CDN flow scheduling method, system and device, and storage medium
By constructing a QoS-QoE mapping dataset and training an offline QoE prediction model, the problem of latency-based user experience evaluation in CDN scheduling systems was solved, achieving optimization of global user experience and resource saving.
Patent Information
- Application Number
- CN202511547517.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-28
- Publication Date
- 2026-01-30
AI Technical Summary
Existing CDN scheduling systems evaluate user experience based on latency, which fails to achieve global user experience optimization and is not cost-effective in improving user experience.
Construct a QoS-QoE mapping dataset, train a QoE prediction model offline, mount it to a CDN scheduling system, use real-time QoS data to predict QoE quality, and generate scheduling policies.
It achieves optimal overall user experience, improves the QoE level of audio and video playback, and saves manpower and resource consumption.
Smart Images
Figure CN121442004A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of audio and video content distribution technology, and in particular to a CDN traffic scheduling method, system, device, and storage medium. Background Technology
[0002] CDN scheduling systems distribute user traffic to different server nodes by balancing factors such as server load, network quality, and bandwidth costs. The basic principle is shown in the following formula (1):
[0003] (1)
[0004] According to this formula, the CDN scheduling system maps the domain names requested by users in different regions to the corresponding server IPs, thereby directing user request traffic to the corresponding nodes. ).
[0005] The basic structure of existing CDN scheduling systems is as follows: Figure 1 As shown, the CDN scheduling system monitors the QoS quality from user terminal 10 to node 20 in each user region through quality sensors distributed across each node, and transmits this information to the scheduling policy generator 30. Node 20 includes different levels of cache nodes such as L1 nodes, L2 nodes, and origin server nodes. The scheduling policy generator 30 matches the "domain name-user region" with different nodes based on the QoS quality from the user region to different nodes, taking into account the traffic capacity of each node and the traffic requirements of different domain names. Under the premise of satisfying traffic constraints, it ensures the optimal network quality from the user terminal to the server node, and finally generates the scheduling policy shown in formula (1), which is then sent to the DNS server 40. When a user requests the content of the corresponding domain name, the DNS resolves the domain name and returns the corresponding node IP for the user to access.
[0006] CDN systems aim to provide users with a better quality of experience (QoE), but the quality sensor can only obtain information related to the quality of service (QoS). Since QoS and QoE are correlated, the CDN scheduling system strives to ensure a better user experience by optimizing QoS.
[0007] Because latency is easier to measure, current CDN scheduling systems mostly use the latency from the user region to the node as the quality evaluation standard for scheduling. They use latency to evaluate user experience, hoping to guarantee QoE by minimizing the latency from the user to the node. This type of approach mainly suffers from two problems:
[0008] (1) Latency cannot truly represent QoE. Latency is a QoS metric, and while QoS quality can measure user experience to some extent, the relationship between QoS and QoE is complex. User experience (QoE) is affected not only by QoS but also by other factors. Using latency as a quality evaluation standard for traffic scheduling can minimize global user access latency, but the difference between QoS and QoE may prevent the optimization of the global user experience.
[0009] (2) Costs and benefits are not proportional. CDN scheduling systems believe that improving QoS can achieve a satisfactory QoE level for users, but existing research has found that improving QoS does not bring about a significant improvement in QoE, and may even lead to a decrease in QoE quality, such as... Figure 2 As shown, CDN traffic scheduling mechanisms, while optimizing latency through various means, do not necessarily lead to improved user experience. The investment of costs may not yield the expected benefits. Summary of the Invention
[0010] To address the shortcomings of existing technologies, this invention proposes a CDN traffic scheduling method, system, device, and storage medium. This method enables the CDN scheduling policy generator to generate scheduling policies using QoE as the quality evaluation standard, thereby achieving optimal global user experience.
[0011] This invention provides a CDN traffic scheduling method, comprising:
[0012] Collect QoS and QoE data, construct a QoS-QoE mapping dataset, and train an offline QoE prediction model.
[0013] The QoE prediction model is mounted to the CDN scheduling system, and the real-time QoS data collected by the quality sensor is input into the QoE prediction model to predict the QoE quality.
[0014] A CDN scheduling strategy is generated based on the predicted QoE quality.
[0015] In one embodiment of the present invention, the collection of QoS data and QoE data includes: collecting QoS data from CDN nodes and collecting QoE data from clients in cooperation with content providers.
[0016] In one embodiment of the present invention, the construction of the QoS-QoE mapping relationship dataset includes: establishing an association between the QoS data and the QoE data using at least one of IP address, communication time, and access method as constraints; performing data cleaning and data smoothing processing on the associated data to obtain the QoS-QoE mapping relationship dataset.
[0017] In one embodiment of the present invention, the training of the QoE prediction model includes: using QoS data in the QoS-QoE mapping dataset, as well as time information and access method information during data matching, as model input features; using a machine learning model to fit the mapping relationship between QoS and QoE; training to generate the QoE prediction model; and optimizing the model error.
[0018] In one embodiment of the present invention, mounting the QoE prediction model to the CDN scheduling system includes: embedding the QoE prediction model into the CDN scheduling system by modifying the data flow from the quality sensor to the scheduling policy generator in the CDN scheduling system; wherein the QoE prediction model can be independently unloaded or replaced.
[0019] In one embodiment of the present invention, the real-time QoS data collected by the quality sensor is the QoS quality data from each user area to the CDN node, and the QoS quality data includes at least one of latency, packet loss rate, and bandwidth.
[0020] The QoE prediction model outputs the QoE quality from the corresponding user region to the CDN node in real time.
[0021] In one embodiment of the present invention, generating a CDN scheduling strategy based on the predicted QoE quality includes: inputting the predicted QoE quality into a scheduling strategy generator, wherein the scheduling strategy generator uses the predicted QoE quality as a quality evaluation standard to plan and generate a scheduling strategy.
[0022] In another aspect, the present invention provides a CDN scheduling system, including a quality sensor, a QoE prediction model, a scheduling policy generator, and a DNS server;
[0023] The quality sensor collects real-time QoS data from the user area to the CDN node and transmits the real-time QoS data to the QoE prediction model.
[0024] The QoE prediction model receives the real-time QoS data and predicts the corresponding QoE quality, and transmits the predicted QoE quality to the scheduling policy generator; wherein, QoS data and QoE data are collected, a QoS-QoE mapping relationship dataset is constructed, and the QoE prediction model is obtained through offline training.
[0025] The scheduling policy generator generates a scheduling policy based on the predicted QoE quality and sends the scheduling policy to the DNS server.
[0026] The DNS server receives the scheduling policy and returns the corresponding CDN node IP to the user.
[0027] Another aspect of the present invention provides a CDN traffic scheduling device, comprising:
[0028] The offline training module is used to collect QoS data and QoE data, construct a QoS-QoE mapping dataset, and train the QoE prediction model offline.
[0029] The online prediction module is used to mount the QoE prediction model to the CDN scheduling system, input the real-time QoS data collected by the quality sensor into the QoE prediction model, and predict the QoE quality.
[0030] The strategy generation module is used to generate CDN scheduling strategies based on the predicted QoE quality.
[0031] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the CDN traffic scheduling method described above.
[0032] In another aspect, the present invention provides a computer program product, comprising a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps of the above-described CDN traffic scheduling method.
[0033] As can be seen from the above solutions, the advantages of the present invention are:
[0034] The CDN traffic scheduling method disclosed in this invention constructs a QoS-QoE mapping dataset and trains a QoE prediction model offline. This QoE prediction model is then integrated into the CDN scheduling system. Real-time QoS data collected by a quality sensor is input into the QoE prediction model to predict the QoE quality. Finally, a CDN scheduling strategy is generated based on the predicted QoE quality. This method ensures the real-time provision of QoE information to the scheduling system, provides a more accurate user experience assessment for the scheduling strategy, and, since the QoE prediction model is integrated into the CDN scheduling system, only the data flow needs to be changed without requiring system reconstruction, saving manpower and resources. Attached Figure Description
[0035] Figure 1 A schematic diagram of the structure of a CDN scheduling system in the prior art is shown;
[0036] Figure 2 This diagram illustrates the QoS-QoE variation of the CDN traffic scheduling mechanism based on latency evaluation in existing technologies.
[0037] Figure 3 A schematic diagram of the overall flow of the CDN traffic scheduling method provided in Embodiment 1 of the present invention is shown;
[0038] Figure 4 It shows Figure 3 A schematic diagram showing the location and relationship of QoS data and QoE data in step S1;
[0039] Figure 5 This diagram illustrates the mapping between QoS data and QoE data in step S1.
[0040] Figure 6 This diagram illustrates the offline training of the QoE prediction model in step S1.
[0041] Figure 7 It shows Figure 3 Schematic diagram of QoE prediction model mounting in step S2;
[0042] Figure 8 A schematic diagram of the CDN scheduling system provided in Embodiment 2 of the present invention is shown;
[0043] Figure 9 A schematic diagram of the CDN traffic scheduling device provided in Embodiment 3 of the present invention is shown;
[0044] The attached figures are labeled as follows:
[0045] 10: Client-side;
[0046] 20: Node;
[0047] 30: Scheduling policy generator;
[0048] 40: DNS server;
[0049] 50: QoE prediction model;
[0050] 300: CDN traffic scheduling device;
[0051] 310: Offline training module;
[0052] 320: Online prediction module;
[0053] 330: Strategy Generation Module. Detailed Implementation
[0054] It should be noted that, in this invention, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.
[0055] In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0056] Since QoE can only be collected by the client, and the quality sensor can only obtain QoS information, it is necessary to first break down the information asymmetry barrier between the server and the client. A conventional approach is to establish a communication mechanism between the client and the server, informing the quality sensor of the QoE obtained by the client. However, this method usually requires architectural modifications to both the client and server, incurring significant time, manpower, and material costs. Furthermore, the client is typically controlled by the CDN content provider, necessitating collaboration with the content provider for adaptation. Simultaneously, the generation of scheduling policies needs a certain degree of real-time capability to ensure rapid adjustment of scheduling policies in real-world scenarios, such as network failures, avoiding significant impact on user experience. However, communication between the client and server typically has a certain lag, making it difficult to meet these requirements.
[0057] To address this issue and minimize modifications to the existing framework while meeting the real-time requirements of scheduling policy generation, this invention designs an offline-trained, online-predictive QoE prediction model. Based on QoS and QoE data, a QoE prediction model is generated through offline training. This model can describe QoE quality using QoS data perceived by the CDN, enabling the CDN to obtain QoE-related information. Subsequently, this QoE prediction model is integrated into the CDN scheduling system. With simple adjustments, the QoS information collected by the quality sensor is input into the prediction model, and the predicted QoE is input into the scheduling policy generator. This allows the scheduling policy generator to perform scheduling planning based on real-time predicted QoE rather than QoS.
[0058] Example 1:
[0059] See Figure 3 As shown, Figure 3 The diagram shows the overall flow of a CDN traffic scheduling method provided in an embodiment of the present invention.
[0060] A CDN traffic scheduling method is proposed, which is based on QoE-based traffic scheduling. This method enables the CDN scheduling policy generator to generate scheduling policies using QoE as the quality evaluation standard, thereby achieving the optimization of the global user experience.
[0061] The CDN traffic scheduling method specifically includes the following steps:
[0062] Step S1: Collect QoS data and QoE data, construct a QoS-QoE mapping dataset, and train the QoE prediction model offline.
[0063] Step S2: Mount the QoE prediction model to the CDN scheduling system, input the real-time QoS data collected by the quality sensor into the QoE prediction model, and predict the QoE quality.
[0064] Step S3: Generate a CDN scheduling strategy based on the predicted QoE quality.
[0065] In this embodiment, the offline training component is responsible for collecting QoS and QoE data and using a machine learning model to fit the mapping relationship between QoS and QoE, generating a QoE prediction model. The online prediction component mounts the trained QoE prediction model onto the actual CDN scheduling system and modifies the data flow of the scheduling architecture, enabling the scheduling policy generator to perform scheduling based on the predicted QoE quality. The two components work together to allow the CDN scheduling system to obtain the QoE quality status and plan scheduling policies in real time based on QoE.
[0066] In one embodiment, in step S1, the prediction model requires QoS and QoE data for training, and the collection of QoS and QoE data includes: collecting QoS data from CDN nodes and collecting QoE data from clients in cooperation with content providers.
[0067] In a specific application, taking video services as an example, the location and relationship of QoS and QoE data are as follows: Figure 4 As shown. QoS data can be directly collected from the scheduling system's quality sensor. The scheduling system's quality sensor uses active and passive probing methods to collect QoS quality data from each user area to all CDN nodes in the network, such as latency, packet loss rate, and bandwidth. QoE data is collected from clients that the quality sensor cannot reach. Due to the information asymmetry between clients and servers, this data needs to be obtained in cooperation with CDN content providers. Both QoS and QoE data are collected offline, without requiring architectural adjustments to clients and servers or consuming resources to establish communication mechanisms.
[0068] Because they are obtained through different methods, QoS data and QoE data are relatively independent and cannot be accurately correlated. Therefore, it is necessary to establish a QoS-QoE mapping dataset. Both QoS data and QoE data contain information such as communication time, the locations of the communicating parties, and the requested domain name. In one embodiment, at least one of IP address, communication time, and access method is used as a constraint to establish the association between the QoS data and QoE data, and to match the QoS and QoE data. The matching method is as follows: Figure 5As shown. The more factors considered during matching, the more accurate the QoS-QoE mapping dataset will be. After the matching operation is completed, unmatched data is removed. The obtained QoS-QoE associated data is then cleaned and smoothed to finally construct the QoS-QoE mapping dataset.
[0069] For the constructed QoS-QoE mapping dataset, in one embodiment, machine learning methods are used to fit the QoS-QoE mapping relationship. The QoS data in the QoS-QoE mapping dataset is used as the primary input. To reduce errors and accurately represent the relationship between QoS and QoE, the time and access method information during QoS and QoE data matching are also input as features into the model. The machine learning model fits the QoS-QoE mapping relationship, training to generate the QoE prediction model. The model predicts the corresponding QoE quality based on the input QoS data and the time and access method information during data matching, thereby helping the scheduling policy generator evaluate user experience. The dataset is divided into training and testing sets, and the model training and evaluation process is as follows: Figure 6 As shown, through the above offline data collection and model training process, a QoS-QoE mapping dataset can be constructed without modifying the existing architecture, and a QoE prediction model can be obtained through offline training, thereby helping the CDN scheduling system to evaluate QoE quality based on QoS information.
[0070] In one embodiment, in step S2, the QoE prediction model is mounted to the CDN scheduling system through online prediction, and the real-time QoS data collected by the quality sensor is input into the QoE prediction model. The QoE prediction model outputs the QoE quality from the corresponding user area to the CDN node in real time.
[0071] In one specific implementation, to enable the CDN to acquire user experience information, a trained QoE prediction model needs to be embedded as a component into the CDN scheduling system. This model predicts QoE based on the QoS information it possesses, which is then used for quality assessment during scheduling. Embedding the QoE prediction model should minimize modifications to the existing CDN scheduling system architecture. Furthermore, considering that the relationship between QoS and QoE may change, and the QoE prediction model will continuously optimize and iterate, the embedding of the QoE prediction model also needs to ensure flexibility and scalability. To address this, a mounting method is used to embed the model, such as... Figure 7As shown, by modifying the data flow from the quality sensor to the scheduling policy generator in the CDN scheduling system, the QoE prediction model can be embedded into the CDN scheduling system without requiring reconstruction or complex modifications to the entire CDN scheduling system, thus greatly reducing maintenance and upgrade costs. Furthermore, the QoE prediction model can be independently uninstalled or replaced, a process that is seamless for other components.
[0072] In one embodiment, in step S3, after the QoE prediction model is mounted, the real-time QoS data detected by the quality sensor is input into the QoE prediction model in real time. The model performs real-time QoE quality prediction, and the predicted QoE quality is input into the scheduling policy generator. The scheduling policy generator uses the predicted QoE quality as a quality evaluation standard to plan and generate scheduling policies. The input quality information has changed from QoS to QoE. Both are quality evaluation indicators and are essentially not significantly different, therefore they do not affect the workflow of the policy generator, and the real-time performance of the entire CDN scheduling system remains unaffected.
[0073] In summary, the CDN traffic scheduling method provided by this invention constructs a QoS-QoE mapping dataset and trains a QoE prediction model offline. This QoE prediction model is then integrated into the CDN scheduling system, and real-time QoS data collected by a quality sensor is input into the model to predict QoE quality. Finally, a CDN scheduling strategy is generated based on the predicted QoE quality. This method ensures the real-time provision of QoE information to the scheduling system, provides a more accurate user experience assessment for the scheduling strategy, and, since the QoE prediction model is integrated into the CDN scheduling system, only the data flow needs to be changed without requiring system reconstruction, saving manpower and resources.
[0074] As shown in Table 1 below, taking audio and video playback as an example, compared with the prior art, the present invention can improve the QoE of user audio and video playback when the user's QoS level is similar.
[0075] Table 1
[0076]
[0077] Example 2:
[0078] This embodiment is a system embodiment corresponding to the above method embodiment, such as... Figure 8 As shown, Figure 8
[0079] A schematic diagram of a CDN scheduling system according to an embodiment of the present invention is shown. This system implementation can be combined with the above-described method implementation. The relevant technical details mentioned in the above method implementation remain valid in this system implementation, and will not be repeated here to avoid repetition.
[0080] A CDN scheduling system includes a quality sensor, a QoE prediction model, and a scheduling policy generator.
[0081] Devices and DNS servers;
[0082] The quality sensor is used to collect real-time QoS data from each user terminal 10 to the CDN node 20 in the user area, and transmit the real-time QoS data to the QoE prediction model 50.
[0083] The QoE prediction model 50 is used to receive the real-time QoS data and predict the corresponding QoE quality, and transmit the predicted QoE quality to the scheduling policy generator 30; wherein, QoS data and QoE data are collected, a QoS-QoE mapping relationship dataset is constructed, and the QoE prediction model is obtained through offline training.
[0084] The scheduling policy generator 30 is used to generate a scheduling policy based on the predicted QoE quality and to send the scheduling policy to the DNS server 40.
[0085] The DNS server 40 is used to receive the scheduling policy and return the corresponding CDN node IP to the user segment 10.
[0086] Example 3:
[0087] This embodiment corresponds to the device embodiment described above, such as... Figure 9 As shown, Figure 9 Show
[0088] A schematic diagram of a CDN traffic scheduling device according to an embodiment of the present invention is provided. This device embodiment can be implemented in conjunction with the above-described method embodiment. The relevant technical details mentioned in the above method embodiment remain valid in this device embodiment, and will not be repeated here to avoid repetition.
[0089] A CDN traffic scheduling device 300 includes:
[0090] The offline training module 310 is used to collect QoS data and QoE data, construct a QoS-QoE mapping dataset, and train offline to obtain a QoE prediction model.
[0091] The online prediction module 320 is used to attach the QoE prediction model to the CDN scheduling system, input the real-time QoS data collected by the quality sensor into the QoE prediction model, and predict the QoE quality.
[0092] The strategy generation module 330 is used to generate a CDN scheduling strategy based on the predicted QoE quality.
[0093] This device embodiment can be implemented in conjunction with the implementation methods described above. The relevant technical details mentioned in the implementation methods of the above embodiments remain valid in the implementation methods of this method embodiment, and will not be repeated here to avoid repetition. It should be noted that the scope of the methods and devices in the embodiments of the present invention is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be applied, omitted, or combined. Furthermore, features described with reference to certain examples may be combined in other examples.
[0094] This invention also provides a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the CDN traffic scheduling method described above, and achieve the same technical effect.
[0095] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the CDN traffic scheduling method described above, and achieves the same technical effect.
[0096] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.
[0097] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software program product is stored in a storage medium and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.
[0098] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.
Claims
1. A CDN traffic scheduling method, characterized in that, The method comprises: collecting QoS data and QoE data, constructing a QoS-QoE mapping relationship dataset, and training offline to obtain a QoE prediction model; mounting the QoE prediction model to a CDN scheduling system, inputting real-time QoS data collected by a quality sensor into the QoE prediction model, and predicting QoE quality; generating a CDN scheduling strategy based on the predicted QoE quality.
2. The method of claim 1, wherein, The collecting of the QoS data and the QoE data comprises: collecting QoS data from CDN nodes and collecting QoE data from clients in cooperation with content providers.
3. The method of claim 1, wherein, The constructing of the QoS-QoE mapping relationship dataset comprises: taking at least one of an IP address, a communication time, and an access mode as a constraint condition, establishing an association between the QoS data and the QoE data, performing data cleaning and data smoothing processing on the associated data, and obtaining the QoS-QoE mapping relationship dataset.
4. The method of claim 1, wherein, The training of the QoE prediction model comprises: taking QoS data in the QoS-QoE mapping relationship dataset and time information and access mode information when the data are matched as model input features, fitting a mapping relationship between QoS and QoE by using a machine learning model, training to generate the QoE prediction model, and optimizing model errors.
5. The method of claim 1, wherein, The mounting of the QoE prediction model to the CDN scheduling system comprises: embedding the QoE prediction model in the CDN scheduling system by modifying a data flow direction from the quality sensor to a scheduling strategy generator in the CDN scheduling system; and wherein the QoE prediction model can be independently unloaded or replaced.
6. The method of claim 1, wherein, The real-time QoS data collected by the quality sensor is QoS quality data from each user area to a CDN node, and the QoS quality data comprises at least one of a delay, a packet loss rate, and a bandwidth. The QoE prediction model outputs QoE quality from a corresponding user area to a CDN node in real time.
7. The method of claim 1, wherein, The generating of the CDN scheduling strategy based on the predicted QoE quality comprises: inputting the predicted QoE quality into a scheduling strategy generator, taking the predicted QoE quality as a quality evaluation standard, and planning and generating a scheduling strategy.
8. A CDN scheduling system, characterized by, The system comprises a quality sensor, a QoE prediction model, a scheduling strategy generator, and a DNS server. The quality sensor collects real-time QoS data from a user area to a CDN node, and transmits the real-time QoS data to the QoE prediction model. The QoE prediction model receives the real-time QoS data and predicts corresponding QoE quality, and transmits the predicted QoE quality to the scheduling strategy generator; wherein QoS data and QoE data are collected, a QoS-QoE mapping relationship dataset is constructed, and a QoE prediction model is trained offline. The scheduling strategy generator generates a scheduling strategy based on the predicted QoE quality, and issues the scheduling strategy to the DNS server. The DNS server receives the scheduling strategy and returns a corresponding CDN node IP to a user.
9. A CDN traffic scheduling apparatus, characterized by, The method comprises: An offline training module is configured to collect QoS data and QoE data, build a QoS-QoE mapping relationship dataset, and train an offline QoE prediction model; An online prediction module is configured to mount the QoE prediction model to a CDN scheduling system, input real-time QoS data collected by a quality sensor into the QoE prediction model, and predict a QoE quality; A policy generation module is configured to generate a CDN scheduling policy according to the predicted QoE quality. 10.A computer readable storage medium having stored thereon a computer program which, when executed by a processor, causes the processor to carry out the steps of any one of claims 1-7.