge switch distinguishes the passing traffic. If the sent data flow is identified as a long flow, a long flow label is added to the data packet (traffic data to be scheduled). The same applies to short flows, so that the inner layer switch that passes through later can directly determine the traffic type of the traffic data to be scheduled by identifying the label, and then determine how to route the traffic data to be scheduled.

[0080] S209: If the traffic type of the traffic data to be scheduled is a long flow, determine a response probability that the edge switch forwards the traffic data to be scheduled to each associated inner layer switch.

[0081] In this embodiment, the response probability can be understood as the probability that the edge switch forwards the traffic data to be scheduled to the inner switch and the inner switch responds.

[0082] Specifically, the white box switch (edge ​​switch and inner switch) uses the open flow protocol OpenFlow to monitor the link ports of the two connected white box switches and summarize the information obtained. Monitor the link port connected to switch i and switch j, obtain the number of bytes sent and received by the port within the sampling time interval, add the number of bytes received and sent within the time interval, and then divide it by the time difference to obtain the used bandwidth BW of the link between switches i and j. uij , as shown below: Among them, rbyte ij The number of bytes received by the port during the sampling time interval, tbyte ij is the number of bytes sent. The maximum bandwidth minus the used bandwidth is the remaining bandwidth BW of the link between switch i and switch j. ij , as shown below: ij =BW max -BW u ij Thus, the congestion index of each port is obtained. All operations are based on private links established for different flow types. According to the network congestion situation, traffic can be randomly routed to each available port and ij =BW ij / ∑BW assigns a corresponding weight P to each port ij , that is, the probability that switch i will forward the response long flow to switch j.

[0083] S210: Determine the inner switch corresponding to the maximum response probability as the target inner switch, and forward the traffic data to be scheduled to the target inner switch through the long flow forwarding port.

[0084] In this embodiment, the maximum response probability can be understood as the maximum value among multiple response probabilities. The target inner switch can be understood as the inner switch to receive the traffic data to be scheduled forwarded by the edge switch. The long flow forwarding port can be understood as a port in the white box switch dedicated to forwarding the traffic data to be scheduled with the traffic type of long flow.

[0085] Specifically, the same edge switch may be connected to multiple inner switches, so there will be multiple response probabilities corresponding to the edge switch. The maximum value of these response probabilities is determined, which is the maximum response probability. The maximum response probability represents the probability that a certain inner switch responds to the scheduled traffic data forwarded by the edge switch. The inner switch corresponding to the maximum response probability is determined and the inner switch is determined as the target inner switch. The scheduled traffic data is forwarded to the target inner switch through the long flow forwarding port to achieve long flow forwarding based on throughput.

[0086] S211: If the traffic type of the traffic data to be scheduled is a short flow, determine the forwarding link weight of the edge switch to forward the traffic data to be scheduled to each associated inner layer switch.

[0087] In this embodiment, the forwarding link weight may be understood as a weight value representing the forwarding priority of the traffic data to be scheduled.

[0088] Specifically, the forwarding decision of short flows is considered mainly based on link delay and packet loss rate, thereby determining the forwarding link weight that represents the forwarding priority of short flows. total =t2-t1 to get the total transmission delay, where t1 is the time when white-box switch i receives the message of the traffic data to be scheduled, white-box switch i forwards the traffic data packet to be scheduled to switch j, and white-box switch j then sends the data packet to the white-box network controller. t2 is the time when the white-box network controller monitors the message. Next, the white-box network controller sends a request message Request to switch i and switch j respectively, and the sending time is recorded as t qi and t qj After receiving the Request message, switch i and switch j immediately send a reply message Reply to the white box network controller. The time when the controller receives the Reply message from switch i and switch j is t pi and t pj The time difference between a pair of messages is the time it takes for the message to go back and forth on the link from the controller to the switch. Therefore, the link delay from the white box network controller to the switch is half of the time difference. The delay between switch i and switch j is Delay ij It is expressed as: Similar to the link bandwidth monitoring method, the packet loss rate monitoring module calculates the link packet loss rate by actively querying the port statistics on switch i connected to switch j. The white box network controller sends a request message to query the switch to receive the number of packets sent by other switches, which is rbyte. ij , the number of packets lost is rbyte lossij ;The number of packets sent is tbyte ij, the number of lost packets is tbyte lossij The calculation of the link between switch i and switch j is as follows: Considering the delay and packet loss rate as the target short flow forwarding link weight W ij It is expressed as: Where Delay ij and Loss ij are the delay and packet loss rate of the remaining bandwidth on the link between i and node j, respectively, and d and l are the corresponding weights. According to the Quality of Service (QoS) requirements of short-flow services, the weights d and l of link delay and packet loss rate are set with different scoring weight values ​​according to the characteristics of short-flow services, so as to ensure that the optimal path of short-flow services is first allocated to short-flow services with higher priority, thereby improving the real-time processing capability of short-flow services.

[0089] S212: determine the inner switch corresponding to the maximum forwarding link weight as the target inner switch, and forward the traffic data to be scheduled to the target inner switch through the short flow forwarding port.

[0090] In this embodiment, the maximum forwarding link weight can be understood as the maximum value among multiple forwarding link weights. The target inner switch can be understood as the inner switch that is to receive the scheduled traffic data forwarded by the edge switch. The short flow forwarding port can be understood as a port in the white box switch that is dedicated to forwarding the scheduled traffic data of the traffic type of short flow.

[0091] Specifically, the same edge switch may be connected to multiple inner switches, so there will be multiple forwarding link weights corresponding to the edge switch. The maximum value among these forwarding link weights is determined, which is the maximum forwarding link weight. The maximum forwarding link weight represents the priority of a certain inner switch in responding to the scheduled traffic data forwarded by the edge switch. The inner switch corresponding to the maximum forwarding link weight is determined and the inner switch is determined as the target inner switch. The scheduled traffic data is forwarded to the target inner switch through the short-flow forwarding port to achieve short-flow forwarding with latency stability as the main focus.

[0092] S213: Obtain the traffic data to be scheduled through the inner switch, parse the label of the traffic data to be scheduled, and determine the traffic type of the traffic data to be scheduled.

[0093] In this embodiment, the inner switch obtains the traffic data to be scheduled forwarded by the edge switch or the traffic data to be scheduled forwarded by the inner switch of the upper layer, parses the label of the traffic data to be scheduled, and determines whether the traffic type of the traffic data to be scheduled is a long flow or a short flow.

[0094] S214: Call the corresponding forwarding port based on the traffic type, and forward the traffic data to be scheduled to the data center service cluster.

[0095] In this embodiment, if the traffic type is a long flow, the traffic data to be scheduled is calculated based on the processing unit corresponding to the long flow, and the calculated traffic data to be scheduled is forwarded to the inner switch or data center service cluster of the next level through the long flow forwarding port corresponding to the long flow; if the traffic type is a short flow, the traffic data to be scheduled is calculated based on the processing unit corresponding to the short flow, and the calculated traffic data to be scheduled is forwarded to the inner switch or data center service cluster of the next level through the short flow forwarding port corresponding to the short flow.

[0096] A flow scheduling method based on an electric power white box network provided by an embodiment of the present invention comprises obtaining the current network status data of an edge switch through a white box network controller; updating network parameters according to the current network status data, and determining the current flow threshold through the updated network parameters; sending the current flow threshold to each edge switch; obtaining the flow data to be scheduled of the end-side device through the edge switch; determining the link bandwidth ratio according to the data stream bytes of the flow data to be scheduled, and comparing the link bandwidth ratio with the current flow threshold; if the link bandwidth ratio is greater than the current flow threshold, determining that the flow type of the flow data to be scheduled is a long flow; if the link bandwidth ratio is less than the current flow threshold, determining that the flow type of the flow data to be scheduled is a short flow; adding a label corresponding to the flow type to the flow data to be scheduled; if the flow type of the flow data to be scheduled is a long flow , determine the response probability of the edge switch forwarding the traffic data to be scheduled to the associated inner switches; determine the inner switch corresponding to the maximum response probability as the target inner switch, and forward the traffic data to be scheduled to the target inner switch through the long flow forwarding port; if the traffic type of the traffic data to be scheduled is short flow, determine the forwarding link weight of the edge switch forwarding the traffic data to be scheduled to the associated inner switches; determine the inner switch corresponding to the maximum forwarding link weight as the target inner switch, and forward the traffic data to be scheduled to the target inner switch through the short flow forwarding port; obtain the traffic data to be scheduled through the inner switch, parse the label of the traffic data to be scheduled, and determine the traffic type of the traffic data to be scheduled; call the corresponding forwarding port based on the traffic type, and forward the traffic data to be scheduled to the data center service cluster. The above technical solution is to solve the problem that when heterogeneous flows are mixed, long flow data packets and short flow data packets are congested due to vicious competition on the same path. The present invention deploys a long and short flow dynamic classification model based on DDPG on the edge switch. On the subsequent inner switch, various flows are dynamically isolated and transmitted on different paths according to the labels of different flows, while ensuring low latency of short flows and maximizing the throughput of long flows. Through the link load sharing mechanism, the flow is distributed to each path according to certain rules, thereby improving network utilization, avoiding network congestion, and providing a better user experience. In order to solve the problem that long flow data packets consume a large amount of switch cache space to obtain high throughput, so that short flow data packets experience head-of-line blocking and cause serious queuing delay problems, fast pipeline resources are used in the white box switch to process the special computing functions required for long and short flows in parallel and separately. The long and short flow special computing unit and the ALU in the action engine are executed in parallel and do not affect each other, which not only retains the high performance of pipeline stream processing, but also increases the optional function realization of long and short flows, so as to avoid idle white box switch resources and cause traffic congestion.

[0097] Embodiment 3

[0098] Figure 4Schematic diagram of a flow scheduling system based on a power white box network provided by the third embodiment of the present invention. Figure 4 As shown, the system includes: a white box network controller 31, at least one edge switch 32 and at least one inner switch 33; the white box network controller 31 is connected to the edge switch 32 and the inner switch 33 respectively, and the edge switch 31 is connected to the inner switch 32; the system is connected to the end-side device through the edge switch 31, and is connected to the data center service cluster through the inner switch 32;

[0099] The white box network controller 31 is used to determine a current traffic threshold and send the current traffic threshold to the edge switch 32;

[0100] The edge switch 32 is used to obtain the traffic data to be scheduled of the end-side device, determine the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forward the traffic data to be scheduled to the inner switch 33 based on the traffic type of the traffic data to be scheduled;

[0101] The inner switch 33 is used to forward the traffic data to be scheduled to the data center service cluster.

[0102] Various types of network applications run in the cloud-network collaborative network: such as network search, data transmission, streaming media playback, etc. According to the characteristics of the central transmission network traffic, long flows, which account for 20% of the traffic, carry about 80% of the data volume, and 80% of the short flows carry 20% of the data volume. The traffic division granularity does not take into account the difference between traffic flows. A uniform static granularity is used for global network traffic. When the traffic segmentation threshold is large, the data carried by short flows has fewer opportunities for routing selection, and there is a greater probability of head-of-line blocking, which reduces the flow completion time of short flows overall. When the traffic segmentation threshold is small, long flows may experience congestion and packet disorder, reducing long flow throughput and increasing packet delay. Since long flows are more sensitive to throughput, and short flows are more sensitive to flow completion time, setting specific decision granularity for the attributes of different flows will help improve the service quality of specific flows. When a device fails somewhere in the network, the network traffic status will also change. In order to achieve the best effect, a new control architecture is formed in combination with white box switches. Figure 5 : is a schematic diagram of the interaction structure between an edge switch and a white box network controller in a traffic scheduling system based on a power white box network provided by Embodiment 3 of the present invention, such as Figure 5 As shown,

[0103] The white-box network controller is linked to the edge switch. The white-box network controller has a global perspective. By interacting with the edge switch, it obtains relevant information about the input load intensity of the overall network topology (obtaining the traffic log data regularly uploaded by the edge switch through the system environment environment, and sensing the network status data of the edge switch in real time (link congestion perception)), divides the overall network operation time into multiple time intervals, and transmits the network status data and the optimized target value, i.e., Reward, to the intelligent agent at the end of each time period; the intelligent agent transmits the collected network status data to the neural network, and uses Reward as the result of the last policy decision to optimize the parameters of the neural network, and at the same time generates a decision result, i.e., Action, corresponding to the neural network input at this time; the Action data obtained by the controller is deployed to the network topology according to the relevant strategy, and the updated current traffic threshold is obtained (the current traffic threshold is calculated by the neural network network based on reinforcement learning), and the updated current traffic threshold is sent to the edge switch; the white-box network controller also needs to send the corresponding forwarding strategies for different traffic types to all white-box switches (edge ​​switches and inner switches).

[0104] The edge switch, i.e., the white box switch of the edge layer, is deployed with a traffic classification module and a forwarding module. The traffic classification module is used to identify the traffic type of the traffic data to be scheduled, specifically including: obtaining the traffic data to be scheduled of the end-side device through a parser; collecting the traffic status information (throughput, etc.) of the traffic data to be scheduled through a traffic status information table, parsing the five-tuple information of the traffic data packet to be scheduled, distinguishing different data flows through the hash value of the five-tuple, and inserting the byte amount information corresponding to the data flow into the traffic information table. In order to prevent the traffic from occupying too much storage resources of the switch, the traffic information table information is checked at regular intervals. When it is found that a certain traffic information has not been updated during this period, it is assumed that the data flow has been completed, and the table entry is deleted from the traffic information table to fully utilize the switch memory resources; based on the traffic status information, all the traffic information injected into the network is counted, the bytes sent in this time period are counted in real time or periodically, and the throughput of the switch is calculated, which is sent to the controller as a measure of the load information of the input network, and the link bandwidth ratio is calculated through the statistics of the traffic bytes; the link bandwidth ratio is compared with the current traffic threshold sent by the white box network controller to determine the traffic type of the traffic data to be scheduled; the traffic data to be scheduled is labeled to mark long and short flows. The forwarding module is used to forward the traffic data to be scheduled based on the traffic type of the traffic data to be scheduled according to the forwarding strategy issued by the white box network control, and is responsible for routing selection, which specifically includes: identifying the label of the traffic data to be scheduled, determining the traffic type of the traffic data to be scheduled, parsing and calculating the traffic data to be scheduled based on N stages, and after completing the calculation of the traffic data to be scheduled, processing the traffic data to be scheduled through the inverse parser, and after completing the inverse parsing, forwarding the traffic data to be scheduled to the connected inner layer switch through the forwarding port corresponding to the traffic type of the traffic data to be scheduled according to the forwarding strategy corresponding to the traffic type. It can be understood that since all host traffic is injected into the network through the edge switch, the traffic classification module can be deployed only in the edge switch, and the forwarding module can be deployed only in the inner layer switch.

[0105] Figure 6It is a structural diagram of an inner switch in a traffic scheduling system based on an electric power white box network provided in the third embodiment of the present invention, that is, a structural diagram of a forwarding module deployed in each white box switch. The software architecture of the forwarding module specifically includes a security pre-processing filtering unit, a protocol-independent parsing unit, a traffic type matching unit, a long flow processing unit, a short flow processing unit, a general computing unit, a protocol-independent reverse parsing unit, a long flow forwarding unit, and a short flow forwarding unit. The forwarding module is mapped to a hardware architecture, the protocol-independent parsing unit is mapped to a parser, and the traffic type matching unit, the long flow processing unit, the short flow processing unit, and the general computing unit are mapped to processors of multiple stages, including a key extractor, a matching action table, and multiple arithmetic logic units (Arithmetic Logic Unit, ALU), the protocol-independent reverse parsing unit is mapped to a reverse parser, the long flow forwarding unit is mapped to a long flow forwarding port, and the short flow forwarding unit is mapped to a short flow forwarding port. Considering the pipeline architecture's streaming processing of data packets, placing unique computing units for long and short flows separately between each stage of the pipeline will increase the processing delay of the pipeline. Therefore, each unique computing unit can be dispersed in each pipeline stage to hide the delay it brings, and can choose whether to perform this operation based on the data classification. Each unique computing unit completes the special computing function it is responsible for. After obtaining the inference result at a certain stage, the subsequent stage can use the result to make corresponding decisions and processes. The unique computing unit and the arithmetic logic unit ALU in the action engine are executed in parallel and do not affect each other. It not only retains the high performance of pipeline streaming processing, but also adds optional special computing functions for long and short flows. After the long and short flows are calculated separately, they are forwarded to the matching link by the corresponding port.

[0106] Optionally, the white box network controller 31 is specifically used for:

[0107] Obtaining current network status data of the edge switch;

[0108] Update network parameters according to the current network status data, and determine the current traffic threshold through the updated network parameters;

[0109] The current traffic threshold is sent to each edge switch.

[0110] Optionally, the edge switch 32 is specifically used for:

[0111] Obtaining the traffic data to be scheduled of the terminal device;

[0112] Determine a link bandwidth ratio according to the data stream bytes of the traffic data to be scheduled, and compare the link bandwidth ratio with the current traffic threshold;

[0113] If the link bandwidth ratio is greater than the current traffic threshold, determining that the traffic type of the traffic data to be scheduled is a long flow;

[0114] If the link bandwidth ratio is less than the current traffic threshold, determining that the traffic type of the traffic data to be scheduled is a short flow;

[0115] Add a label corresponding to the traffic type to the traffic data to be scheduled.

[0116] Optionally, the edge switch 32 is further configured to:

[0117] If the traffic type of the traffic data to be scheduled is a long flow, determining a response probability of the edge switch forwarding the traffic data to be scheduled to each associated inner layer switch;

[0118] Determine the inner switch corresponding to the maximum response probability as the target inner switch, and forward the traffic data to be scheduled to the target inner switch through the long flow forwarding port;

[0119] If the traffic type of the traffic data to be scheduled is a short flow, determining a forwarding link weight of the edge switch to forward the traffic data to be scheduled to each associated inner layer switch;

[0120] The inner switch corresponding to the maximum forwarding link weight is determined as the target inner switch, and the traffic data to be scheduled is forwarded to the target inner switch through the short flow forwarding port.

[0121] Optionally, the inner switch 33 is specifically used for:

[0122] Acquire the traffic data to be scheduled, parse the label of the traffic data to be scheduled, and determine the traffic type of the traffic data to be scheduled;

[0123] The corresponding forwarding port is called based on the traffic type to forward the traffic data to be scheduled to the data center service cluster.

[0124] Optionally, the white box network controller 31 is further used for:

[0125] The forwarding strategy corresponding to the traffic type is determined through the white box network controller, and the forwarding strategy is sent to the edge switch and the inner switch.

[0126] The traffic scheduling system based on the power white box network adopted in the present technical solution includes a white box network controller, at least one edge switch and at least one inner switch; the white box network controller is connected to the edge switch and the inner switch respectively, and the edge switch is connected to the inner switch; the system is connected to the end-side device through the edge switch, and is connected to the data center service cluster through the inner switch; the white box network controller is used to determine the current traffic threshold and send the current traffic threshold to the edge switch; the edge switch is used to obtain the traffic data to be scheduled of the end-side device, and determine the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forward the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled; the inner switch is used to forward the traffic data to be scheduled to the data center service cluster. The above technical solution deploys a DDPG-based long and short flow dynamic classification model on the edge switch. On the subsequent inner layer switch, various flows are dynamically isolated and transmitted on different paths according to the labels of different flows, thereby maximizing the throughput of long flows while ensuring low latency of short flows. In the white box switch, fast pipeline resources are used to process the special computing functions required for long and short flows in parallel. The long and short flow special computing units are executed in parallel with the ALU in the action engine without affecting each other. This not only retains the high performance of pipeline stream processing, but also increases the optional function implementation of long and short flows to avoid idle white box switch resources and traffic congestion. This solves the problem that long flow data packets consume a large amount of switch cache space to obtain high throughput, causing short flow data packets to experience head-of-line blocking and cause serious queuing delay problems.

[0127] The traffic scheduling system based on the power white box network provided in the embodiment of the present invention can execute the traffic scheduling method based on the power white box network provided in any embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method.

[0128] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0129] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A traffic scheduling method based on power white box network, characterized in that: Applied to a traffic scheduling system based on a power white box network, the system includes a white box network controller, at least one edge switch and at least one inner switch, the white box network controller is connected to the edge switch and the inner switch respectively, the edge switch is connected to the inner switch, the system is connected to the end-side device through the edge switch, and is connected to the data center service cluster through the inner switch, the method includes: Determine a current traffic threshold through the white box network controller, and send the current traffic threshold to the edge switch; Acquire the traffic data to be scheduled of the end-side device through the edge switch, determine the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forward the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled; The traffic data to be scheduled is forwarded to the data center service cluster through the inner layer switch.

2. The method according to claim 1, characterized in that: The determining of the current traffic threshold and sending the current traffic threshold to the edge switch includes: Obtaining current network status data of the edge switch; Update network parameters according to the current network status data, and determine the current traffic threshold through the updated network parameters; The current traffic threshold is sent to each edge switch.

3. The method according to claim 1, characterized in that The acquiring the to-be-scheduled traffic data of the terminal device, and determining the traffic type of the to-be-scheduled traffic data according to the current traffic threshold, includes: Obtaining the traffic data to be scheduled of the terminal device; Determine a link bandwidth ratio according to the data stream bytes of the traffic data to be scheduled, and compare the link bandwidth ratio with the current traffic threshold; If the link bandwidth ratio is greater than the current traffic threshold, determining that the traffic type of the traffic data to be scheduled is a long flow; If the link bandwidth ratio is less than the current traffic threshold, determining that the traffic type of the traffic data to be scheduled is a short flow; A label corresponding to the traffic type is added to the traffic data to be scheduled.

4. The method according to claim 1, characterized in that The forwarding the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled includes: If the traffic type of the traffic data to be scheduled is a long flow, determining a response probability of the edge switch forwarding the traffic data to be scheduled to each associated inner layer switch; The inner switch corresponding to the maximum response probability is determined as the target inner switch, and the traffic data to be scheduled is forwarded to the target inner switch through the long flow forwarding port.

5. The method according to claim 1, characterized in that The forwarding the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled includes: If the traffic type of the traffic data to be scheduled is a short flow, determining a forwarding link weight of the edge switch to forward the traffic data to be scheduled to each associated inner layer switch; The inner switch corresponding to the maximum forwarding link weight is determined as the target inner switch, and the traffic data to be scheduled is forwarded to the target inner switch through the short flow forwarding port.

6. The method according to claim 1, characterized in that The forwarding the to-be-scheduled traffic data to the data center service cluster includes: Acquire the traffic data to be scheduled, parse the label of the traffic data to be scheduled, and determine the traffic type of the traffic data to be scheduled; The corresponding forwarding port is called based on the traffic type to forward the traffic data to be scheduled to the data center service cluster.

7. The method according to claim 1, characterized in that Also includes: The forwarding strategy corresponding to the traffic type is determined through the white box network controller, and the forwarding strategy is sent to the edge switch and the inner switch.

8. A traffic scheduling system based on power white box network, characterized in that: include: A white box network controller, at least one edge switch and at least one inner switch; the white box network controller is connected to the edge switch and the inner switch respectively, and the edge switch is connected to the inner switch; the system is connected to the end-side device through the edge switch, and is connected to the data center service cluster through the inner switch; The white box network controller is used to determine a current traffic threshold and send the current traffic threshold to the edge switch; The edge switch is used to obtain the traffic data to be scheduled of the end-side device, determine the traffic type of the traffic data to be scheduled according to the current traffic threshold, and forward the traffic data to be scheduled to the inner switch based on the traffic type of the traffic data to be scheduled; The inner switch is used to forward the traffic data to be scheduled to the data center service cluster.

9. The system according to claim 8, characterized in that The white box network controller is specifically used for: Obtaining current network status data of the edge switch; Update network parameters according to the current network status data, and determine the current traffic threshold through the updated network parameters; The current traffic threshold is sent to each edge switch.

10. The system according to claim 8, characterized in that The edge switch is specifically used for: Obtaining the traffic data to be scheduled of the terminal device; Determine a link bandwidth ratio according to the data stream bytes of the traffic data to be scheduled, and compare the link bandwidth ratio with the current traffic threshold; If the link bandwidth ratio is greater than the current traffic threshold, determining that the traffic type of the traffic data to be scheduled is a long flow; If the link bandwidth ratio is less than the current traffic threshold, determining that the traffic type of the traffic data to be scheduled is a short flow; A label corresponding to the traffic type is added to the traffic data to be scheduled.

Images