System Integration Architecture and Cooperative Computing Method for the Collaboration of Cloud Computing and Edge Computing

Abstract

The present invention relates to the field of distributed computing and network technologies, and specifically, to a system integration architecture for the collaboration between cloud computing and edge computing and a collaborative computing method. It includes: an edge processing and caching unit that filters and caches the received raw data based on an adaptive data filtering strategy and an intelligent caching strategy; a cloud management and collaboration unit that constructs a delay and energy consumption optimization objective function based on the current states of cloud nodes and edge nodes, and uses a multi-objective optimization algorithm to solve the delay and energy consumption optimization objective function, dynamically collaborating on task scheduling and resource allocation; a heterogeneous hardware acceleration unit that calculates the optimal hardware matching degree using a hierarchical accelerator scheduling strategy, introduces a computing acceleration efficiency coefficient, and schedules heterogeneous accelerator resources. The system integration architecture for the collaboration between cloud computing and edge computing and the collaborative computing method combine a multi-objective optimization algorithm with intelligent matching of heterogeneous accelerators to achieve efficient collaborative scheduling between cloud computing and edge computing.

Description

Technical Field

[0001] The present invention relates to the field of distributed computing and network technologies, and more specifically, to a system integration architecture for cloud computing and edge computing collaboration and a collaborative computing method. Background Art

[0002] The system integration architecture for cloud computing and edge computing collaboration and the collaborative computing method aim to minimize the overall execution latency and energy consumption of the system and optimize resource utilization. Through multi-objective optimization algorithms and dynamic scheduling strategies, task allocation and resource allocation are controlled to achieve efficient task scheduling and resource management, improving system performance and energy efficiency.

[0003] Existing cloud computing and edge computing collaborative architectures usually have difficulty in efficiently dynamically scheduling computing tasks under high load conditions and flexibly selecting the most suitable computing resources according to the characteristics of tasks. Moreover, due to the existence of task complexity and computing resource heterogeneity, problems such as low accelerator resource utilization, long task execution latency, and low system energy efficiency will occur during different task allocations. Therefore, a system integration architecture for cloud computing and edge computing collaboration and a collaborative computing method are provided. Summary of the Invention

[0004] The purpose of the present invention is to provide a system integration architecture and method for cloud computing and edge computing collaboration to solve the problems of low accelerator resource utilization, long task execution latency, and low system energy efficiency that occur during different task allocations due to the existence of task complexity and computing resource heterogeneity as mentioned in the above background art.

[0005] To achieve the above purpose, the present invention provides a system integration architecture for cloud computing and edge computing collaboration, including: an edge processing cache unit, which filters and caches the received raw data based on an adaptive data filtering strategy and an intelligent caching strategy;

[0006] It further includes a cloud management collaboration unit, which is used to construct a latency-energy consumption optimization objective function and set constraint conditions between the edge and the cloud based on the current states of cloud nodes and edge nodes, and use a multi-objective optimization algorithm to solve the latency-energy consumption optimization objective function, dynamically collaborating task scheduling and resource allocation;

[0007] It further includes a heterogeneous hardware acceleration unit, which calculates the optimal hardware matching degree using a hierarchical accelerator scheduling strategy and introduces a computing acceleration efficiency coefficient to schedule heterogeneous accelerator resources.

[0008] As a further improvement of this technical solution, the edge processing cache unit includes an adaptive data filtering module and an intelligent cache management module;

[0009] Among them, the adaptive data filtering module uses an adaptive data filtering strategy to filter the collected raw data;

[0010] The intelligent cache management module uses an intelligent cache strategy to store the data after filtering into the cache.

[0011] As a further improvement of this technical solution, the adaptive data filtering strategy is implemented based on machine learning algorithms and statistical analysis techniques, and is used to identify the received raw data in real time and filter out the noise and irrelevant data in the data:

[0012] The adaptive data filtering module uses an adaptive data filtering strategy to filter the collected raw data, and the specific method steps are as follows:

[0013] S1.1.1. Receive the raw data from various terminal devices, obtain the business requirement weight vector, and use the data feature extraction algorithm to extract the raw data feature vector, and adaptively select the filtering strategy according to the business requirement weight vector and the raw data feature vector:

[0014] ;

[0015] Among them, is the time; is the time selected filtering strategy; is the set of all optional filtering strategies; is the filtering strategy; is the business requirement weight vector; is the time raw data feature vector of;

[0016] S1.1.2. Apply the filtering strategy selected at time to filter the raw data and obtain the filtered data:

[0017] ;

[0018] Among them, is the time filtered data volume; is the raw data; is the time data filtering ratio.

[0019] As a further improvement of this technical solution, the intelligent cache strategy is implemented based on the least recently used algorithm and predictive caching technology, and is used to optimize data storage and access;

[0020] The intelligent cache management module stores the data after filtering processing into the cache using an intelligent cache strategy. The specific method steps are as follows:

[0021] S1.2.1. Compress the filtered data and store it in the cache;

[0022] S1.2.2. Calculate the current cache hit rate based on the number of cache hit accesses and the total number of accesses:

[0023] ;

[0024] where, is the cache hit rate at time ; is the number of cache hit accesses at time ; is the total number of accesses at time ;

[0025] S1.2.3. Adjust the cache strategy parameters based on the cache hit rate and the target hit rate:

[0026] ;

[0027] where, is the cache strategy parameter at time ; is the updated cache strategy parameter at time t + 1; is the learning rate; is the target cache hit rate;

[0028] S1.2.4. Calculate the amount of data to be migrated based on the current cache capacity and the preset cache upper limit, and migrate the amount of data to be migrated to the cloud and edge nodes:

[0029] ;

[0030] where, is the amount of data to be migrated at time ; is the total amount of data in the current cache at time ; is the maximum capacity of the cache.

[0031] As a further improvement of this technical solution, the cloud management and coordination unit includes a task scheduling and allocation module, a delay and energy consumption solution module, and a network path optimization module;

[0032] Among them, the task scheduling and allocation module is used to construct a delay and energy consumption optimization objective function and set constraint conditions between the edge node and the cloud based on the current states of the cloud node and the edge node using a multi-objective optimization algorithm;

[0033] The time delay and energy consumption solving module solves the time delay and energy consumption optimization objective function based on the multi-objective optimization algorithm, and dynamically coordinates task scheduling and resource allocation;

[0034] The network path optimization module uses the Dijkstra algorithm to optimize the data transmission path and bandwidth allocation between the edge node and the cloud.

[0035] As a further improvement of this technical solution, the multi-objective optimization algorithm is implemented based on the task time delay requirement and energy consumption evaluation, and is used to dynamically coordinate task scheduling and resource allocation between the edge node and the cloud, so as to minimize the overall execution time delay and energy consumption of the system and optimize the resource utilization rate;

[0036] The task scheduling and allocation module is used to construct a time delay and energy consumption optimization objective function and set constraint conditions based on the current states of the cloud node and the edge node between the edge node and the cloud by using the multi-objective optimization algorithm. The specific method steps are as follows:

[0037] S2.1.1. Based on the collected original data, use the multi-objective optimization algorithm to construct a time delay and energy consumption optimization objective function:

[0038] ;

[0039] Among them, is the label of the task, indicating the th task; is the label of the edge node, indicating the th edge node; is the label of the cloud node, indicating the th cloud node; is the edge binary decision variable; is the cloud binary decision variable; is the weight factor of the time delay; is the weight factor of the energy consumption; is the execution duration of the th task on the th edge node; is the execution duration of the th task on the th cloud node; is the power consumption of the th task on the th edge node; is the power consumption of the th task on the th cloud node; is the total number of tasks; is the total number of edge nodes; is the total number of cloud nodes;

[0040] S2.1.2. Add the following constraint conditions to the delay and energy consumption optimization objective function:

[0041] Task unique allocation constraint:

[0042] ;

[0043] Node resource capacity constraint:

[0044] ;

[0045] ;

[0046] where is the demand for computing resources of the th task at time ; is the maximum available computing resource of the th edge at time ; is the maximum available computing resource of the th cloud node at time ;

[0047] Binary variable constraint:

[0048] ;

[0049] .

[0050] As a further improvement of this technical solution, the delay and energy consumption solving module solves the delay and energy consumption optimization objective function based on a multi-objective optimization algorithm, and dynamically coordinates task scheduling and resource allocation. The specific method steps are as follows:

[0051] S2.2.1. Based on the total number of tasks, the total number of edge nodes, and the total number of cloud nodes, select a strategy based on the optimization algorithm, and select an optimization algorithm for solving the delay and energy consumption optimization objective function applicable to the current moment;

[0052] S2.2.2. Based on the selected optimization algorithm, solve the delay and energy consumption optimization objective function at the current moment to obtain the optimal edge binary decision variable and the optimal cloud binary decision variable;

[0053] S2.2.3. Based on the optimal edge binary decision variable and the optimal cloud binary decision variable, construct an optimal task scheduling and resource allocation scheme;

[0054] Among them, the optimization algorithm selection strategy is implemented by integrating the linear programming algorithm, the mixed integer programming algorithm, the genetic algorithm, and the particle swarm optimization algorithm, and is used to select an optimization algorithm for solving the delay and energy consumption optimization objective function applicable to the current moment according to the total number of tasks, the total number of edge nodes, and the total number of cloud nodes. The specific optimization algorithm selection strategy is as follows:

[0055] If both the delay and energy consumption optimization objective function and the constraint conditions are linear, select the linear programming algorithm;

[0056] If the delay and energy consumption optimization objective function involves integer decision variables, select the mixed integer programming algorithm;

[0057] If the delay and energy consumption optimization objective function involves the global optimal solution, select the genetic algorithm;

[0058] If the delay and energy consumption optimization objective function seeks an efficient approximate solution, select the particle swarm optimization algorithm.

[0059] As a further improvement of this technical solution, the heterogeneous hardware acceleration unit includes an accelerator matching module and an accelerator scheduling module;

[0060] Among them, the accelerator matching module is used to collect accelerator performance data, calculate the acceleration efficiency coefficient of each task on different accelerators, and determine the optimal accelerator matching scheme. The specific method steps are as follows:

[0061] S3.1.1. Collect the performance metrics of the accelerator, including computing power, memory bandwidth, and power consumption;

[0062] Obtain the characteristic metrics of each task, including data volume, computational intensity, and real-time requirements;

[0063] S3.1.2. Based on the performance metrics of the accelerator and the characteristic metrics of each task, calculate the acceleration efficiency coefficient of each task:

[0064] ;

[0065] Among them, is the label of the accelerator, indicating the th accelerator; is the label of the task, indicating the th task; is the acceleration efficiency coefficient of the th task on the th accelerator; is the computing power of the th accelerator; is the memory bandwidth of the th accelerator; is the power consumption of the th accelerator; is the computational intensity of the th task; is the data volume of the th task;

[0066] S3.1.3. Match the optimal accelerator for each task based on the acceleration efficiency coefficient:

[0067] ;

[0068] and the acceleration efficiency coefficient of the th task on the th accelerator becomes the acceleration efficiency coefficient of the th task on the optimal accelerator ; is the optimal accelerator for the th task;

[0069] S3.1.4. Store the optimal accelerator matching scheme for each task in the cache and transmit it to the accelerator scheduling module.

[0070] As a further improvement of this technical solution, the accelerator scheduling module dynamically schedules the allocation of accelerator resources based on the optimal accelerator matching scheme through a hierarchical scheduling strategy and real-time resource monitoring. The specific method steps are as follows:

[0071] S3.2.1. Calculate the task priority based on the real-time requirement and acceleration efficiency coefficient of the task:

[0072] ;

[0073] where is the priority of the th task; is the acceleration efficiency coefficient of the th task on the optimal accelerator; is the real-time requirement of the th task;

[0074] S3.2.2. Select the task with the highest task priority:

[0075] Define the decision variable:

[0076] ;

[0077] where is the set of all accelerators;

[0078] Construct the resource allocation optimization objective function:

[0079] ;

[0080] Among them, is the decision variable for the th task assigned to the th accelerator;

[0081] Construct constraint conditions:

[0082] Task unique assignment constraint:

[0083] ;

[0084] Accelerator resource capacity constraint:

[0085] ;

[0086] Among them, is the demand for computing resources of the th task; is the maximum available computing resource of the th accelerator;

[0087] Binary variable constraint:

[0088] ;

[0089] S3.2.3. Based on the selected optimization algorithm, solve the resource allocation optimization objective function to obtain the optimal decision variable , and perform heterogeneous accelerator scheduling.

[0090] On the other hand, the present invention provides a collaborative computing method for cloud computing and the edge. Based on the above system integration architecture for cloud computing and edge computing collaboration, it includes the following steps:

[0091] S10.1. Based on the adaptive data filtering strategy and intelligent caching strategy, filter and cache the received original data;

[0092] S10.2. Based on the current cloud node and edge node states, construct a delay and energy consumption optimization objective function and set constraint conditions, and use a multi-objective optimization algorithm to solve the delay and energy consumption optimization objective function to dynamically coordinate task scheduling and resource allocation;

[0093] S10.3. Adopt a hierarchical accelerator scheduling strategy to calculate the optimal hardware matching degree, and introduce a computing acceleration efficiency coefficient to schedule heterogeneous accelerator resources.

[0094] Compared with the prior art, the beneficial effects of the present invention:

[0095] 1. In the system integration architecture and collaborative computing method of cloud computing and edge computing, based on the multi-objective optimization algorithm and dynamic scheduling strategy, the best computing path can be intelligently selected and the resource allocation can be dynamically adjusted according to the real-time requirements of tasks and the heterogeneous characteristics of computing resources, so as to achieve efficient computing task scheduling and resource utilization among different nodes.

[0096] 2. In the system integration architecture and collaborative computing method of cloud computing and edge computing, by introducing the accelerator performance matching and task characteristic analysis mechanism, the optimal scheduling of computing tasks between the cloud and edge nodes is realized, ensuring that the computing tasks are allocated according to the optimal matching degree of the accelerator, thereby improving the overall computing efficiency of the system, reducing latency and optimizing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 is the overall flowchart of the present invention;

[0098] The meanings of the reference numerals in the figure are as follows:

[0099] 1. Edge processing cache unit; 2. Cloud management collaboration unit; 3. Heterogeneous hardware acceleration unit; 11. Adaptive data filtering module; 12. Intelligent cache management module; 21. Task scheduling and allocation module; 22. Latency and energy consumption solving module; 23. Network path optimization module; 31. Accelerator matching module; 32. Accelerator scheduling module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0100] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0101] Embodiment 1: Please refer to Figure 1 As shown, a system integration architecture for the collaboration of cloud computing and edge computing is provided, including: an edge processing cache unit 1, which filters and caches the received raw data based on an adaptive data filtering strategy and an intelligent cache strategy;

[0102] The edge processing cache unit 1 includes an adaptive data filtering module 11 and an intelligent cache management module 12;

[0103] In this embodiment, the adaptive data filtering module 11 uses an adaptive data filtering strategy to filter the collected raw data;

[0104] The intelligent cache management module 12 stores the data after filtering processing into the cache using an intelligent cache strategy.

[0105] The adaptive data filtering strategy is implemented based on machine learning algorithms and statistical analysis techniques, and is used to identify the received raw data in real time and filter out the noise and irrelevant data in the data:

[0106] In this embodiment, the adaptive data filtering module 11 uses the adaptive data filtering strategy to filter the collected raw data, and the specific method steps are as follows:

[0107] S1.1.1. Receive the raw data from various terminal devices, obtain the business requirement weight vector, and use the data feature extraction algorithm to extract the raw data feature vector, and adaptively select the filtering strategy according to the business requirement weight vector and the raw data feature vector:

[0108] ;

[0109] Among them, is the time; is the time selected filtering strategy; is the set of all optional filtering strategies; is the filtering strategy; is the business requirement weight vector; is the time raw data feature vector at;

[0110] S1.1.2. Apply the filtering strategy selected at time to filter the raw data and obtain the filtered data:

[0111] ;

[0112] Among them, is the time filtered data volume; is the raw data; is the time data filtering ratio.

[0113] In this embodiment, the data feature extraction algorithm adopts principal component analysis to extract key features from high-dimensional raw data, reduce the data dimension, and retain as much data information as possible to support the selection of the adaptive filtering strategy;

[0114] All optional filtering strategy sets specifically include the following filtering strategies:

[0115] Threshold-based filtering strategy: Simply filter the data through a preset threshold;

[0116] Filtering strategy based on pattern recognition: Decide whether to retain by identifying specific patterns or trends in the data;

[0117] Filtering strategy based on machine learning: Use existing machine learning models to intelligently filter data;

[0118] Filtering strategy based on statistical analysis: Evaluate the abnormality or importance of data through statistical analysis methods and decide whether to retain data items.

[0119] In this embodiment, the intelligent caching strategy is implemented based on the least recently used algorithm and predictive caching technology, and is used to optimize data storage and access;

[0120] In this embodiment, the intelligent caching management module 12 uses the intelligent caching strategy to store the filtered data into the cache. The specific method steps are as follows:

[0121] S1.2.1. Compress the filtered data and store it in the cache;

[0122] S1.2.2. Calculate the current cache hit rate based on the number of cache hits and the total number of accesses:

[0123] ;

[0124] where is the cache hit rate at time ; is the number of cache hits at time ; is the total number of accesses at time ;

[0125] S1.2.3. Adjust the cache strategy parameters based on the cache hit rate and the target hit rate:

[0126] ;

[0127] where is the cache strategy parameter at time ; is the updated cache strategy parameter at time t+1; is the learning rate; is the target cache hit rate;

[0128] S1.2.4. Calculate the amount of data to be migrated based on the current cache capacity and the preset cache upper limit, and migrate the amount of data to be migrated to the cloud and edge nodes:

[0129] ;

[0130] wherein, is the time the amount of data to be migrated; is the time the total amount of data currently cached; is the maximum capacity of the cache.

[0131] It further includes a cloud management coordination unit 2, which is used to construct a delay and energy consumption optimization objective function and set constraint conditions between the edge and the cloud based on the current states of the cloud node and the edge node, and use a multi-objective optimization algorithm to solve the delay and energy consumption optimization objective function, and dynamically coordinate task scheduling and resource allocation;

[0132] In this embodiment, the cloud management coordination unit 2 includes a task scheduling and allocation module 21, a delay and energy consumption solving module 22, and a network path optimization module 23;

[0133] wherein, the task scheduling and allocation module 21 is used to construct a delay and energy consumption optimization objective function and set constraint conditions between the edge node and the cloud based on the current states of the cloud node and the edge node by using a multi-objective optimization algorithm;

[0134] The delay and energy consumption solving module 22 solves the delay and energy consumption optimization objective function based on the multi-objective optimization algorithm, and dynamically coordinates task scheduling and resource allocation;

[0135] The network path optimization module 23 uses the Dijkstra algorithm to optimize the data transmission path and bandwidth allocation between the edge node and the cloud.

[0136] In this embodiment, the Dijkstra algorithm is a classic algorithm for calculating the shortest path in a graph, and the advantages of using it to optimize the data transmission path and bandwidth allocation between the edge node and the cloud are as follows:

[0137] The Dijkstra algorithm can find the shortest path between any two nodes in the graph within polynomial time, which is suitable for real-time network path optimization requirements; it is applicable to different scales of network topologies, and can be effectively applied in both small-scale and large-scale network environments; the Dijkstra algorithm can adjust the path selection in real time according to the dynamic changes of the network state, can meet the real-time requirements of the network, and improve the reliability and efficiency of data transmission; combined with real-time network monitoring data, the Dijkstra algorithm can continuously optimize the transmission path and bandwidth allocation strategy.

[0138] The multi-objective optimization algorithm is implemented based on task delay requirements and energy consumption evaluation, and is used to dynamically coordinate task scheduling and resource allocation between the edge node and the cloud to minimize the overall execution delay and energy consumption of the system and optimize resource utilization;

[0139] In this embodiment, the task scheduling and allocation module 21 is used to construct a delay and energy consumption optimization objective function and set constraint conditions based on the current cloud node and edge node states between the edge node and the cloud using a multi-objective optimization algorithm. The specific method steps are as follows:

[0140] S2.1.1. Based on the collected raw data, use a multi-objective optimization algorithm to construct a delay and energy consumption optimization objective function:

[0141] ;

[0142] Among them, is the label of the task, indicating the th task; is the label of the edge node, indicating the th edge node; is the label of the cloud node, indicating the th cloud node; is the edge binary decision variable; is the cloud binary decision variable; is the weight factor of the delay; is the weight factor of the energy consumption; is the execution duration of the th task on the th edge node; is the execution duration of the th task on the th cloud node; is the power consumption of the th task on the th edge node; is the power consumption of the th task on the th cloud node; is the total number of tasks; is the total number of edge nodes; is the total number of cloud nodes;

[0143] S2.1.2. Add constraint conditions to the delay and energy consumption optimization objective function:

[0144] Task unique allocation constraint:

[0145] ;

[0146] Node resource capacity constraint:

[0147] ;

[0148] ;

[0149] Among them, is the demand for computing resources of the th task at time ; is the maximum available computing resources of the th edge at time ; is the maximum available computing resources of the th cloud node at time ;

[0150] Binary variable constraint:

[0151] ;

[0152] .

[0153] In this embodiment, the delay and energy consumption solving module 22 solves the delay and energy consumption optimization objective function based on a multi-objective optimization algorithm, and dynamically coordinates task scheduling and resource allocation. The specific method steps are as follows:

[0154] S2.2.1. According to the total number of tasks, the total number of edge nodes, and the total number of cloud nodes, select a strategy based on an optimization algorithm, and select an optimization algorithm for solving the delay and energy consumption optimization objective function applicable to the current moment;

[0155] S2.2.2. Based on the selected optimization algorithm, solve the delay and energy consumption optimization objective function at the current moment to obtain the optimal edge binary decision variable and the optimal cloud binary decision variable;

[0156] S2.2.3. Based on the optimal edge binary decision variable and the optimal cloud binary decision variable, construct an optimal task scheduling and resource allocation scheme;

[0157] Among them, the optimization algorithm selection strategy is implemented by integrating a linear programming algorithm, a mixed integer programming algorithm, a genetic algorithm, and a particle swarm optimization algorithm, and is used to select an optimization algorithm for solving the delay and energy consumption optimization objective function applicable to the current moment according to the total number of tasks, the total number of edge nodes, and the total number of cloud nodes. The optimization algorithm selection strategy is specifically as follows:

[0158] If both the delay and energy consumption optimization objective function and the constraint conditions are linear, select the linear programming algorithm;

[0159] If the delay and energy consumption optimization objective function involves integer decision variables, select the mixed integer programming algorithm;

[0160] If the delay and energy consumption optimization objective function involves a global optimal solution, select the genetic algorithm;

[0161] If the time-delay energy consumption optimization objective function seeks an efficient approximate solution, the particle swarm optimization algorithm is selected.

[0162] It further includes a heterogeneous hardware acceleration unit 3, which calculates the optimal hardware matching degree by adopting a hierarchical accelerator scheduling strategy and introduces a computing acceleration efficiency coefficient to schedule heterogeneous accelerator resources;

[0163] The heterogeneous hardware acceleration unit 3 includes an accelerator matching module 31 and an accelerator scheduling module 32;

[0164] In this embodiment, the accelerator matching module 31 is used to collect accelerator performance data, calculate the acceleration efficiency coefficient of each task on different accelerators, and determine the optimal accelerator matching scheme. The specific method steps are as follows:

[0165] S3.1.1. Collect the performance indicators of the accelerator, including computing power, memory bandwidth, and power consumption;

[0166] Obtain the characteristic indicators of each task, including data volume, computing intensity, and real-time requirement;

[0167] S3.1.2. Based on the performance indicators of the accelerator and the characteristic indicators of each task, calculate the acceleration efficiency coefficient of each task:

[0168] ;

[0169] Among them, is the label of the accelerator, indicating the th accelerator; is the label of the task, indicating the th task; is the th task on the th accelerator's acceleration efficiency coefficient; is the computing power of the th accelerator; is the memory bandwidth of the th accelerator; is the power consumption of the th accelerator; is the computing intensity of the th task; is the th task's data volume;

[0170] S3.1.3. Based on the acceleration efficiency coefficient, match the optimal accelerator for each task:

[0171] ;

[0172] And the th task on the The acceleration efficiency coefficient of the th task becomes the acceleration efficiency coefficient of the task on the optimal accelerator ; is the optimal accelerator for the th task;

[0173] S3.1.4. Store the optimal accelerator matching scheme for each task in the cache and transmit it to the accelerator scheduling module 32.

[0174] In this embodiment, based on the optimal accelerator matching scheme, the accelerator scheduling module 32 dynamically schedules the allocation of accelerator resources through a hierarchical scheduling strategy and real-time resource monitoring. The specific method steps are as follows:

[0175] S3.2.1. Calculate the task priority based on the real-time requirement and acceleration efficiency coefficient of the task:

[0176] ;

[0177] where is the priority of the th task; is the acceleration efficiency coefficient of the th task on the optimal accelerator; is the real-time requirement of the th task;

[0178] S3.2.2. Select the task with the highest task priority:

[0179] Define the decision variable:

[0180] ;

[0181] where is the set of all accelerators;

[0182] Construct the resource allocation optimization objective function:

[0183] ;

[0184] where is the decision variable for allocating the th task to the th accelerator;

[0185] Construct the constraint conditions:

[0186] Task unique allocation constraint:

[0187] ;

[0188] Accelerator resource capacity constraint:

[0189] ;

[0190] wherein, is the required computing resources for the th task pair; is the maximum available computing resources of the th accelerator;

[0191] Binary variable constraint:

[0192] ;

[0193] S3.2.3. Solve the resource allocation optimization objective function based on the selected optimization algorithm to obtain the optimal decision variable , and perform heterogeneous accelerator scheduling. Embodiment 2:

[0194] A collaborative computing method for cloud computing and edge, based on the above system integration architecture for cloud computing and edge computing collaboration, includes the following steps:

[0195] S10.1. Filter and cache the received raw data based on the adaptive data filtering strategy and intelligent caching strategy;

[0196] S10.2. Construct a delay and energy consumption optimization objective function and set constraint conditions based on the current cloud node and edge node states, and use a multi-objective optimization algorithm to solve the delay and energy consumption optimization objective function to dynamically collaborate task scheduling and resource allocation;

[0197] S10.3. Calculate the optimal hardware matching degree using a hierarchical accelerator scheduling strategy and introduce a computing acceleration efficiency coefficient to schedule heterogeneous accelerator resources.

[0198] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed.

Claims

1. The system integration architecture of cloud computing and edge computing collaboration is characterized by: include: An edge processing cache unit (1), wherein the edge processing cache unit (1) performs filtering processing and caching on the received original data based on an adaptive data filtering strategy and an intelligent caching strategy; The edge processing cache unit (1) comprises an adaptive data filtering module (11) and an intelligent cache management module (12); The adaptive data filtering module (11) uses an adaptive data filtering strategy to filter the collected raw data; The intelligent cache management module (12) uses an intelligent cache strategy to store the filtered data into a cache; A cloud management collaboration unit (2), the cloud management collaboration unit (2) is used to construct a delay energy consumption optimization objective function and set constraints between the edge and the cloud based on the current cloud node and edge node status, solve the delay energy consumption optimization objective function using a multi-objective optimization algorithm, and dynamically coordinate task scheduling and resource allocation; The cloud management collaboration unit (2) comprises a task scheduling and allocation module (21), a delay and energy consumption solution module (22) and a network path optimization module (23); The task scheduling and allocation module (21) is used to construct a delay and energy consumption optimization objective function and set constraint conditions between the edge node and the cloud based on the current cloud node and edge node status using a multi-objective optimization algorithm; The delay energy consumption solving module (22) solves the delay energy consumption optimization objective function based on a multi-objective optimization algorithm, and dynamically coordinates task scheduling and resource allocation; The network path optimization module (23) uses the Dijkstra algorithm to optimize the data transmission path and bandwidth allocation between the edge node and the cloud; A heterogeneous hardware acceleration unit (3), wherein the heterogeneous hardware acceleration unit (3) adopts a hierarchical accelerator scheduling strategy to calculate the optimal hardware matching degree, introduces a calculation acceleration efficiency coefficient, and schedules heterogeneous accelerator resources; The heterogeneous hardware acceleration unit (3) comprises an accelerator matching module (31) and an accelerator scheduling module (32); The accelerator matching module (31) is used to collect accelerator performance data, calculate the acceleration efficiency coefficient of each task on different accelerators, and determine the optimal accelerator matching solution; The accelerator scheduling module (32) dynamically schedules the allocation of accelerator resources based on the optimal accelerator matching solution through a hierarchical scheduling strategy and real-time resource monitoring.

2. The system integration architecture for collaborative cloud computing and edge computing according to claim 1 is characterized by: The adaptive data filtering strategy is implemented based on machine learning algorithms and statistical analysis techniques, and is used to identify the received raw data in real time and filter out noise and irrelevant data in the data; The intelligent cache strategy is implemented based on the least recently used algorithm and predictive cache technology to optimize data storage and access.

3. The system integration architecture for collaborative cloud computing and edge computing according to claim 2 is characterized by: The adaptive data filtering module (11) uses an adaptive data filtering strategy to filter the collected raw data. The specific method steps are as follows: S1.1.

1. Receive raw data from various terminal devices, obtain the business demand weight vector, and use the data feature extraction algorithm to extract the raw data feature vector, and adaptively select the filtering strategy based on the business demand weight vector and the raw data feature vector: ； in, for the moment; For the moment The selected filtering strategy; A collection of all optional filtering strategies; For filtering strategy; is the business demand weight vector; For the moment The original data feature vector of S1.1.2 Application time The selected filtering strategy is used to filter the original data to obtain the filtered data: ； in, For the moment Filter data volume; is the original data; For the moment Data filtering ratio.

4. The system integration architecture for collaborative cloud computing and edge computing according to claim 3 is characterized by: The intelligent cache management module (12) uses an intelligent cache strategy to store the filtered data in the cache. The specific method steps are as follows: S1.2.1, compressing the filtered data and storing it in the cache; S1.2.

2. Calculate the current cache hit rate based on the number of cache hit accesses and the total number of accesses: ； in, For the moment Cache hit rate; For the moment The number of cache hit accesses; For the moment Total number of visits; S1.2.

3. Adjust cache strategy parameters based on cache hit rate and target hit rate: ； in, For the moment Cache strategy parameters; The cache strategy parameters updated at time t+1; is the learning rate; is the target cache hit rate; S1.2.

4. Calculate the amount of data to be migrated based on the current cache capacity and the preset cache limit, and migrate the amount of data to be migrated to the cloud and edge nodes: ； in, For the moment The amount of data that needs to be migrated; For the moment The total amount of data currently cached; The maximum capacity of the cache.

5. The system integration architecture for collaborative cloud computing and edge computing according to claim 4, characterized in that: The multi-objective optimization algorithm is implemented based on task delay requirements and energy consumption evaluation, and is used to dynamically coordinate task scheduling and resource allocation between edge nodes and the cloud to minimize the overall execution delay and energy consumption of the system and optimize resource utilization.

6. The system integration architecture for collaborative cloud computing and edge computing according to claim 5, characterized in that: The task scheduling allocation module (21) is used to construct a delay energy consumption optimization objective function and set constraint conditions between the edge node and the cloud based on the current cloud node and edge node status using a multi-objective optimization algorithm. The specific method steps are as follows: S2.1.

1. Based on the collected raw data, a multi-objective optimization algorithm is used to construct the delay energy consumption optimization objective function: ； in, is the task number, indicating the tasks; is the label of the edge node, indicating the edge nodes; is the number of the cloud node, indicating the cloud nodes; is a marginal binary decision variable; is a binary decision variable in the cloud; is the weight factor of delay; is the weight factor of energy consumption; For the The task is in Execution time on edge nodes; For the The task is in Execution time on each cloud node; For the The task is in Power consumption on edge nodes; For the The task is in Power consumption on each cloud node; is the total number of tasks; is the total number of edge nodes; is the total number of cloud nodes; S2.1.

2. Add constraints to the latency energy consumption optimization objective function: Task unique assignment constraints: ； Node resource capacity constraints: ； ； in, For the Tasks at time The amount of computing resources required; For the The edge at the moment The maximum available computing resources; For the Cloud nodes at time The maximum available computing resources; Binary variable constraints: ； 。 7. The system integration architecture for collaborative cloud computing and edge computing according to claim 6, characterized in that: The delay energy consumption solving module (22) solves the delay energy consumption optimization objective function based on a multi-objective optimization algorithm, dynamically coordinates task scheduling and resource allocation, and the specific method steps are as follows: S2.2.

1. According to the total number of tasks, the total number of edge nodes and the total number of cloud nodes, based on the optimization algorithm selection strategy, select the optimization algorithm for solving the delay energy consumption optimization objective function applicable to the current moment; S2.2.

2. Based on the selected optimization algorithm, solve the delay energy consumption optimization objective function at the current moment to obtain the optimal edge binary decision variables and the optimal cloud binary decision variables; S2.2.

3. Based on the optimal edge binary decision variables and the optimal cloud binary decision variables, construct the optimal task scheduling and resource allocation solution; The optimization algorithm selection strategy is based on the integration of linear programming algorithm, mixed integer programming algorithm, genetic algorithm and particle swarm optimization algorithm, and is used to select the optimization algorithm for solving the delay energy consumption optimization objective function applicable to the current moment according to the total number of tasks, the total number of edge nodes and the total number of cloud nodes. The optimization algorithm selection strategy is as follows: If the delay energy consumption optimization objective function and constraints are both linear, the linear programming algorithm is selected; If the delay energy consumption optimization objective function involves integer decision variables, a mixed integer programming algorithm is selected; If the delay energy consumption optimization objective function involves a global optimal solution, a genetic algorithm is selected; If the delay energy consumption optimization objective function seeks an efficient approximate solution, the particle swarm optimization algorithm is selected.

8. The system integration architecture of cloud computing and edge computing collaboration according to claim 7 is characterized by: The specific method steps for the accelerator matching module (31) to determine the optimal accelerator matching solution are as follows: S3.1.

1. Collect performance metrics of the accelerator, including computing power, memory bandwidth, and power consumption; Obtain the characteristic indicators of each task, including data volume, computational intensity, and real-time requirements; S3.1.

2. Based on the performance indicators of the accelerator and the characteristic indicators of each task, calculate the acceleration efficiency coefficient of each task: ； in, is the number of the accelerator, indicating the accelerator; is the task number, indicating the tasks; For the The task is in The acceleration efficiency coefficient of each accelerator; For the The computing power of an accelerator; For the Memory bandwidth of each accelerator; For the The power consumption of each accelerator; For the The computational intensity of each task; For the The amount of data for each task; S3.1.

3. Based on the acceleration efficiency coefficient, match the optimal accelerator for each task: ； And the first The task is in The acceleration efficiency coefficient of the first accelerator becomes The acceleration efficiency coefficient of the task in the optimal accelerator ; For the The optimal accelerator for each task; S3.1.

4. The optimal accelerator matching solution for each task is stored in a cache and transmitted to the accelerator scheduling module (32).

9. The system integration architecture for collaborative cloud computing and edge computing according to claim 8, characterized in that: The specific method steps of the accelerator scheduling module (32) for dynamically scheduling accelerator resource allocation are as follows: S3.2.

1. Calculate the task priority based on the real-time requirements and acceleration efficiency coefficient of the task: ； in, For the The priority of each task; For the The acceleration efficiency coefficient of each task on the optimal accelerator; For the The real-time requirements of each task; S3.2.

2. Select the task with the highest priority: Define the decision variables: ； in, is the set of all accelerators; Construct resource allocation optimization objective function: ； in, For the Assign tasks to The decision variables of the accelerator; Build constraints: Task unique assignment constraints: ； Accelerator resource capacity constraints: ； in, For the The amount of computing resources required by each task; For the The maximum available computing resources of an accelerator; Binary variable constraints: ； S3.2.

3. Based on the selected optimization algorithm, solve the resource allocation optimization objective function to obtain the optimal decision variables , perform heterogeneous accelerator scheduling.

10. A collaborative computing method for cloud computing and edge computing, based on a system integration architecture for collaborative cloud computing and edge computing as claimed in any one of claims 1 to 9, characterized in that: The steps include: S10.

1. Filter and cache the received raw data based on an adaptive data filtering strategy and an intelligent caching strategy; S10.

2. Based on the current status of cloud nodes and edge nodes, construct the latency and energy consumption optimization objective function and set constraints, use the multi-objective optimization algorithm to solve the latency and energy consumption optimization objective function, and dynamically coordinate task scheduling and resource allocation; S10.

3. Use a hierarchical accelerator scheduling strategy to calculate the optimal hardware matching degree, introduce a calculation acceleration efficiency coefficient, and schedule heterogeneous accelerator resources.

Images