CN117033693B - Method and system for cloud processing in mixed mode - Google Patents
Method and system for cloud processing in mixed mode Download PDFInfo
- Publication number
- CN117033693B CN117033693B CN202311289674.3A CN202311289674A CN117033693B CN 117033693 B CN117033693 B CN 117033693B CN 202311289674 A CN202311289674 A CN 202311289674A CN 117033693 B CN117033693 B CN 117033693B
- Authority
- CN
- China
- Prior art keywords
- user
- data
- music
- cloud
- requirements
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000012545 processing Methods 0.000 title claims abstract description 74
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 81
- 230000004044 response Effects 0.000 claims abstract description 62
- 238000004364 calculation method Methods 0.000 claims abstract description 55
- 230000003993 interaction Effects 0.000 claims abstract description 47
- 230000008859 change Effects 0.000 claims abstract description 42
- 238000003860 storage Methods 0.000 claims abstract description 40
- 238000013135 deep learning Methods 0.000 claims abstract description 18
- 238000012795 verification Methods 0.000 claims abstract description 8
- 230000005540 biological transmission Effects 0.000 claims description 73
- 230000006399 behavior Effects 0.000 claims description 52
- 238000013468 resource allocation Methods 0.000 claims description 36
- 238000012549 training Methods 0.000 claims description 35
- 238000005457 optimization Methods 0.000 claims description 28
- 238000011156 evaluation Methods 0.000 claims description 19
- 230000000694 effects Effects 0.000 claims description 18
- 238000012544 monitoring process Methods 0.000 claims description 18
- 238000009826 distribution Methods 0.000 claims description 14
- 230000007246 mechanism Effects 0.000 claims description 14
- 238000003066 decision tree Methods 0.000 claims description 11
- 238000010219 correlation analysis Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 8
- 238000013500 data storage Methods 0.000 claims description 6
- 238000003064 k means clustering Methods 0.000 claims description 4
- 230000002787 reinforcement Effects 0.000 claims description 4
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000001172 regenerating effect Effects 0.000 claims description 3
- 230000002452 interceptive effect Effects 0.000 claims 1
- 238000003672 processing method Methods 0.000 claims 1
- 238000004458 analytical method Methods 0.000 description 16
- 230000002159 abnormal effect Effects 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 8
- 239000011435 rock Substances 0.000 description 8
- 230000005856 abnormality Effects 0.000 description 7
- 230000008451 emotion Effects 0.000 description 7
- 238000005206 flow analysis Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000012546 transfer Methods 0.000 description 4
- 239000002699 waste material Substances 0.000 description 4
- 239000008186 active pharmaceutical agent Substances 0.000 description 3
- 238000012098 association analyses Methods 0.000 description 3
- 238000011217 control strategy Methods 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000001932 seasonal effect Effects 0.000 description 3
- 241001522296 Erithacus rubecula Species 0.000 description 2
- 238000005094 computer simulation Methods 0.000 description 2
- 230000008602 contraction Effects 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000003203 everyday effect Effects 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000003139 buffering effect Effects 0.000 description 1
- 230000002354 daily effect Effects 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/63—Querying
- G06F16/635—Filtering based on additional data, e.g. user or group profiles
- G06F16/637—Administration of user profiles, e.g. generation, initialization, adaptation or distribution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5072—Grid computing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Bioethics (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computer Hardware Design (AREA)
- Computer Security & Cryptography (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Telephonic Communication Services (AREA)
Abstract
The application provides a method and a system for cloud processing in a mixed mode, comprising the following steps: according to the operation habit of the user, the personalized requirements of the user on privacy protection, tone quality and response time are obtained; according to the user requirements, determining data encryption, identity verification and voice quality enhancement methods required by the user, and judging network requirements, storage requirements and calculation requirements of cloud calculation required by music data; according to the user interaction data, predicting the music preference of the user, preloading the music content which can be played by the user in advance through mixed cloud computing, and reducing the buffer time; if the number of users or the music preference changes to cause the data to be preloaded to change, the change of tone quality, response time and calculation cost is estimated again; and adopting a deep learning algorithm to dynamically adjust public cloud and private cloud allocation recommendation strategies so as to obtain an optimal data preloading scheme.
Description
Technical Field
The invention relates to the technical field of information, in particular to a method and a system for cloud processing in a mixed mode.
Background
As users' demands for music playing services continue to increase, existing systems have problems. Firstly, due to the huge volume of music data, the traditional data transmission mode cannot meet the requirements of users in real time, so that the situation of buffering occurs during music playing, and the user experience is reduced. Existing music playback platforms face challenges in terms of network requirements, storage requirements, and computing requirements when handling large amounts of music data. Processing music data requires a strong computing power and memory space, while satisfying the high-speed transmission requirements of music data. However, existing platforms often do not have an efficient way to balance the cost of computing resources and data transfer. Existing music playback platforms lack the ability to dynamically adjust the data preloading strategy. In addition, as the number of users and the music preference change, the conventional system cannot dynamically adjust the data preloading strategy, and cannot provide an optimal data preloading scheme. The existing music data does not use a resource scheduling strategy of hybrid cloud computing, and cannot provide corresponding computing resources. The existing music data cannot meet the mixed cloud computing requirements and cannot well distribute data transmission between public cloud and private cloud. When the number of users or the music preference change causes that the preloaded data needs to be changed, the preloaded data is not evaluated again in time, and the technology of individual requirements cannot be acquired according to the operation habit of the users, so that the user experience of the music playing service is poor. The existing music data are not subjected to dynamic adjustment of public cloud and private cloud distribution, and an optimal data preloading scheme cannot be obtained. In summary, the existing music playing platform has the problems of difficulty in meeting the personalized requirements of users, difficulty in processing a large amount of music data and lack of dynamic adjustment of the data preloading strategy.
Disclosure of Invention
The invention provides a method and a system for cloud processing in a mixed mode, which mainly comprise the following steps:
according to the operation habit of the user, the personalized requirements of the user on privacy protection, tone quality and response time are obtained; according to the user requirements, determining data encryption, identity verification and voice quality enhancement methods required by the user, and judging network requirements, storage requirements and calculation requirements of cloud calculation required by music data; according to the user interaction data, predicting the music preference of the user, preloading the music content which can be played by the user in advance through mixed cloud computing, and reducing the buffer time; according to the mixed cloud computing requirements of music data, computing cost and flow cost meeting the requirements are generated, and a data transmission distribution mode of computing service between public cloud and private cloud is adjusted; if the number of users or the music preference changes to cause the data to be preloaded to change, the change of tone quality, response time and calculation cost is estimated again; and adopting a deep learning algorithm to dynamically adjust public cloud and private cloud allocation recommendation strategies so as to obtain an optimal data preloading scheme.
Preferably, the obtaining, according to the user operation habit, the personalized requirements of the user on privacy protection, tone quality and response time includes:
Acquiring historical operation data of user privacy protection habits, and acquiring user habit data; judging the proportion of non-default options in the privacy setting of the user according to the output of the habit data of the user, and if the proportion is greater than a preset threshold value, adjusting the default setting of the system; analyzing operation data of a user, identifying specific frequency response preference under sound quality requirements, and dividing the user into different sound quality requirement groups through a K-means clustering algorithm; according to the K-means algorithm result, if a certain group selects high tone quality playing frequency higher than a preset proportion, the default tone quality output of the system is improved for the group; continuously using the K-means result, and carrying out correlation analysis on tone quality and network bandwidth on each group; predicting the response time tolerance of the users of each group under different network conditions according to the correlation analysis; acquiring predicted response time tolerance, matching with the operation speed of a user, and optimizing a back-end data processing strategy; collecting data of system response time and user satisfaction, and analyzing correlation by using CNN; according to the output result of the CNN, carrying out detail optimization on the response time of the system; and according to the optimization result of each step, comprehensively distributing weights of tone quality and response time, and analyzing comprehensive user satisfaction as a parameter of system optimization.
Preferably, the determining, according to the user requirement, the data encryption, the authentication and the voice enhancement method required by the user, and determining the network requirement, the storage requirement and the calculation requirement of cloud calculation required by the music data include:
acquiring an integrity verification mechanism of data interaction, and adopting a secure HashAlgorithm method to ensure the data integrity; judging an advanced algorithm strategy of user privacy protection, and optimizing multimedia transmission of music data by adopting a music flow control strategy according to a data protection result to ensure optimized transmission of flow; judging the algorithm complexity of real-time audio processing, and if the algorithm complexity is higher than a preset algorithm complexity threshold, adding an additional audio processing calculation unit; and according to the user identity verification result, the requirements of cloud computing resources are evaluated, and whether more computing resources need to be dynamically allocated or not is determined.
Preferably, the predicting the music preference of the user according to the user interaction data, preloading the music content that will be played by the user in advance through hybrid cloud computing, and reducing the buffer duration includes:
acquiring the latest music song listening behavior of a user according to the timestamp details and the historical behavior mode of the user interaction data; judging the type of music liked by the user by using the metadata and the audio characteristics of the music; predicting the next song listening time of the user according to the interaction frequency and the activity of the user in different scenes; acquiring a network flow mode and a real-time network environment state of music playing through a network monitoring tool Wireshark and network state APIs; dynamically determining an optimal storage location of music content based on resource usage and allocation policies of each node in the hybrid cloud computing environment; if the user playing behavior shows seasonal and specific event-driven modes, preloading music content related to the event; obtaining the current hottest music content through popular trend of the music content and social media heat analysis; determining a favorite music list of a user by adopting the repeated play times and the complete play rate in the play history of the user; if the correlation analysis result between the music content and the emotion state of the user is positive correlation, preloading the music; acquiring a multi-source data integration strategy of music content, and integrating music data from different platforms and sources; providing computing resources for a computing model through a resource scheduling strategy of hybrid cloud computing; determining the format and quality of the preloaded music according to the suitability of the multi-terminal equipment for playing the music; the method comprises the steps that a data caching strategy of preloading music content is adopted, and music in a favorite track list of a user is loaded onto user equipment in advance; constructing a user behavior prediction model through a decision tree algorithm, if a real-time updating mechanism of the user behavior prediction model judges that a user can change music preference, analyzing new user interaction data by using the decision tree algorithm, and updating a track recommendation list of the user; judging whether the preloaded music can be smoothly played by a user according to an usability assessment mechanism of the preloaded music through a data compression and transmission strategy of the music content; if the real-time performance and timeliness of the user interaction data obtain a new interaction record, immediately updating a user behavior prediction model, and continuously optimizing a music recommendation list of the user; further comprises: providing computing resources for a computing model through a resource scheduling strategy of hybrid cloud computing; and constructing a user behavior prediction model by using a decision tree algorithm, and updating a track recommendation list of the user if the real-time updating mechanism judges that the user can change the music preference.
The resource scheduling strategy through hybrid cloud computing provides computing resources for a computing model, and specifically comprises the following steps:
and determining a computing resource pool in the mixed cloud environment according to the requirements and availability requirements of the computing model, wherein the computing resource pool comprises private cloud and public cloud resources. And acquiring the resource requirement and the optimization target of the calculation model by monitoring and analyzing the load condition and the performance index of the calculation model. According to the load balancing algorithm, judging which computing resources the computing tasks should be allocated to so as to realize load balancing and maximize the utilization rate of the resources. And judging whether the scale of the computing resource needs to be automatically expanded or reduced according to the elastic adjustment rule and the strategy so as to adapt to the load change of the computing model. And determining the data storage position to be accessed by the computing model according to the position and the moving cost of the data, and improving the data access efficiency through data copying or caching. And optimizing a resource scheduling strategy according to the load condition and the performance index of the computing resource so as to improve the resource utilization rate and the performance of the computing model. According to the security requirement, the data and privacy security of the calculation model are ensured by adopting a mode of encrypting, transmitting and storing the data. The availability of the computing resources is determined through real-time monitoring and adjustment, so that continuous operation of the computing model is ensured, and stable performance is provided. And determining a final resource allocation scheme according to the demand of the calculation model and the result of the resource scheduling strategy, and implementing resource scheduling. Resource scheduling policies are periodically evaluated and optimized to accommodate changes in computing models and hybrid cloud environments.
The step of constructing a user behavior prediction model by using a decision tree algorithm, and if a real-time updating mechanism judges that a user can change music preference, updating a track recommendation list of the user specifically comprises the following steps:
and acquiring user interaction data and preprocessing, including data cleaning, missing value processing and feature selection operation, according to the user basic information, the user historical behavior data, the current interaction data, the music attribute and the context information. Extracting valuable features from the preprocessed user interaction data by adopting feature extraction, and playing singer and type information in the record recently played by the user. And obtaining a result of feature extraction, taking the result as input, performing model training through a decision tree algorithm, and establishing a user behavior prediction model. In the training process, according to the labels of the historical behavior data and the current interaction data of the user, whether the user changes the music preference or not carries out supervised learning on the user behavior prediction model. And determining a trained user behavior prediction model, predicting new user interaction data by using the trained user behavior prediction model, and judging whether the user can change the music preference. The prediction result is a probability value indicating the likelihood of the user changing the music preference. And updating the track recommendation list of the user according to the prediction result. If the predicted result shows that the user can replace the music preference, regenerating a song recommendation list according to the new preference of the user so as to improve the satisfaction degree and the user experience of the user.
Preferably, the generating, according to the mixed cloud computing requirement of the music data, a computing cost and a traffic cost that meet the requirement, and adjusting a data transmission allocation manner of the computing service between the public cloud and the private cloud, includes:
acquiring the size and type of the data according to the analysis requirements of the scale and the format of the music data; judging computing nodes and storage positions by adopting computing resource quota and distribution of the hybrid cloud; adjusting the data processing priority by calculating the time delay requirement and response time of the task; generating predicted calculation cost according to calculation efficiency and cost difference of public cloud and private cloud; determining the flow capacity and the rate requirement of data transmission through a storage mode and an access mode of music data; optimizing the matching degree of the data redundancy and the backup strategy in the hybrid cloud by adopting a load balancing algorithm, and ensuring the data integrity and accessibility; acquiring real-time processing and storage requirements of music data, and determining the duration and format of data storage; adjusting the data transmission speed according to the bandwidth requirement and the peak flow of the data transmission; determining a time window of data transmission according to network stability evaluation between the public cloud and the private cloud; making a processing strategy when data is in error through error tolerance rate and rollback mechanism of music data processing; and adopting a deep reinforcement learning algorithm, and dynamically adjusting data transmission allocation according to a self-adaptive load balancing strategy of the hybrid cloud so as to optimize cost and efficiency.
Preferably, if the number of users or the music preference change to change the data to be preloaded, the change of the sound quality, the response time and the calculation cost is evaluated again, including:
judging whether abnormal user growth or abnormal user reduction exists or not according to the user quantity change data and the historical trend thereof; acquiring the data size and the coding form of various types of music, and judging the influence on the tone quality; determining the optimal number of concurrent requests through correlation analysis of the number of users and the concurrent amount of the server requests; according to the influence relation between the caching strategy and response time of the music file, adjusting the caching strategy of the music file; allocating proper computing resources for different users through the number of users and a computing resource allocation optimization strategy; obtaining geographic position data of a user, and optimizing a data transmission path through association analysis of data transmission rate; calculating by adopting the relation between a music file storage structure and a reading speed, and optimizing the storage mode of the music file; according to the efficiency difference of the load balancing strategy of the system under different user numbers, the load balancing strategy is adjusted; determining the optimal proportion of the sound quality to the calculation cost through the balance optimization measurement and calculation of the sound quality and the calculation cost; adopting a prediction model LSTM of user music preference change to match with a data preloading strategy, and optimizing the data preloading strategy; further comprises: and constructing a prediction model of user music preference change by adopting LSTM, and optimizing a data preloading strategy.
The method for constructing the prediction model of the user music preference change by adopting the LSTM comprises the following steps of:
and acquiring music preference model input data of the user according to the historical music preference data and personal information of the user. And determining the super parameters, the number of units, the learning rate and the training round number of the LSTM model. And constructing an LSTM model, sending input data into the LSTM, and updating the model state at the current moment through an internal gating structure. And acquiring input data of the data preloading strategy according to personal information, music characteristics and social network activity attributes of the user. It is determined which data needs to be prepared in advance for the user during the preloading phase, and the preloaded data content is decided according to the preference and preference of the user. And determining the data preloading time according to the using habit and the behavior prediction of the user, and loading related data in advance according to the time period or the place. And according to the historical behavior of the user and the result of the prediction model, obtaining the preloaded data quantity, and improving the data preloaded quantity to reduce the number of times of waiting loading of the user. Taking the output of the data preloading strategy as the input of the LSTM model, training the model, and updating the model parameters through a back propagation algorithm until the training converges. And obtaining a decision result of loading the related music data in advance according to the personal preference and behavior habit of the user by using the trained LSTM model.
Preferably, the adopting a deep learning algorithm dynamically adjusts public cloud and private cloud allocation recommendation strategies to obtain an optimal data preloading scheme, including:
dynamically adjusting the allocation of public cloud resources according to the available resource capacity and the load balancing strategy of the public cloud; according to the matching degree of the service demand and the computing capacity of the private cloud, evaluating the resource allocation strategy of the private cloud; judging the allocation strategy of data transmission through the data transmission bandwidth and capacity evaluation of public cloud and private cloud; evaluating the execution effect of the preloading scheme through the efficiency and stability indexes of the data preloading scheme; performing strategy analysis according to a deep learning algorithm; the deep learning algorithm selects a ResNet network structure and reads real-time allocation conditions of public cloud resources and private cloud resources; analyzing real-time bandwidths and response time of public cloud and private cloud, judging the current cloud resource performance state, and providing a performance evaluation reference for the model; according to the evaluation standard, acquiring a historical performance index of the data preloading scheme, and further judging a load balancing strategy of the cloud resource and a change trend of the load balancing strategy; based on the ResNet network structure, extracting features of different data types and data sizes, and ensuring the efficiency of a data preloading scheme; analyzing access frequency and expected change of data in public cloud and private cloud, and providing targeted training data for a deep learning model; adopting an Adagrad algorithm to perform optimization training on the ResNet network, and adjusting weights to obtain a more accurate resource allocation strategy; comparing training parameters and an initialization state of the deep learning algorithm model, and monitoring an overfitting or underfilling state of the model; when the over fitting or under fitting of the model is detected, the learning rate of an Adagrad algorithm is adjusted, and the stability and the accuracy of the model are ensured; continuously collecting resource allocation data of public cloud and private cloud, and performing real-time training to ensure that the model always maintains an optimal state; further comprises: and acquiring real-time allocation conditions of public cloud and private cloud resources through a ResNet network structure.
The method for acquiring the real-time allocation status of public cloud and private cloud resources through the ResNet network structure specifically comprises the following steps:
and according to the real-time resource allocation, obtaining the use of each resource in the public cloud and the private cloud, wherein the use comprises CPU utilization rate, memory utilization rate, network bandwidth utilization rate, storage space utilization rate, virtual machine number and GPU utilization rate. And (5) utilizing the ResNet network structure to analyze the resource utilization rate. And training by inputting the resource utilization rate data and utilizing a ResNet network model to obtain a predicted output result of the model, and reflecting the change trend and the abnormality of the resource utilization rate. And judging whether the resource utilization rate is normal or not according to the prediction result of the ResNet network model. And judging whether the resource utilization rate exceeds or falls below a normal range and whether an abnormality exists according to the set threshold. If the CPU utilization rate exceeds the threshold value, the CPU is overloaded, and if the GPU utilization rate is lower than the threshold value, the resource waste exists. And generating early warning information of the resource utilization rate according to the output result of the abnormality detection and early warning module. When the resource utilization rate is abnormal, the system sends out early warning in time. And carrying out resource planning and scheduling according to the real-time resource allocation condition and the resource utilization analysis result. Based on the predicted result of the ResNet network model, a resource allocation policy is determined, including task allocation and priority of resource allocation. If the resources are overloaded, more CPU and memory resources are allocated preferentially, and if the resources are wasted, the resource allocation is reduced or the task allocation is adjusted. And dynamically adjusting the resources according to the resource planning and scheduling results. And judging whether the allocation amount of the resources needs to be increased or decreased according to the real-time resource allocation condition.
The invention provides a cloud processing system in a mixed mode, which comprises:
the personalized demand acquisition module is used for acquiring personalized demands of users on privacy protection, tone quality and response time according to user operation habits;
the music data processing module is used for determining a data encryption, identity verification and voice enhancement method required by a user according to the user requirement and judging the network requirement, storage requirement and calculation requirement of cloud calculation required by the music data;
the music preference prediction and preloading module is used for predicting the music preference of a user according to the user interaction data, preloading the music content which can be played by the user in advance through mixed cloud computing, and reducing the buffer time;
the cost optimization and data distribution module is used for generating calculation cost and flow cost meeting the requirements according to the mixed cloud calculation requirements of the music data, and adjusting a data transmission distribution mode of the calculation service between public cloud and private cloud;
the dynamic adjustment and evaluation module is used for evaluating the fluctuation of tone quality, response time and calculation cost again if the number of users or the data to be preloaded are changed due to the change of music preference;
And the cloud resource optimization module is used for dynamically adjusting public cloud and private cloud allocation recommendation strategies by adopting a deep learning algorithm to obtain an optimal data preloading scheme.
The technical scheme provided by the embodiment of the invention can have the following beneficial effects:
the invention discloses a technology for acquiring personalized demands of a user according to operation habits of the user. The technique determines the user's need for data encryption, authentication, and voice enhancement by analyzing the user's privacy protection, voice quality, and response time preferences. Meanwhile, the music preference of the user is predicted according to the user interaction data, and the music content which can be played by the user is preloaded in advance through the hybrid cloud computing, so that the buffer time is reduced. Aiming at cloud computing requirements of music data, the invention also determines network requirements, storage requirements and computing requirements. By generating the computation cost and the traffic cost meeting the requirements, the data transmission distribution mode of the computation service between the public cloud and the private cloud is adjusted. When the number of users or the music preference changes, the invention can reevaluate the changes of tone quality, response time and calculation cost, and adopts a deep learning algorithm to dynamically adjust the data preloading recommendation strategy of public cloud and private cloud so as to obtain the optimal data preloading scheme. In summary, the invention can provide personalized music playing service according to the user demand by integrating privacy protection, tone enhancement, cloud computing, deep learning and other technologies, and improve the user experience.
Drawings
Fig. 1 is a flowchart of a method and a system for cloud processing in a mixed mode according to the present invention.
Fig. 2 is a schematic diagram of a method and a system for cloud processing in a hybrid mode according to the present invention.
Fig. 3 is a schematic diagram of a method and system for hybrid mode cloud processing according to the present invention.
Fig. 4 is a block diagram of a hybrid mode cloud processing system according to the present invention.
Detailed Description
For a further understanding of the present invention, the present invention will be described in detail with reference to the drawings and examples. The present application is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be noted that, for convenience of description, only the portions related to the invention are shown in the drawings.
The method and system for cloud processing in a hybrid mode in this embodiment may specifically include:
s101, according to user operation habits, personalized requirements of users on privacy protection, tone quality and response time are obtained.
And acquiring historical operation data of the user privacy protection habit and acquiring user habit data. Judging the proportion of non-default options in the privacy setting of the user according to the output of the habit data of the user, and if the proportion is larger than a preset threshold value, adjusting the default setting of the system. And analyzing the operation data of the user, identifying the specific frequency response preference under the voice quality demand, and dividing the user into different voice quality demand groups through a K-means clustering algorithm. And according to the K-means algorithm result, if a certain group selects high-tone-quality playing frequency to be higher than a preset proportion, the default tone-quality output of the system is improved for the group. And continuously using the K-means result, and carrying out correlation analysis of tone quality and network bandwidth on each group. Based on the correlation analysis, the response time tolerance of users of each group under different network conditions is predicted. And acquiring the predicted response time tolerance, matching with the operation speed of the user, and optimizing a back-end data processing strategy. Data of system response time and user satisfaction are collected, and correlation is analyzed by using CNN. And carrying out detail optimization on the system response time according to the output result of the CNN. And according to the optimization result of each step, comprehensively distributing weights of tone quality and response time, and analyzing comprehensive user satisfaction as a parameter of system optimization. For example, assume that there is a user habit data set containing privacy setting operation records of 100 users. The privacy settings for each user include 10 options, with 3 non-default options. The proportion of non-default options can be calculated, i.e., (3/10) ×100% =30%. If the preset threshold is 20%, the system default setting may be adjusted because the ratio is greater than the threshold. Next, the user is subjected to a tone quality requirement division using a K-means clustering algorithm. Suppose a user is divided into two groups, group 1 and group 2. In the K-means results, it was found that group 1 favors sound quality with specific frequency response between 200Hz and 500Hz, while group 2 favors sound quality with specific frequency response between 1kHz and 3 kHz. In the analysis of the correlation of the tone quality with the network bandwidth, it was found that the correlation of the tone quality requirements of the users of group 1 with the network bandwidth was lower, whereas the correlation of the tone quality requirements of the users of group 2 with the network bandwidth was higher. Thus, it can be predicted that group 1 users have higher response time tolerance under different network conditions, while group 2 users have lower response time tolerance when the network bandwidth is lower. By matching the user's operating speed with the predicted response time tolerance, the back-end data processing strategy can be optimized. For example, if group 1 users operate at a slower speed, but the predicted response time is more tolerant, a slower data processing strategy may be employed to improve the tonal output of the system. Next, data of system response time and user satisfaction are collected and correlation between them is analyzed using CNN. It is assumed that a strong positive correlation exists between the response time of the system and the user satisfaction, i.e. the faster the response time, the higher the user satisfaction. Based on this result, the response time of the system can be optimized in detail to improve user satisfaction. Finally, according to the optimization result of each step, the weights of tone quality and response time can be comprehensively distributed, and the comprehensive user satisfaction is analyzed as a parameter of system optimization. For example, the weights of the sound quality and the response time may be set to 70% and 30%, respectively, and the performance and the room for improvement of the system may be evaluated according to the user satisfaction.
S102, determining data encryption, authentication and voice quality enhancement methods required by a user according to the user requirements, and judging network requirements, storage requirements and calculation requirements of cloud calculation required by music data.
In particular, the secure hashalgorithm method may be employed provided that the integrity of the data interaction is to be verified. Assuming that two data blocks A and B are provided, hash operations of SHA-256 are respectively carried out on the two data blocks to obtain hash values Ha and Hb. If Ha and Hb are equal, it can be determined that the data has not been tampered with during transmission, thereby ensuring the integrity of the data. For example, if the SHA-256 hash value of data block a is 0x 1234567880 abcdef and the SHA-256 hash value of data block B is also 0x 1234567880 abcdef, then the integrity of the data can be confirmed. For advanced algorithm policies of user privacy protection, if a differential privacy algorithm is adopted to protect the privacy of the user. A privacy parameter epsilon can be set, assuming epsilon=5. If the personal information of a certain user is inquired, the noise range of the inquired result is within +/-5, so that the privacy security of the inquired result can be ensured. A user's age is 30 years and if the user's age is queried, the returned query result may be any number within 30±5, such as 32 or 28. In the music flow control strategy, if an optimized transmission strategy is adopted, the transmission rate of music data is adjusted according to the data protection result. If the current network congestion condition is found according to the flow analysis, the transmission rate of the music data can be reduced, and the network congestion is avoided. Assuming that the current network bandwidth is 10Mbps, if the current network congestion is judged, the transmission rate of music data can be reduced to 5Mbps, thereby ensuring the optimal transmission of traffic. The transmission rate of the music data is originally 10Mbps, but the current network congestion is judged according to the flow analysis, and the transmission rate is reduced to 5Mbps. For the algorithm complexity of real-time audio processing, if the algorithm complexity is higher than a preset algorithm complexity threshold, an additional audio processing calculation unit may be added to increase the processing speed. Assuming that there is originally only one audio processing computing unit, the processing speed is 1000 audio samples/second. If the algorithm complexity is high, an additional audio processing calculation unit can be added, so that the total processing speed is increased to 2000 audio samples/second. The processing speed of the original audio processing computing unit is 1000 audio samples/second, but due to the high algorithm complexity, an additional audio processing computing unit is added, so that the total processing speed reaches 2000 audio samples/second. And evaluating the requirements of the cloud computing resources according to the user authentication result, and if the user authentication is passed, indicating that the user can access the cloud computing resources and perform related computing tasks. If there are a large number of users that need to access cloud computing resources, more computing resources may need to be dynamically allocated to meet the needs of the users. Assuming 1000 users need to access cloud computing resources at a certain time, and the computing resources needed by each user are 1 CPU and 2GB of memory, at least 1000 CPUs and 2000GB of memory need to be dynamically allocated to meet the needs of the users. If the integrity of the data interaction is to be verified, the SecureHashAlgorithm method may be employed. Assuming that two data blocks A and B are provided, hash operations of SHA-256 are respectively carried out on the two data blocks to obtain hash values Ha and Hb. If Ha and Hb are equal, it can be determined that the data has not been tampered with during transmission, thereby ensuring the integrity of the data. For example, if the SHA-256 hash value of data block a is 0x 1234567880 abcdef and the SHA-256 hash value of data block B is also 0x 1234567880 abcdef, then the integrity of the data can be confirmed. For advanced algorithm policies of user privacy protection, if a differential privacy algorithm is adopted to protect the privacy of the user. A privacy parameter epsilon can be set, assuming epsilon=5. If the personal information of a certain user is inquired, the noise range of the inquired result is within +/-5, so that the privacy security of the inquired result can be ensured. A user's age is 30 years and if the user's age is queried, the returned query result may be any number within 30±5, such as 32 or 28. In the music flow control strategy, if an optimized transmission strategy is adopted, the transmission rate of music data is adjusted according to the data protection result. If the current network congestion condition is found according to the flow analysis, the transmission rate of the music data can be reduced, and the network congestion is avoided. Assuming that the current network bandwidth is 10Mbps, if the current network congestion is judged, the transmission rate of music data can be reduced to 5Mbps, thereby ensuring the optimal transmission of traffic. The transmission rate of the music data is originally 10Mbps, but the current network congestion is judged according to the flow analysis, and the transmission rate is reduced to 5Mbps. For the algorithm complexity of real-time audio processing, if the algorithm complexity is higher than a preset algorithm complexity threshold, an additional audio processing calculation unit may be added to increase the processing speed. Assuming that there is originally only one audio processing computing unit, the processing speed is 1000 audio samples/second. If the algorithm complexity is high, an additional audio processing calculation unit can be added, so that the total processing speed is increased to 2000 audio samples/second. The processing speed of the original audio processing computing unit is 1000 audio samples/second, but due to the high algorithm complexity, an additional audio processing computing unit is added, so that the total processing speed reaches 2000 audio samples/second. And evaluating the requirements of the cloud computing resources according to the user authentication result, and if the user authentication is passed, indicating that the user can access the cloud computing resources and perform related computing tasks. If there are a large number of users that need to access cloud computing resources, more computing resources may need to be dynamically allocated to meet the needs of the users. Assuming 1000 users need to access cloud computing resources at a certain time, and the computing resources needed by each user are 1 CPU and 2GB of memory, at least 1000 CPUs and 2000GB of memory need to be dynamically allocated to meet the needs of the users.
S103, predicting the music preference of the user according to the user interaction data, preloading the music content which can be played by the user in advance through mixed cloud computing, and reducing the buffer time.
Specifically, if the user's most recent music listening activity occurs 10 minutes ago, the user typically listens to songs 2 times per hour, based on historical activity patterns. For example, based on metadata and audio characteristics analysis of music, the type of music that a user likes may be popular music, embodied as fast-paced, clear melodies and lyrics to the lyrics of the lyrics. If the interaction frequency of the user in the working scene is high, the user is most active in listening to songs at 10 am and 3 pm every day. According to this mode, the predicted time for the user to listen to the song next may be 10 am on tomorrow. The network flow mode and the real-time network environment state of music playing are monitored through the Wireshark and the network state APIs, and the network environment between 7 and 9 points at night is found to be poor for a user, so that the music content is decided to be stored on the node closest to the user, and better playing experience is provided. If the user's play behavior shows a seasonal pattern in which Christmas songs are listened to every year on Christmas, the music content associated with Christmas, such as Christmas song albums, is preloaded. By analyzing the popularity trend of music content and social media popularity, the most popular music content is found to be a popular song named "BlindingLights". If the user repeatedly plays a song 10 times in the past month with a full play rate of 90%, it may be determined that the user likes the song and adds it to the list of music that the user likes. Music that is comfortable to listen to when the user's emotion is found to fall is found by emotion analysis, and such music is preloaded to provide an emotion regulating effect. Music data from platforms Spotify, appleMusic and youtube music, etc. are integrated to provide a wider variety of music selections to the user. If the resource scheduling strategy according to the hybrid cloud computing is adopted, 100 computing servers are provided for the computing model and are used for processing the updating and analysis of the user behavior prediction model. The format and quality of the preloaded music are determined according to the type of equipment used by the user, and smaller audio files are used for the mobile phone user to save storage space and data flow. And loading the music in the favorite track list of the user on the user equipment in advance according to the data caching strategy of the pre-loading of the music content so as to ensure that the user can use the music at any time. And constructing a user behavior prediction model by using a decision tree algorithm, if a real-time updating mechanism judges that the user can change the music preference, analyzing new user interaction data and updating a track recommendation list of the user. And judging whether the preloaded music can be smoothly played by a user according to an usability evaluation mechanism of the preloaded music content so as to ensure the continuity of user experience. And if the real-time performance and timeliness of the user interaction data obtain a new interaction record, immediately updating the user behavior prediction model to continuously optimize the music recommendation list of the user. If the user's most recent music listening activity occurs 10 minutes ago, the user typically listens to the song 2 times per hour, according to the historical activity pattern. For example, based on metadata and audio characteristics analysis of music, the type of music that a user likes may be popular music, embodied as fast-paced, clear melodies and lyrics to the lyrics of the lyrics. If the interaction frequency of the user in the working scene is high, the user is most active in listening to songs at 10 am and 3 pm every day. According to this mode, the predicted time for the user to listen to the song next may be 10 am on tomorrow. The network flow mode and the real-time network environment state of music playing are monitored through the Wireshark and the network state APIs, and the network environment between 7 and 9 points at night is found to be poor for a user, so that the music content is decided to be stored on the node closest to the user, and better playing experience is provided. If the user's play behavior shows a seasonal pattern in which Christmas songs are listened to every year on Christmas, the music content associated with Christmas, such as Christmas song albums, is preloaded. By analyzing the popularity trend of music content and social media popularity, the most popular music content is found to be a popular song named "BlindingLights". If the user repeatedly plays a song 10 times in the past month with a full play rate of 90%, it may be determined that the user likes the song and adds it to the list of music that the user likes. Music that is comfortable to listen to when the user's emotion is found to fall is found by emotion analysis, and such music is preloaded to provide an emotion regulating effect. Music data from platforms Spotify, appleMusic and youtube music, etc. are integrated to provide a wider variety of music selections to the user. If the resource scheduling strategy according to the hybrid cloud computing is adopted, 100 computing servers are provided for the computing model and are used for processing the updating and analysis of the user behavior prediction model. The format and quality of the preloaded music are determined according to the type of equipment used by the user, and smaller audio files are used for the mobile phone user to save storage space and data flow. And loading the music in the favorite track list of the user on the user equipment in advance according to the data caching strategy of the pre-loading of the music content so as to ensure that the user can use the music at any time. And constructing a user behavior prediction model by using a decision tree algorithm, if a real-time updating mechanism judges that the user can change the music preference, analyzing new user interaction data and updating a track recommendation list of the user. And judging whether the preloaded music can be smoothly played by a user according to an usability evaluation mechanism of the preloaded music content so as to ensure the continuity of user experience. And if the real-time performance and timeliness of the user interaction data obtain a new interaction record, immediately updating the user behavior prediction model to continuously optimize the music recommendation list of the user.
And providing computing resources for the computing model through a resource scheduling strategy of the hybrid cloud computing.
Specifically, the selection of private cloud and public cloud resources may use OpenStack as the private cloud platform and AWS or Azure as the public cloud platform. Monitoring and analyzing load conditions and performance metrics the resource requirements and optimization objectives can be obtained by collecting and analyzing metric data using promethaus as a monitoring tool. The load balancing algorithm may use a round robin algorithm or a LeastConnection algorithm to determine which computing resources a computing task should be allocated to achieve load balancing and maximize resource utilization. The elasticity adjustment rules and policies may use AutoScaling functions to determine whether an automatic expansion or contraction of the scale of computing resources is required to accommodate the load changes of the computing model. The data storage position can be determined by object storage technologies such as Minio or Ceph, and the data access efficiency is improved by technologies such as data copying or caching. The resource scheduling strategy can use Kubernetes as a container scheduling tool to optimize resource scheduling according to the load condition and performance index of the computing resources so as to improve the resource utilization rate and the performance of the computing model. Encrypted transmission and storage of data may use TLS/SSL to ensure security of data and privacy. Real-time monitoring and tuning can use Grafana as a monitoring and visualization tool by which the availability of computing resources is determined to ensure that the computing model continues to run and provide stable performance. The resource allocation scheme may be determined from the requirements of the computational model and the results of the resource scheduling policy, using Terraform as an infrastructure automation tool to implement the resource scheduling. Periodic evaluation and optimization resource scheduling policies can be evaluated and optimized using monitoring data of Prometheus and Grafana to accommodate changes in computing models and hybrid cloud environments. The selection of private cloud and public cloud resources may use OpenStack as the private cloud platform and AWS or Azure as the public cloud platform. Monitoring and analyzing load conditions and performance metrics the resource requirements and optimization objectives can be obtained by collecting and analyzing metric data using promethaus as a monitoring tool. The load balancing algorithm may use a round robin algorithm or a LeastConnection algorithm to determine which computing resources a computing task should be allocated to achieve load balancing and maximize resource utilization. The elasticity adjustment rules and policies may use AutoScaling functions to determine whether an automatic expansion or contraction of the scale of computing resources is required to accommodate the load changes of the computing model. The data storage position can be determined by object storage technologies such as Minio or Ceph, and the data access efficiency is improved by technologies such as data copying or caching. The resource scheduling strategy can use Kubernetes as a container scheduling tool to optimize resource scheduling according to the load condition and performance index of the computing resources so as to improve the resource utilization rate and the performance of the computing model. Encrypted transmission and storage of data may use TLS/SSL to ensure security of data and privacy. Real-time monitoring and tuning can use Grafana as a monitoring and visualization tool by which the availability of computing resources is determined to ensure that the computing model continues to run and provide stable performance. The resource allocation scheme may be determined from the requirements of the computational model and the results of the resource scheduling policy, using Terraform as an infrastructure automation tool to implement the resource scheduling. Periodic evaluation and optimization resource scheduling policies can be evaluated and optimized using monitoring data of Prometheus and Grafana to accommodate changes in computing models and hybrid cloud environments.
And constructing a user behavior prediction model by using a decision tree algorithm, and updating a track recommendation list of the user if the real-time updating mechanism judges that the user can change the music preference.
And acquiring user interaction data and preprocessing, including data cleaning, missing value processing and feature selection operation, according to the user basic information, the user historical behavior data, the current interaction data, the music attribute and the context information. Extracting valuable features from the preprocessed user interaction data by adopting feature extraction, and playing singer and type information in the record recently played by the user. And obtaining a result of feature extraction, taking the result as input, performing model training through a decision tree algorithm, and establishing a user behavior prediction model. In the training process, according to the labels of the historical behavior data and the current interaction data of the user, whether the user changes the music preference or not carries out supervised learning on the user behavior prediction model. And determining a trained user behavior prediction model, predicting new user interaction data by using the trained user behavior prediction model, and judging whether the user can change the music preference. The prediction result is a probability value indicating the likelihood of the user changing the music preference. And updating the track recommendation list of the user according to the prediction result. If the predicted result shows that the user can replace the music preference, regenerating a song recommendation list according to the new preference of the user so as to improve the satisfaction degree and the user experience of the user. For example, assume that there is a music recommendation system that predicts whether a user will change a music preference based on the user's basic information, historical behavior data, current interaction data, and music attributes. Duplicate items, missing values and outliers are deleted from the user interaction data, ensuring the accuracy and integrity of the data. And selecting singers and type information in the latest play records from the preprocessed user interaction data as features. Singers and types played by the user in the last week are recorded, wherein the singers comprise singer A, singer B, type X and type Y. Using feature extraction techniques, valuable features can be extracted from singer and genre information. The preference degree of the user for different singers and types can be calculated, and the ratio of the playing times of the singer A to the total playing times or the good rate of the user for the type X can be calculated. A user behavior prediction model can be established using a decision tree algorithm. And performing supervised learning on the user behavior prediction model according to the labels of the user historical behavior data and the current interaction data, whether music preference is replaced or not, and the like. The data set may be marked according to whether the user has replaced a favorite singer or genre in the last week. The trained user behavior prediction model is used for predicting new user interaction data and judging whether the user can change music preference or not. The prediction result may be a probability value indicating the likelihood of the user changing the music preference. The user behavior prediction model may predict that the probability of the user changing the music preference is 8. Based on the prediction result, the track recommendation list of the user may be updated. If the predicted result indicates that the user will change the music preference, the song recommendation list can be regenerated according to the new preference of the user so as to improve the satisfaction degree and the user experience of the user. If the prediction result indicates that the user has replaced a favorite singer, the relevant song of the singer may be added to the recommendation list.
S104, according to the mixed cloud computing requirements of the music data, computing cost and flow cost meeting the requirements are generated, and a data transmission distribution mode of computing service between public cloud and private cloud is adjusted.
And acquiring the size and type of the data according to the analysis requirements of the size and the format of the music data. And judging the computing nodes and the storage positions by adopting computing resource quota and distribution of the hybrid cloud. And adjusting the data processing priority by calculating the time delay requirement and the response time of the task. And generating the expected calculation cost according to the calculation efficiency and cost difference of the public cloud and the private cloud. The flow capacity and the rate requirement of data transmission are determined by the storage mode and the access mode of the music data. And optimizing the matching degree of the data redundancy and the backup strategy in the hybrid cloud by adopting a load balancing algorithm, and ensuring the data integrity and accessibility. And acquiring real-time processing and storage requirements of the music data, and determining the duration and format of data storage. And adjusting the data transmission speed through the bandwidth requirement and the peak flow of the data transmission. And determining a time window of data transmission according to the network stability evaluation between the public cloud and the private cloud. And setting a processing strategy when data is in error through error tolerance rate and rollback mechanism of music data processing. And adopting a deep reinforcement learning algorithm, and dynamically adjusting data transmission allocation according to a self-adaptive load balancing strategy of the hybrid cloud so as to optimize cost and efficiency. For example, assume that the size of music data is 1TB and the format is MP3. According to the parsing requirement, the size and type of the data, namely 1TB and MP3 format, need to be acquired. Next, computing resource quota and distribution need to be considered. Assume that there are 3 compute nodes in the hybrid cloud, each node having the same computing power. Depending on the processing requirements of the music data, computing tasks may be assigned to these 3 nodes. For storage locations, it is assumed that there is 500GB of storage space available in the private cloud. Music data may be stored in a private cloud. The priority of data processing can be adjusted according to the latency requirements and response times of the computational tasks. For example, if a certain computing node is more powerful, more important tasks may be assigned to that node. Based on the computational efficiency and cost differences of public and private clouds, an estimated computational cost may be generated. Assuming that the computation cost of the private cloud is 100 yuan/hour, the computation cost of the public cloud is 200 yuan/hour. By means of the storage mode and the access mode, the traffic capacity and the rate requirements of the data transmission can be determined. To optimize data redundancy and backup strategies, a load balancing algorithm may be employed. Music data is stored in both private and public clouds to ensure data integrity and accessibility. The storage duration and format of the data can be determined according to the real-time processing and storage requirements of the music data. If real-time processing of the data is required, the data may alternatively be stored in memory rather than on a hard disk. The data transmission speed can be adjusted according to the bandwidth requirement and peak flow of the data transmission. If the bandwidth is limited, the data transmission speed can be reduced to avoid network congestion. According to the network stability evaluation between the public cloud and the private cloud, a time window for data transmission can be determined. If the network stability is poor, the data transmission can be selected to be carried out in a low peak period. For error tolerance and rollback mechanisms of data processing, corresponding processing strategies can be formulated. If a data error occurs, a rollback operation may be performed using the backup data. Finally, through a deep reinforcement learning algorithm, the data transmission allocation can be dynamically adjusted to optimize cost and efficiency. According to the load balancing strategy, the data transmission tasks can be automatically distributed according to the processing capacities of different computing nodes.
S105, if the number of users or the music preference changes, the data to be preloaded changes, the changes of the sound quality, the response time and the calculation cost are again evaluated.
And judging whether abnormal user growth or abnormal user reduction exists or not according to the user quantity change data and the historical trend thereof. And acquiring the data size and the coding form of various types of music, and judging the influence on the tone quality. And determining the optimal number of concurrent requests through correlation analysis of the number of users and the concurrent amount of the server requests. And adjusting the caching strategy of the music file according to the influence relation between the caching strategy of the music file and the response time. And allocating proper computing resources for different users through the number of users and the computing resource allocation optimization strategy. And obtaining the geographical position data of the user, and optimizing the data transmission path through the association analysis of the data transmission rate. And (3) calculating by adopting the relation between the music file storage structure and the reading speed, and optimizing the storage mode of the music file. And adjusting the load balancing strategy according to the efficiency difference of the load balancing strategy of the system under different user numbers. And determining the optimal proportion of the sound quality to the calculation cost through the balance optimization measurement and calculation of the sound quality and the calculation cost. And adopting a prediction model LSTM of user music preference change to match with the data preloading strategy, and optimizing the data preloading strategy. For example, by means of the user quantity change data and the historical trend thereof, the number of users of the assumed music platform is gradually increased in the past year, and from the newly increased number of users per month, the user growth in the first three months is about 1000 people, and then abnormal user growth starts to occur in the fourth month, so that 2000 people are reached, and the growth speed is doubled compared with that in the first three months. This indicates that there may be a particular event or promotional activity that leads to a proliferation of the number of users. In acquiring the data size and the encoding form of various types of music, it was found that the average size of the music file in the MP3 format was 5MB, and the average size of the music file in the FLAC format was 30MB. This illustrates that the FLAC format may occupy more memory than the MP3 format, but may also provide higher sound quality. By correlation analysis of the number of users and the request concurrency of the server, it is assumed that the request concurrency of the server reaches 500 times/second when 1000 users play music at the same time. Then further assume that after increasing the number of concurrent requests to 2000 times/second, the response time of the server begins to slow down, reaching 3 seconds. This illustrates that at 2000 concurrent requests per second, the performance of the server has reached a bottleneck. According to the relation between the caching strategy and the response time of the music file, the response time of the user when the music file is cached on the CDN node is found to be 5 seconds when the user accesses the music from a remote area, and the response time of the user when the music file is directly obtained from a server is found to be 10 seconds. This illustrates that the response time of the user can be significantly reduced by caching the music file. Through the number of users and the computing resource allocation optimization strategy, it is found that for normal users, each user needs to be allocated 1 computing resource unit, while for advanced users, each user needs to be allocated 2 computing resource units. By the allocation strategy, the computing resources can be reasonably allocated according to the level and the number of the users, so that the overall performance of the system is improved. The geographic position data of the user is obtained, through the association analysis of the data transmission rate, the user A is found to be located in the city X, the user B is found to be located in the city Y, through direct connection transmission, the data transmission rate of the user A is 100Mbps, and the data transmission rate of the user B is only 50Mbps. This shows that city a has a better network environment and a faster data transfer rate, so that the data transfer path can be optimized to transfer music files from city a to user B. By adopting the relation measurement and calculation of the music file storage structure and the reading speed, the reading speed of storing the music file on the SSD hard disk is found to be 100MB/s, and the reading speed of storing the music file on the HDD hard disk is found to be only 50MB/s. This means that the speed of reading music files can be increased by using the SSD hard disk, thereby optimizing the storage manner of music files. According to the efficiency difference of the load balancing strategy of the system under different user numbers, the average response time is found to be 2 seconds when the polling load balancing strategy is adopted under the condition of 10 users, and the average response time is found to be 5 seconds when the weighted polling strategy is adopted. This indicates that the weighted polling strategy works better on load balancing, which can improve the overall performance of the system. By optimizing the measurement by the tradeoff of sound quality and computational cost, it is assumed that increasing sound quality requires increasing computational cost, which would increase by 50% if sound quality were increased to the highest level. By balancing the relationship between the sound quality and the calculation cost, the optimal ratio of the sound quality to the calculation cost can be determined, so that the calculation cost is reduced as much as possible while the user demand is met. Matching with the data preloading policy using a predictive model of user music preference variation, such as LSTM, may use the LSTM model to predict the types of music that the user may like in the future based on the user's historical music preferences. And then, data preloading can be carried out according to the prediction results, and music files which are possibly liked are cached on the equipment of the user in advance, so that the data preloading strategy is optimized, and the user experience is improved.
And constructing a prediction model of user music preference change by adopting LSTM, and optimizing a data preloading strategy.
Specifically, based on the user's historical music preference data and personal information, for example, the number of songs that the user has heard in the past month is 100, of which 60 are popular music, 20 are rock music, and 20 are classical music. The personal information of the user includes sex female, age 25 years, and region city. And determining the super parameters of the LSTM model, setting the unit number of the LSTM model to 128, learning rate to 001 and training round number to 10. And constructing an LSTM model, sending input data into the LSTM, and updating the model state at the current moment through an internal gating structure. It is assumed that the input data includes the user's history music preference data and personal information, wherein the history music preference data may be expressed as a vector, such as [60,20,20], and the personal information may be expressed as a vector, [1,25,1], wherein 1 represents female, 25 represents age, and 1 represents city. And acquiring input data of the data preloading strategy according to personal information, music characteristics and social network activity attributes of the user. The user's musical characteristics may be denoted as [8,2,3], representing the user's likeness to popular, rock and classical music, and social network activity attributes may be denoted as [5,7,6], representing the user's liveness on the social network. It is determined which data needs to be prepared in advance for the user during the preloading phase, and the preloaded data content is decided according to the preference and preference of the user. Based on the user's historical music preference data and personal information, some related data of popular music, rock music and classical music may be preloaded to satisfy the user's preference. And determining the data preloading time according to the using habit and the behavior prediction of the user, and loading related data in advance according to the time period or the place. If a user normally listens to music on the way to work in the morning, relevant music data can be loaded in advance in the morning to reduce the time the user waits for loading. And according to the historical behavior of the user and the result of the prediction model, obtaining the preloaded data quantity, and improving the data preloaded quantity to reduce the number of times of waiting loading of the user. Based on the number of songs heard by the user in the past month and the result of the predictive model, the amount of music data preloaded each time can be determined to meet the needs of the user. Taking the output of the data preloading strategy as the input of the LSTM model, training the model, and updating the model parameters through a back propagation algorithm until the training converges. And taking personal information, music characteristics, social network activity attributes and preloaded data quantity of the user as inputs of the LSTM model, updating model parameters through a back propagation algorithm, and improving the prediction accuracy of the model. And obtaining a decision result of loading the related music data in advance according to the personal preference and behavior habit of the user by using the trained LSTM model. According to personal information, music characteristics and social network activity attributes of a user, the trained LSTM model can be utilized to judge which types of music data are loaded in advance at what time point and place so as to improve user experience. According to personal information and music characteristics of a user, the probability that the user likes to listen to popular music in the middle of working in the morning can be predicted to be 8 by using the trained LSTM model, the probability of rock music is 2, and the probability of classical music is 3. Based on this prediction, some related data of popular music and classical music can be loaded in advance to improve the user experience. The amount of preloaded data may be determined based on the number of songs that the user has heard in the past month and the result of the predictive model, e.g., 30 pieces of music data per preloaded. Based on the user's historical music preference data and personal information, for example, the number of songs that the user has heard in the past month is 100, of which 60 are popular music, 20 are rock music, and 20 are classical music. The personal information of the user includes sex female, age 25 years, and region city. And determining the super parameters of the LSTM model, setting the unit number of the LSTM model to 128, learning rate to 001 and training round number to 10. And constructing an LSTM model, sending input data into the LSTM, and updating the model state at the current moment through an internal gating structure. It is assumed that the input data includes the user's history music preference data and personal information, wherein the history music preference data may be expressed as a vector, such as [60,20,20], and the personal information may be expressed as a vector, [1,25,1], wherein 1 represents female, 25 represents age, and 1 represents city. And acquiring input data of the data preloading strategy according to personal information, music characteristics and social network activity attributes of the user. The user's musical characteristics may be denoted as [8,2,3], representing the user's likeness to popular, rock and classical music, and social network activity attributes may be denoted as [5,7,6], representing the user's liveness on the social network. It is determined which data needs to be prepared in advance for the user during the preloading phase, and the preloaded data content is decided according to the preference and preference of the user. Based on the user's historical music preference data and personal information, some related data of popular music, rock music and classical music may be preloaded to satisfy the user's preference. And determining the data preloading time according to the using habit and the behavior prediction of the user, and loading related data in advance according to the time period or the place. If a user normally listens to music on the way to work in the morning, relevant music data can be loaded in advance in the morning to reduce the time the user waits for loading. And according to the historical behavior of the user and the result of the prediction model, obtaining the preloaded data quantity, and improving the data preloaded quantity to reduce the number of times of waiting loading of the user. Based on the number of songs heard by the user in the past month and the result of the predictive model, the amount of music data preloaded each time can be determined to meet the needs of the user. Taking the output of the data preloading strategy as the input of the LSTM model, training the model, and updating the model parameters through a back propagation algorithm until the training converges. And taking personal information, music characteristics, social network activity attributes and preloaded data quantity of the user as inputs of the LSTM model, updating model parameters through a back propagation algorithm, and improving the prediction accuracy of the model. And obtaining a decision result of loading the related music data in advance according to the personal preference and behavior habit of the user by using the trained LSTM model. According to personal information, music characteristics and social network activity attributes of a user, the trained LSTM model can be utilized to judge which types of music data are loaded in advance at what time point and place so as to improve user experience. According to personal information and music characteristics of a user, the probability that the user likes to listen to popular music in the middle of working in the morning can be predicted to be 8 by using the trained LSTM model, the probability of rock music is 2, and the probability of classical music is 3. Based on this prediction, some related data of popular music and classical music can be loaded in advance to improve the user experience. The amount of preloaded data may be determined based on the number of songs that the user has heard in the past month and the result of the predictive model, e.g., 30 pieces of music data per preloaded.
And S106, dynamically adjusting public cloud and private cloud allocation recommendation strategies by adopting a deep learning algorithm to obtain an optimal data preloading scheme.
And dynamically adjusting the allocation of public cloud resources according to the available resource capacity and the load balancing strategy of the public cloud. And evaluating a resource allocation strategy of the private cloud according to the matching degree of the service demand and the computing capacity of the private cloud. And judging the distribution strategy of data transmission through the data transmission bandwidth and capacity evaluation of the public cloud and the private cloud. And evaluating the execution effect of the preloading scheme through the efficiency and stability indexes of the data preloading scheme. Performing strategy analysis according to a deep learning algorithm; and the deep learning algorithm selects a ResNet network structure and reads the real-time allocation condition of public cloud resources and private cloud resources. And analyzing the real-time bandwidth and response time of the public cloud and the private cloud, judging the current cloud resource performance state, and providing a performance evaluation reference for the model. And acquiring historical performance indexes of the data preloading scheme according to the evaluation standard, and further judging a load balancing strategy of the cloud resource and a change trend of the load balancing strategy. And based on the ResNet network structure, extracting the characteristics of different data types and data sizes, and ensuring the efficiency of a data preloading scheme. And analyzing the access frequency and expected change of the data in the public cloud and the private cloud, and providing targeted training data for the deep learning model. And (3) carrying out optimization training on the ResNet network by adopting an Adagrad algorithm, and adjusting weights to obtain a more accurate resource allocation strategy. Comparing the training parameters of the deep learning algorithm model with the initialized state, and monitoring the over-fitting or under-fitting state of the model. When the over fitting or under fitting of the model is detected, the learning rate of the Adagrad algorithm is adjusted, and the stability and the accuracy of the model are ensured. Continuously collecting resource allocation data of public cloud and private cloud, and performing real-time training to enable the model to always keep an optimal state, for example, according to available resource capacity and load balancing strategy of the public cloud, assuming that 100 virtual machines exist in the public cloud, wherein the resource capacity of each virtual machine is 8 CPU cores and 16GB of memory. And according to the load balancing strategy, each virtual machine evenly distributes resources. It is assumed that the business needs of private clouds require running a large database and at least 20 CPU cores and 64GB memory to meet the requirements. According to the matching degree of the computing power, the private cloud can allocate 25 virtual machines, and each virtual machine resource is 8 CPU cores and 16GB of memory. For the allocation policy of data transmission, it is assumed that the data transmission bandwidth between the public cloud and the private cloud is 1000Mbps and the capacity is 1TB. And according to the requirements, distributing the data transmission bandwidth according to the use proportion of public cloud and private cloud resources, namely, the public cloud accounts for 70% and the private cloud accounts for 30%. The data transmission bandwidth of the public cloud is 700Mbps, and the private cloud is 300Mbps; the data transmission capacity of the public cloud is 700GB, and the private cloud is 300GB. For evaluation of the data preloading scheme, it is assumed that the preloading scheme requires 500GB of data to be loaded into public and private clouds within 30 minutes. If the data loading rate per minute is 50GB, the execution effect of the preloading scheme can meet the requirements. In the deep learning algorithm, a ResNet network structure is selected for policy analysis. The real-time allocation conditions of public cloud resources and private cloud resources are assumed to be read, wherein the allocation conditions of the public cloud resources are as follows: the CPU core utilization rate is 80%, and the memory utilization rate is 60%; the distribution condition of the private cloud resources is as follows: the CPU core utilization rate is 50%, and the memory utilization rate is 70%. Based on these data, the current cloud resource performance state can be determined and a performance assessment benchmark can be provided for the model. And acquiring historical performance indexes of the data preloading scheme according to the evaluation standard, and further judging a load balancing strategy of the cloud resource and a change trend of the load balancing strategy. Assuming that the load balancing policy of public cloud remains stable within the past week, the resource allocation situation is basically unchanged, while the load balancing policy of private cloud fluctuates, and the resource allocation situation is sometimes increased and sometimes reduced. And based on the ResNet network structure, extracting the characteristics of different data types and data sizes, and ensuring the efficiency of a data preloading scheme. It is assumed that in the feature extraction process, data can be divided into small data and large data according to the size and type of the data and processed separately. And analyzing the access frequency and expected change of the data in the public cloud and the private cloud, and providing more targeted training data for the deep learning model. It is assumed that the frequency of data access in public clouds is higher, while the frequency of data access in private clouds is lower, but it is expected that the frequency of data access in future private clouds will increase. And (3) carrying out optimization training on the ResNet network by adopting an Adagrad algorithm, and adjusting weights to obtain a more accurate resource allocation strategy. It is assumed that the manner in which the weights are adjusted during the training process is determined based on the accuracy of the resource allocation strategy, i.e., the higher the accuracy, the less the weight adjustment. Comparing the training parameters of the deep learning algorithm model with the initialized state, and monitoring the over-fitting or under-fitting state of the model. Assuming during the monitoring process, the model was found to have an over-fit, i.e., perform well on the training set but not on the test set. In order to solve the over-fitting problem, the learning rate of the Adagrad algorithm can be adjusted so as to improve the stability and accuracy of the model. And continuously collecting resource allocation data of public cloud and private cloud, and performing real-time training to ensure that the model always maintains the optimal state. It is assumed that public and private cloud resource allocation data is collected daily and used for real-time training to maintain model accuracy and adaptability.
And acquiring real-time allocation conditions of public cloud and private cloud resources through a ResNet network structure.
And according to the real-time resource allocation, obtaining the use of each resource in the public cloud and the private cloud, wherein the use comprises CPU utilization rate, memory utilization rate, network bandwidth utilization rate, storage space utilization rate, virtual machine number and GPU utilization rate. And (5) utilizing the ResNet network structure to analyze the resource utilization rate. And training by inputting the resource utilization rate data and utilizing a ResNet network model to obtain a predicted output result of the model, and reflecting the change trend and the abnormality of the resource utilization rate. And judging whether the resource utilization rate is normal or not according to the prediction result of the ResNet network model. And judging whether the resource utilization rate exceeds or falls below a normal range and whether an abnormality exists according to the set threshold. If the CPU utilization rate exceeds the threshold value, the CPU is overloaded, and if the GPU utilization rate is lower than the threshold value, the resource waste exists. And generating early warning information of the resource utilization rate according to the output result of the abnormality detection and early warning module. When the resource utilization rate is abnormal, the system sends out early warning in time. And carrying out resource planning and scheduling according to the real-time resource allocation condition and the resource utilization analysis result. Based on the predicted result of the ResNet network model, a resource allocation policy is determined, including task allocation and priority of resource allocation. If the resources are overloaded, more CPU and memory resources are allocated preferentially, and if the resources are wasted, the resource allocation is reduced or the task allocation is adjusted. And dynamically adjusting the resources according to the resource planning and scheduling results. And judging whether the allocation amount of the resources needs to be increased or decreased according to the real-time resource allocation condition. For example, assume that there is one public cloud and one private cloud, each with 10 virtual machines in it. The resource utilization of these virtual machines is intended to be monitored, including CPU utilization, memory utilization, network bandwidth utilization, storage space utilization, and GPU utilization. Resource utilization data of each virtual machine in a day is collected as training data. These data include percent resource utilization per hour. Next, these data are trained using the res net network model. After training is completed, the model can be used to predict the trend of future resource utilization and abnormal conditions. The hypothetical model predicts that the CPU utilization of a certain virtual machine will exceed the 90% threshold at the next hour. According to the prediction result, the condition that the virtual machine is likely to be overloaded with resources can be judged. In addition, the hypothetical model predicts that the GPU utilization of a certain virtual machine will be below a 10% threshold in the next hour. According to the prediction result, the condition that the virtual machine has resource waste can be judged. And generating early warning information of the resource utilization rate according to the output result of the abnormality detection and early warning module. If the resource utilization rate is found to be abnormal, the system can send out early warning in time. And according to the real-time resource allocation status and the resource utilization analysis result, resource planning and scheduling can be performed. According to the prediction result of the model, a resource allocation strategy can be determined, such as the situation that more CPU and memory resources are preferentially allocated to handle resource overload, or the situation that resource allocation is reduced or task allocation is adjusted to reduce resource waste. Finally, the resources can be dynamically adjusted according to the results of resource planning and scheduling. If a real-time resource allocation situation is found to require an increase or decrease in the allocation of resources, adjustments can be made accordingly.
The foregoing description is only of the preferred embodiments of the present application and is presented as a description of the principles of the technology being utilized. It will be appreciated by persons skilled in the art that the scope of the invention referred to in this application is not limited to the specific combinations of features described above, but also covers other embodiments which may be formed by any combination of features described above or their equivalents without departing from the spirit of the application. Such as the above-described features and technical features having similar functions (but not limited to) disclosed in the present application are replaced with each other.
Claims (6)
1. A method of mixed mode cloud processing, the method comprising:
according to the operation habit of the user, the personalized requirements of the user on privacy protection, tone quality and response time are obtained;
according to the user requirements, determining data encryption, identity verification and voice quality enhancement methods required by the user, and judging network requirements, storage requirements and calculation requirements of cloud calculation required by music data;
according to the user interaction data, predicting the music preference of the user, and preloading the music content which can be played by the user in advance through mixed cloud computing;
according to the mixed cloud computing requirements of music data, computing cost and flow cost meeting the requirements are generated, and a data transmission distribution mode of computing service between public cloud and private cloud is adjusted;
If the number of users or the music preference changes to cause the data to be preloaded to change, the change of tone quality, response time and calculation cost is estimated again;
adopting a deep learning algorithm to dynamically adjust public cloud and private cloud allocation recommendation strategies to obtain an optimal data preloading scheme;
the method for acquiring the personalized requirements of the user on privacy protection, tone quality and response time according to the user operation habit comprises the following steps:
acquiring historical operation data of user privacy protection habits, acquiring user habit data, judging the proportion of non-default options in user privacy settings according to the output of the user habit data, and if the proportion is greater than a preset threshold value, adjusting the default settings of the system;
analyzing operation data of a user, identifying specific frequency response preference under sound quality requirements, and dividing the user into different sound quality requirement groups through a K-means clustering algorithm;
continuously using the K-means result, carrying out correlation analysis on tone quality and network bandwidth on each group, and predicting response time tolerance of users of each group under different network conditions according to the correlation analysis;
acquiring predicted response time tolerance, matching with the operation speed of a user, optimizing a back-end data processing strategy, acquiring data of system response time and user satisfaction, analyzing the correlation by using CNN, and carrying out detail optimization on the system response time according to the output result of the CNN;
According to the optimization result of each step, comprehensively distributing the weights of tone quality and response time, and analyzing comprehensive user satisfaction as a parameter of system optimization;
according to the user interaction data, predicting the music preference of the user, pre-loading the music content which can be played by the user in advance through mixed cloud computing, wherein the method comprises the following steps:
according to the interactive data and the song listening behavior of the user, determining the music type preference of the user, and predicting the next song listening time of the user; determining an optimal storage position of music content by acquiring a network flow mode and a real-time network environment state of music playing; determining a favorite music list of the user according to the repeated play times and the complete play rate in the play history of the user;
determining the format and quality of the preloaded music according to the type of the equipment used by the user;
according to a data caching strategy of the pre-loading of the music content, loading the music in a favorite music list of a user to user equipment in advance;
acquiring user interaction data according to user basic information, user history behavior data, current interaction data, music attribute and context information, preprocessing, including data cleaning, missing value processing and feature selection operation, extracting singer and type information in a user play record according to the preprocessed user interaction data by adopting feature extraction, acquiring a result of the feature extraction as input, performing model training through a decision tree algorithm, establishing a user behavior prediction model, performing supervised learning on the user behavior prediction model according to the user history behavior data and the tags of the current interaction data, determining a trained user behavior prediction model, predicting new user interaction data by utilizing the trained user behavior prediction model, judging whether the user can replace the music preference, wherein the prediction result is a probability value, representing the possibility of replacing the music preference by the user, updating a song recommendation list of the user according to the prediction result, and regenerating the song recommendation list according to the new preference of the user if the prediction result indicates that the user can replace the music preference;
Judging whether the preloaded music can be smoothly played by a user according to an usability evaluation mechanism of the preloaded music content;
and if the real-time performance and timeliness of the user interaction data obtain a new interaction record, immediately updating the user behavior prediction model to continuously optimize the music recommendation list of the user.
2. The method for cloud processing in hybrid mode according to claim 1, wherein determining the data encryption, authentication and voice enhancement methods required by the user according to the user requirement, determining the network requirement, storage requirement and calculation requirement of cloud calculation required by the music data, comprises:
by adopting a Secure Hash Algorithm method, the integrity of the data is ensured; judging whether an additional audio processing calculation unit is needed to be added according to whether the algorithm complexity of the real-time audio processing is higher than a preset algorithm complexity threshold; and according to the user identity verification result, evaluating the required cloud computing resources and determining whether more computing resources need to be dynamically allocated.
3. The method for processing the cloud in the mixed mode according to claim 1, wherein the generating the computing cost and the traffic cost to meet the requirements according to the mixed cloud computing requirements of the music data, and adjusting the data transmission allocation manner of the computing service between the public cloud and the private cloud, comprises:
The method comprises the steps of obtaining a mixed cloud computing node and a storage position by obtaining the size and the type of music data, further determining the computing cost by adjusting the data processing priority, optimizing the data redundancy and backup strategy of the mixed cloud according to the evaluation data transmission requirement, determining the data storage time and format, determining the data transmission time window by adjusting the data transmission speed, and dynamically adjusting the data transmission allocation by adopting a deep reinforcement learning algorithm according to the formulated error processing strategy.
4. The method for processing the cloud in the mixed mode according to claim 1, wherein if the number of users or the music preference changes to change the data to be preloaded, the changes of the sound quality, the response time and the calculation cost are evaluated again, comprising:
the method comprises the steps of obtaining the size and the code of music data through judging user changes, determining the optimal concurrency request quantity, further adjusting a music file cache strategy, distributing computing resources, optimizing a data transmission path, adjusting a load balancing strategy through optimizing a music file storage mode, determining the proportion of optimal tone quality to computing cost, and optimizing a data preloading strategy by adopting a prediction model.
5. The method for processing the cloud in the mixed mode according to claim 1, wherein the adopting a deep learning algorithm to dynamically adjust public cloud and private cloud allocation recommendation strategies, to obtain an optimal data preloading scheme, comprises:
the method comprises the steps of adjusting public cloud resources, evaluating private cloud resource allocation, judging data transmission allocation, evaluating the effect of a preloading scheme, adopting a ResNet network structure, reading cloud resource allocation conditions, judging cloud resource performance states, acquiring historical performance indexes, extracting characteristics of data types and sizes, analyzing data access frequency, training by adopting an Adagrad algorithm, monitoring the fitting or under-fitting state of a model, including finding problems, and continuously collecting data for real-time training by adjusting learning rate, so as to keep the optimal state of the model.
6. A mixed mode cloud processing system for performing a mixed mode cloud processing method according to any of claims 1-5, said system comprising:
the personalized demand acquisition module is used for acquiring personalized demands of users on privacy protection, tone quality and response time according to user operation habits;
the music data processing module is used for determining a data encryption, identity verification and voice enhancement method required by a user according to the user requirement and judging the network requirement, storage requirement and calculation requirement of cloud calculation required by the music data;
The music preference prediction and preloading module is used for predicting the music preference of a user according to the user interaction data, preloading the music content which can be played by the user in advance through mixed cloud computing, and reducing the buffer time;
the cost optimization and data distribution module is used for generating calculation cost and flow cost meeting the requirements according to the mixed cloud calculation requirements of the music data, and adjusting a data transmission distribution mode of the calculation service between public cloud and private cloud;
the dynamic adjustment and evaluation module is used for evaluating the fluctuation of tone quality, response time and calculation cost again if the number of users or the data to be preloaded are changed due to the change of music preference;
and the cloud resource optimization module is used for dynamically adjusting public cloud and private cloud allocation recommendation strategies by adopting a deep learning algorithm to obtain an optimal data preloading scheme.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311289674.3A CN117033693B (en) | 2023-10-08 | 2023-10-08 | Method and system for cloud processing in mixed mode |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311289674.3A CN117033693B (en) | 2023-10-08 | 2023-10-08 | Method and system for cloud processing in mixed mode |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117033693A CN117033693A (en) | 2023-11-10 |
| CN117033693B true CN117033693B (en) | 2024-03-08 |
Family
ID=88641500
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311289674.3A Active CN117033693B (en) | 2023-10-08 | 2023-10-08 | Method and system for cloud processing in mixed mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117033693B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103731372A (en) * | 2013-12-24 | 2014-04-16 | 东南大学 | Resource supply method for service supplier under hybrid cloud environment |
| CN107659433A (en) * | 2017-09-08 | 2018-02-02 | 中国联合网络通信集团有限公司 | A kind of cloud resource dispatching method and equipment |
| CN107888981A (en) * | 2017-11-16 | 2018-04-06 | 北京小米移动软件有限公司 | Audio frequency and video preload method, apparatus, equipment and storage medium |
| CN108733429A (en) * | 2018-05-16 | 2018-11-02 | Oppo广东移动通信有限公司 | Method of adjustment, device, storage medium and the mobile terminal of system resource configuration |
| CN109672709A (en) * | 2017-10-17 | 2019-04-23 | 上海仪电（集团）有限公司中央研究院 | A kind of mixed cloud service scheduling system and method |
| CN111491006A (en) * | 2020-03-03 | 2020-08-04 | 天津大学 | Load-aware cloud computing resource elastic distribution system and method |
| CN114357236A (en) * | 2021-12-31 | 2022-04-15 | 卓尔智联(武汉)研究院有限公司 | Music recommendation method and device, electronic equipment and computer readable storage medium |
| WO2022133996A1 (en) * | 2020-12-25 | 2022-06-30 | 深圳晶泰科技有限公司 | Hybrid cloud computing platform-based storage method and system |
| CN116126534A (en) * | 2023-01-28 | 2023-05-16 | 哈尔滨工业大学（威海） | Cloud resource dynamic expansion method and system |
| CN116233498A (en) * | 2023-03-09 | 2023-06-06 | 联通沃音乐文化有限公司 | Audio-video playing control method based on AI artificial intelligence |
| CN116684528A (en) * | 2023-05-25 | 2023-09-01 | 联通沃音乐文化有限公司 | Recommendation method for different visual angles of video color ring |
| CN116700920A (en) * | 2023-05-15 | 2023-09-05 | 中国工商银行股份有限公司 | Cloud primary hybrid deployment cluster resource scheduling method and device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9154826B2 (en) * | 2011-04-06 | 2015-10-06 | Headwater Partners Ii Llc | Distributing content and service launch objects to mobile devices |
-
2023
- 2023-10-08 CN CN202311289674.3A patent/CN117033693B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103731372A (en) * | 2013-12-24 | 2014-04-16 | 东南大学 | Resource supply method for service supplier under hybrid cloud environment |
| CN107659433A (en) * | 2017-09-08 | 2018-02-02 | 中国联合网络通信集团有限公司 | A kind of cloud resource dispatching method and equipment |
| CN109672709A (en) * | 2017-10-17 | 2019-04-23 | 上海仪电（集团）有限公司中央研究院 | A kind of mixed cloud service scheduling system and method |
| CN107888981A (en) * | 2017-11-16 | 2018-04-06 | 北京小米移动软件有限公司 | Audio frequency and video preload method, apparatus, equipment and storage medium |
| CN108733429A (en) * | 2018-05-16 | 2018-11-02 | Oppo广东移动通信有限公司 | Method of adjustment, device, storage medium and the mobile terminal of system resource configuration |
| CN111491006A (en) * | 2020-03-03 | 2020-08-04 | 天津大学 | Load-aware cloud computing resource elastic distribution system and method |
| WO2022133996A1 (en) * | 2020-12-25 | 2022-06-30 | 深圳晶泰科技有限公司 | Hybrid cloud computing platform-based storage method and system |
| CN114357236A (en) * | 2021-12-31 | 2022-04-15 | 卓尔智联(武汉)研究院有限公司 | Music recommendation method and device, electronic equipment and computer readable storage medium |
| CN116126534A (en) * | 2023-01-28 | 2023-05-16 | 哈尔滨工业大学（威海） | Cloud resource dynamic expansion method and system |
| CN116233498A (en) * | 2023-03-09 | 2023-06-06 | 联通沃音乐文化有限公司 | Audio-video playing control method based on AI artificial intelligence |
| CN116700920A (en) * | 2023-05-15 | 2023-09-05 | 中国工商银行股份有限公司 | Cloud primary hybrid deployment cluster resource scheduling method and device |
| CN116684528A (en) * | 2023-05-25 | 2023-09-01 | 联通沃音乐文化有限公司 | Recommendation method for different visual angles of video color ring |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117033693A (en) | 2023-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | CHARM: A cost-efficient multi-cloud data hosting scheme with high availability | |
| US8417816B2 (en) | Topology aware cache cooperation | |
| TWI514132B (en) | Data synchronization | |
| Sarhan et al. | Caching and scheduling in NAD-based multimedia servers | |
| US20140136571A1 (en) | System and Method for Optimizing Data Storage in a Distributed Data Storage Environment | |
| US20110107030A1 (en) | Self-organizing methodology for cache cooperation in video distribution networks | |
| US8954556B2 (en) | Utility-based model for caching programs in a content delivery network | |
| CN105376218B (en) | Streaming media system and method for quickly responding to user request | |
| CN109982104B (en) | Motion-aware video prefetching and cache replacement decision method in motion edge calculation | |
| Tu et al. | An optimized cluster storage method for real-time big data in Internet of Things | |
| JP2021515295A (en) | Dynamic authorization batch processing in a distributed storage network | |
| CN103631894A (en) | Dynamic copy management method based on HDFS | |
| Dong et al. | Correlation based file prefetching approach for hadoop | |
| CN114697683A (en) | Intelligent scheduling method, equipment and computer program product for streaming media file | |
| Haouari et al. | QoE-aware resource allocation for crowdsourced live streaming: A machine learning approach | |
| US6973536B1 (en) | Self-adaptive hybrid cache | |
| Akhtar et al. | Avic: a cache for adaptive bitrate video | |
| CN117033693B (en) | Method and system for cloud processing in mixed mode | |
| CN117041233B (en) | Distributed cloud computing method and system | |
| JP2012141885A (en) | Cache system, cache arrangement method, cache control device and cache control program | |
| Chikhaoui et al. | StorNIR, a multi-objective replica placement strategy for cloud federations | |
| Wei et al. | QoE oriented chunk scheduling in P2P-VoD streaming system | |
| Selvi et al. | Popularity (hit rate) based replica creation for enhancing the availability in cloud storage | |
| Cohen et al. | Self-adjusting advertisement of cache indicators with bandwidth constraints | |
| CN114860168A (en) | Cache placement method, system and medium for long and short time slot combination optimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |