A game service node scheduling method based on edge computing

Abstract

The application relates to the technical field of resource allocation, and discloses a game service node scheduling method based on edge computing, which comprises the following steps: collecting all online user information and node information, and calculating the residual resource values of the nodes; combining the game levels, historical records and delay tolerance of different types of games of the users to construct a delay demand matrix and obtain the acceptable delay threshold of each user; establishing a node load evolution model through a historical load sequence, predicting future resource use trends, and calculating schedulable resources; modeling the online behavior of the users, counting the active frequencies of the users in different time periods, clustering the users according to geographical positions, and forming a user cluster set; screening nodes that meet delay requirements, and constructing a candidate node set according to the schedulable resources; allocating resource shares to the users in the clusters through a dynamic allocation mechanism and correcting the resource shares; and monitoring the node load, and if the node load exceeds a threshold, triggering a migration strategy to adjust resource distribution.

Description

Technical Field

[0001] This invention relates to the field of resource allocation technology, specifically to a game service node scheduling method based on edge computing. Background Technology

[0002] With the rapid development of online games and massively multiplayer online interactive applications, users' demands for game service quality are constantly increasing, especially in terms of real-time performance, stability, and smooth interaction. While traditional cloud computing architectures can provide powerful computing and storage capabilities, the long data transmission paths often result in high network latency, making it difficult to meet the millisecond-level response requirements of real-time gaming scenarios. At the same time, uneven distribution of resources in cloud centers can easily lead to service quality degradation due to dense user activity in certain areas. Therefore, how to rationally utilize the distributed advantages of edge computing nodes to reduce user access latency and achieve dynamic load balancing has become a pressing technical challenge.

[0003] Existing edge computing scheduling methods mostly focus on improving resource utilization or balancing node load, often neglecting the individualized latency requirements of users. For example, in existing scheduling schemes, users are treated as having homogeneous service requests, lacking differentiated considerations for different game types, user levels, and historical latency performance. This results in some highly sensitive users still experiencing lag or operational delays, thus affecting the overall experience. Furthermore, the dynamic nature of user behavior is not adequately considered, especially users' online frequency and behavioral patterns. If this information is not effectively predicted and utilized, the scheduling strategy will deviate from its optimal performance in actual execution.

[0004] On the other hand, existing research on node load prediction and scheduling often employs estimations or simplified models based on fixed parameters, failing to fully utilize historical sequence data to infer future trends. This approach often lacks adaptability in the face of sudden traffic surges or concentrated user logins, easily leading to node overload or resource waste. Especially in high-concurrency scenarios, if the scheduling strategy lacks real-time dynamic adjustment capabilities, severe service delays or even outages may occur when the load on some nodes exceeds the threshold.

[0005] In summary, there is an urgent need to propose a new game service node scheduling method based on edge computing, which can comprehensively consider user behavior characteristics, node resource status and latency requirements, and achieve efficient resource utilization and significant improvement in user experience through adaptive prediction and scheduling mechanisms. Summary of the Invention

[0006] This invention provides a game service node scheduling method based on edge computing, which helps to solve the problems mentioned in the background art above.

[0007] This invention provides the following technical solution: a game service node scheduling method based on edge computing, comprising:

[0008] Obtain online user distribution information and edge node resource information, and calculate the comprehensive value of remaining resources for each node;

[0009] Collect edge node resource information, denoted as a set. ,in This represents the resource information of the k-th edge node;

[0010] Construct a user latency requirement matrix and determine latency thresholds by combining user game level, historical latency, and game type tolerance.

[0011] Construct a node load evolution model to predict future loads and calculate schedulable resources based on historical load sequences;

[0012] Predicting user online behavior and clustering them by location yields a set of user clusters;

[0013] For each cluster, select nodes with acceptable latency and construct a candidate node set based on schedulable resources;

[0014] Dynamic load scheduling is performed based on the candidate node set to allocate node resource shares to users within the cluster and perform corrections.

[0015] Node load is monitored in real time, and load migration is triggered when the load exceeds the threshold to achieve scheduling adjustment.

[0016] Optionally, the step of obtaining online user distribution information and edge node resource information, and calculating the comprehensive value of remaining resources for each node, includes:

[0017] Collect the locations of all online users, denoted as set. ,in This indicates the position of the mu-th user;

[0018] The edge node resource information includes CPU utilization. Memory usage and network bandwidth ,in ;

[0019] For each node The current total value of remaining resources is calculated as follows:

[0020] ;

[0021] in:

[0022] Represents a node Current total value of remaining resources;

[0023] Representing nodes respectively Current , and Maximum available value.

[0024] Optionally, the step of constructing a user latency requirement matrix and determining latency thresholds by combining user game level, historical latency, and game type tolerance includes:

[0025] The network latency between online user devices and edge nodes is obtained and denoted as a matrix. L Specifically, it is expressed as: ;

[0026] in, express To the node Real-time network latency;

[0027] Set a latency tolerance threshold for each online user. The details are as follows:

[0028] ;

[0029] in:

[0030] For the first The game level of each online user;

[0031] The lowest game level among all online users;

[0032] The highest game level among all online users;

[0033] For online users The historical average latency is as follows:

[0034] ;

[0035] in, For the traversal sequence number, For online users Historical login count, For online users In the The actual delay during the first login;

[0036] For game type The average tolerable delay is calculated as follows:

[0037] ;

[0038] in, For the traversal sequence number, For game type The total number of online users For game type The Middle The actual latency for each online user;

[0039] Constructing a Delayed Demand Matrix D The details are as follows:

[0040] ;

[0041] in, Indicates the delay tolerance threshold and The difference between them;

[0042] like , then it represents a node Satisfying online users The demand;

[0043] Otherwise, node Not satisfied with online users The demand.

[0044] Optionally, the construction of the node load evolution model, which predicts future load and calculates schedulable resources based on historical load sequences, includes:

[0045] Get each node The historical load time series is denoted as ;

[0046] in, Indicates the node at time t The total load is denoted as ;

[0047] The node load evolution model is constructed as follows:

[0048] ;

[0049] in:

[0050] Indicates the prediction node The overall load at the next time point t;

[0051] P Indicates a time-based sliding window;

[0052] Indicates time Next node The overall load;

[0053] pass and Calculate the current node schedulable resources The specific expression is: .

[0054] Optionally, the step of predicting user online behavior and clustering them by location to obtain a set of user clusters includes:

[0055] Statistics on each user's online frequency at different time periods and input frequency sets The online frequency acquisition process is as follows:

[0056] Divide a day into O equal time slots, with each time slot lasting for a duration of... ;

[0057] Select the fre_day day closest to the current day as the data to be collected;

[0058] Construct the frequency generation function as follows:

[0059] ;

[0060] in, fd For the traversal sequence number, Indicates online users In the fd Heavenly time Total online time for each time slot;

[0061] Divide all online users into M clusters based on their location;

[0062] Online users are clustered into groups as follows:

[0063] ;

[0064] in, This represents the m-th cluster.

[0065] Optionally, for each cluster, filtering nodes with acceptable latency and constructing a candidate node set based on schedulable resources includes:

[0066] For each cluster Select nodes with acceptable latency, as follows:

[0067] ;

[0068] in, Cluster The corresponding set of acceptable nodes;

[0069] Will Internal elements are categorized according to schedulable resources Sort from highest to lowest, as follows:

[0070] ;

[0071] in:

[0072] Indicates schedulable resources Sort by height from highest to lowest , denoted as cluster The set of candidate nodes;

[0073] This indicates a descending order operation.

[0074] Optionally, the step of performing dynamic load scheduling based on the candidate node set, allocating node resource shares to users within the cluster and performing corrections, includes:

[0075] For each cluster Set an initialization allocation matrix for each. ;

[0076] in, This indicates that it is assigned to online users. nodes The share of resources;

[0077] Traverse the candidate node set For each node, resource shares are allocated as follows:

[0078] ;

[0079] The nodes, after the initial allocation of node resource shares, will be corrected and allocated as follows:

[0080] .

[0081] Optionally, the real-time monitoring of node load, triggering load migration to achieve scheduling adjustment when the load exceeds a threshold, includes:

[0082] Get each node in real time The load, denoted as ;

[0083] Set a node load threshold to define the node's capacity to handle loads.

[0084] When node When the load exceeds the node's load threshold, load migration is triggered, as follows:

[0085] ;

[0086] in, This indicates that it is assigned to online users. nodes The share of resources;

[0087] Represents a node Scheduled resources;

[0088] Represents a node The remaining share of resources.

[0089] The present invention has the following beneficial effects:

[0090] 1. This edge computing-based game service node scheduling method acquires online user distribution information and edge node resource information, and further calculates the comprehensive value of remaining resources for each node, achieving accurate quantification of resource status. This scheme integrates user and node information before scheduling, overcoming the limitations of existing methods that rely solely on single-item remaining resource indicators. By comprehensively quantifying three key resources—CPU, memory, and bandwidth—a unified comprehensive value of remaining resources is formed, providing a more comprehensive and accurate data foundation for scheduling decisions. This approach avoids overall node performance imbalance caused by overload of a single resource, reducing service latency while ensuring resource utilization. Furthermore, the synchronous acquisition of user distribution information and node status means the scheduling strategy no longer relies on global broadcast information synchronization, but rather on real-time sampling and calculation, thereby reducing additional system communication overhead. This mechanism is particularly suitable for large-scale multiplayer online scenarios, significantly improving service stability and flexibility. Compared with conventional methods, this solution emphasizes the synchronous consideration of multi-dimensional resources, avoiding the risks of previous technologies that rely on a single indicator to make scheduling decisions, and ensuring that the scheduling strategy remains robust and efficient under high concurrency conditions.

[0091] 2. This edge computing-based game service node scheduling method constructs a user latency demand matrix and combines it with user game level, historical latency, and game type tolerance to form personalized latency thresholds. First, a user's game level reflects their operational proficiency and competitive needs; higher-level users are more sensitive to latency and therefore require lower latency thresholds. Second, a user's historical average latency provides a quantitative reference for individual network stability, accurately reflecting their actual usage environment. Finally, the average tolerance for game type ensures that latency threshold differences across different gameplay modes are reflected. The resulting latency demand matrix, combining these three factors, not only satisfies individualized needs but also ensures overall fairness. In practical applications, this matrix can accurately select edge nodes that meet user requirements, reducing disconnections and game experience degradation caused by latency mismatch. Compared to existing single latency threshold solutions, this matrix achieves a fine-grained balance between user experience and resource scheduling, enabling the system to flexibly adapt to complex user groups. This individualized dynamic modeling method solves the problem of neglecting user differences in traditional scheduling, significantly improving service satisfaction.

[0092] 3. This edge computing-based game service node scheduling method achieves early perception of node future states through a historical load evolution model. Compared to traditional reactive strategies, it can predict and warn of resource strain trends before scheduling. Through a time-sliding window design, the model captures the dynamic changes of nodes at different times, ensuring that the prediction results reflect not only short-term fluctuations but also long-term trends. Combining the predicted schedulable resources with latency requirements, nodes that do not meet latency requirements can be quickly eliminated, and the remaining nodes are sorted according to the size of their schedulable resources to form a candidate set. This approach prioritizes high-resource nodes while ensuring low latency, significantly improving scheduling success rate and resource utilization. Simultaneously, the sorted candidate set provides high-quality input for subsequent dynamic load allocation, avoiding uneven resource utilization or the selection of overloaded nodes. Compared to traditional schemes that rely solely on current load, the evolutionary model and sorting mechanism proposed in this invention achieve a dual combination of predictability and constraint, not only solving the problems of resource overload and latency imbalance but also improving the foresight and robustness of scheduling decisions.

[0093] 4. This edge computing-based game service node scheduling method establishes an online frequency model through user historical behavior statistics and clusters users based on location, dividing them into multiple spatial clusters. This design significantly reduces the complexity of scheduling large-scale users simultaneously, transforming the scheduling algorithm from a global problem into a local one. Furthermore, users within the same cluster typically have similar geographical latency and resource needs, facilitating more reasonable matching strategies during node resource allocation. Subsequently, dynamic load scheduling establishes a mapping relationship between clusters and candidate nodes, ensuring fairness and efficiency through a resource share allocation matrix. After the initial allocation, a correction mechanism further eliminates biases caused by intra-cluster differences, ensuring that node resources are accurately allocated to the users who need them most. This grouping and allocation mechanism not only reduces resource waste but also effectively mitigates the risk of node overload. Compared to existing allocation methods based on single users, this solution is more efficient in large-scale scenarios, ensuring overall latency control while avoiding individual user experience degradation, reflecting an innovative approach that balances group scheduling with individual needs.

[0094] 5. This edge computing-based game service node scheduling method, by further introducing real-time monitoring and migration strategies on top of existing prediction and scheduling mechanisms, forms a closed-loop scheduling system. By continuously acquiring node load and comparing it with set thresholds, the system can instantly detect node anomalies. Once a node load exceeds its capacity, the migration mechanism quickly selects a suitable node for task migration based on existing allocation shares and remaining schedulable resources. This effectively prevents node downtime and user disconnections caused by sudden traffic surges or concentrated user online activity. Compared to traditional static scheduling, this solution's dynamic migration mechanism has adaptive capabilities, continuously correcting scheduling results during operation to ensure system stability under high concurrency and dynamic changes. This design particularly addresses the common scheduling rigidity problem in existing methods, namely, the lack of subsequent adjustment capabilities once the initial allocation is completed. Through real-time monitoring and migration, this solution achieves a complete closed loop from prediction to execution to adjustment, greatly improving service reliability and user experience in edge computing environments. Attached Figure Description

[0095] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0096] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0097] Example 1, see Figure 1 A game service node scheduling method based on edge computing, characterized by comprising:

[0098] Obtain online user distribution information and edge node resource information, and calculate the comprehensive value of remaining resources for each node;

[0099] Collect edge node resource information, denoted as a set. ,in This represents the resource information of the k-th edge node;

[0100] Construct a user latency requirement matrix and determine latency thresholds by combining user game level, historical latency, and game type tolerance.

[0101] Construct a node load evolution model to predict future loads and calculate schedulable resources based on historical load sequences;

[0102] Predicting user online behavior and clustering them by location yields a set of user clusters;

[0103] For each cluster, select nodes with acceptable latency and construct a candidate node set based on schedulable resources;

[0104] Dynamic load scheduling is performed based on the candidate node set to allocate node resource shares to users within the cluster and perform corrections.

[0105] Node load is monitored in real time, and load migration is triggered when the load exceeds the threshold to achieve scheduling adjustment.

[0106] The process of acquiring online user distribution information and edge node resource information, and calculating the comprehensive value of remaining resources for each node, includes:

[0107] Collect the locations of all online users, denoted as set. ,in This indicates the position of the mu-th user;

[0108] The edge node resource information includes CPU utilization. Memory usage and network bandwidth ,in ;

[0109] For each node The current total value of remaining resources is calculated as follows:

[0110] ;

[0111] in:

[0112] Represents a node Current total value of remaining resources;

[0113] Representing nodes respectively Current , and Maximum available value.

[0114] By acquiring online user distribution information and edge node resource information, and further calculating the comprehensive remaining resource value of each node, the scheme achieves accurate quantification of resource status. This scheme integrates user and node information before scheduling, overcoming the limitations of existing methods that rely solely on single-item remaining resource indicators. By comprehensively quantifying three key resources—CPU, memory, and bandwidth—a unified comprehensive remaining resource value is formed, providing a more comprehensive and accurate data foundation for scheduling decisions. This approach avoids overall node performance imbalance caused by overload of a single resource, reducing service latency while ensuring resource utilization. Furthermore, the synchronous acquisition of user distribution information and node status means the scheduling strategy no longer relies on global broadcast information synchronization, but rather on real-time sampling and calculation, thus reducing additional system communication overhead. This mechanism is particularly suitable for large-scale, multi-user online scenarios, significantly improving service stability and flexibility. Compared to conventional methods, this scheme emphasizes the synchronous consideration of multi-dimensional resources, avoiding the risks associated with previous technologies that relied on a single indicator for scheduling decisions, ensuring the scheduling strategy remains robust and efficient under high concurrency conditions.

[0115] The construction of the user latency requirement matrix, which combines user game level, historical latency, and game type tolerance to determine latency thresholds, includes:

[0116] The network latency between online user devices and edge nodes is obtained and denoted as a matrix. L Specifically, it is expressed as: ;

[0117] in, express To the node Real-time network latency can be measured via ICMP or UDP heartbeat packets;

[0118] Set a latency tolerance threshold for each online user. The details are as follows:

[0119] ;

[0120] in:

[0121] For the first The game level of an online user is generally higher. Users with higher levels are more skilled and more sensitive to latency, so the required threshold is relatively lower.

[0122] The lowest game level among all online users;

[0123] The highest game level among all online users;

[0124] For online users The historical average latency is as follows:

[0125] ;

[0126] in, For the traversal sequence number, For online users Historical login count, For online users In the The actual delay during the first login;

[0127] For game type The average tolerable delay is calculated as follows:

[0128] ;

[0129] in, For the traversal sequence number, For game type The total number of online users For game type The Middle The actual latency for each online user;

[0130] Constructing a Delayed Demand Matrix D The details are as follows:

[0131] ;

[0132] in, Indicates the delay tolerance threshold and The difference between them;

[0133] like , then it represents a node Satisfying online users The demand;

[0134] Otherwise, node Not satisfied with online users The demand.

[0135] By constructing a user latency requirement matrix and combining it with user game level, historical latency, and game type tolerance, personalized latency thresholds are formed. First, a user's game level reflects their skill level and competitive needs; higher-level users are more sensitive to latency and therefore require lower latency thresholds. Second, a user's historical average latency provides a quantitative reference for individual network stability, accurately reflecting their actual usage environment. Finally, the average tolerance for game type ensures that latency threshold differences across different gameplay modes are reflected. The resulting latency requirement matrix, when combined, not only meets individual differences but also ensures overall fairness. In practical applications, this matrix can accurately identify edge nodes that meet user requirements, reducing disconnections and game experience degradation caused by latency mismatch. Compared to existing single latency threshold solutions, this matrix achieves a fine-grained balance between user experience and resource scheduling, enabling the system to flexibly adapt to complex user groups. This individualized dynamic modeling method solves the problem of neglecting user differences in traditional scheduling, significantly improving service satisfaction.

[0136] The construction of the node load evolution model, which predicts future load and calculates schedulable resources based on historical load sequences, includes:

[0137] Get each node The historical load time series is denoted as ;

[0138] in, Indicates the node at time t The total load is denoted as The total load is the sum of CPU, memory, and bandwidth usage.

[0139] The node load evolution model is constructed as follows:

[0140] ;

[0141] in:

[0142] Indicates the prediction node The overall load at the next time point t;

[0143] P Indicates a time-based sliding window;

[0144] Indicates time Next node The overall load;

[0145] pass and Calculate the current node schedulable resources The specific expression is: .

[0146] The process of predicting user online behavior and clustering them by location to obtain a set of user clusters includes:

[0147] Statistics on each user's online frequency at different time periods and input frequency sets The online frequency acquisition process is as follows:

[0148] Divide a day into equal proportions O There are 1 time slot, and the duration of each time slot is 1. ;

[0149] Select the closest to the day fre_day The sky is used as the data collection point;

[0150] Construct the frequency generation function as follows:

[0151] ;

[0152] in, fd For the traversal sequence number, Indicates online users In the fd Heavenly time Total online time for each time slot;

[0153] Divide all online users into M clusters based on their location;

[0154] Online users are clustered into groups as follows:

[0155] ;

[0156] in, This represents the m-th cluster.

[0157] For each cluster, the process of filtering nodes with acceptable latency and constructing a candidate node set based on schedulable resources includes:

[0158] For each cluster Select nodes with acceptable latency, as follows:

[0159] ;

[0160] in, Cluster The corresponding set of acceptable nodes;

[0161] Will Internal elements are categorized according to schedulable resources Sort from highest to lowest, as follows:

[0162] ;

[0163] in:

[0164] Indicates schedulable resources Sort by height from highest to lowest , denoted as cluster The set of candidate nodes;

[0165] This indicates a descending order operation.

[0166] By employing a historical load evolution model, the future state of nodes can be predicted in advance. Compared to traditional reactive strategies, this approach can predict and warn of resource strain trends before scheduling. Through a time-sliding window design, the model captures the dynamic changes of nodes at different times, ensuring that the predictions reflect not only short-term fluctuations but also long-term trends. Combining the predicted schedulable resources with latency requirements, nodes that do not meet latency demands can be quickly eliminated, and the remaining nodes are sorted according to the size of their schedulable resources to form a candidate set. This approach prioritizes high-resource nodes while maintaining low latency, significantly improving scheduling success rate and resource utilization. Simultaneously, the sorted candidate set provides high-quality input for subsequent dynamic load allocation, preventing uneven resource utilization or the selection of overloaded nodes. Compared to traditional solutions that rely solely on current load, the evolutionary model and sorting mechanism proposed in this invention combine predictability and constraint, not only solving the problems of resource overload and latency imbalance but also enhancing the foresight and robustness of scheduling decisions.

[0167] The process of performing dynamic load scheduling based on the candidate node set, allocating node resource shares to users within the cluster, and making corrections includes:

[0168] For each cluster Set an initialization allocation matrix for each. ;

[0169] in, This indicates that it is assigned to online users. nodes The share of resources;

[0170] Traverse the candidate node set For each node, resource shares are allocated as follows:

[0171] ;

[0172] The nodes, after the initial allocation of node resource shares, will be corrected and allocated as follows:

[0173] .

[0174] By establishing an online frequency model through user historical behavior statistics and combining it with location for clustering, users are divided into multiple spatial clusters. This design significantly reduces the complexity of scheduling large-scale users simultaneously, transforming the scheduling algorithm from a global problem into a local one. Furthermore, users within the same cluster typically have similar geographical latency and resource needs, facilitating more reasonable matching strategies during node resource allocation. Subsequently, dynamic load scheduling establishes a mapping relationship between clusters and candidate nodes, ensuring fairness and efficiency through a resource share allocation matrix. After the initial allocation, a correction mechanism further eliminates biases caused by intra-cluster differences, ensuring that node resources are accurately allocated to the users who need them most. This mechanism combining grouping and allocation not only reduces resource waste but also effectively mitigates the risk of node overload. Compared to existing allocation methods based on individual users, this solution is more efficient in large-scale scenarios, ensuring overall latency control while avoiding degradation in individual user experience, demonstrating an innovative approach that balances group-based scheduling with individual needs.

[0175] The real-time monitoring node load, when exceeding a threshold, triggers load migration to achieve scheduling adjustments, including:

[0176] Get each node in real time The load, denoted as ;

[0177] Set a node load threshold to define the node's capacity to handle loads.

[0178] When node When the load exceeds the node's load threshold, load migration is triggered, as follows:

[0179] ;

[0180] in, This indicates that it is assigned to online users. nodes The share of resources;

[0181] Represents a node Scheduled resources;

[0182] Represents a node The remaining share of resources.

[0183] By building upon existing prediction and scheduling mechanisms, a real-time monitoring and migration strategy is introduced to form a closed-loop scheduling system. By continuously acquiring node load and comparing it with set thresholds, the system can instantly detect node anomalies. Once a node load exceeds its capacity, the migration mechanism quickly selects a suitable node for task migration based on existing allocation shares and remaining schedulable resources. This effectively prevents node downtime and user disconnections caused by sudden traffic surges or concentrated user online activity. Compared to traditional static scheduling, this solution's dynamic migration mechanism is adaptive, continuously correcting scheduling results during operation to ensure system stability under high concurrency and dynamic changes. This design particularly addresses the common scheduling rigidity problem in existing methods, namely the lack of subsequent adjustment capabilities once initial allocation is complete. Through real-time monitoring and migration, this solution achieves a complete closed loop from prediction to execution to adjustment, significantly improving service reliability and user experience in edge computing environments.

Claims

1. A game service node scheduling method based on edge computing, characterized in that, include: Obtain online user location distribution information and edge node resource information, and calculate the comprehensive value of remaining resources for each node; Collect the locations of all online users, denoted as set. ,in This indicates the position of the mu-th user; Collect edge node resource information and denote all edge nodes as a set. ,in This represents the k-th edge node; Construct a user latency requirement matrix, and determine latency thresholds by combining user game level, historical average latency, and average tolerable latency for game type. Specifically, this includes: The network latency between online user devices and edge nodes is obtained and denoted as a matrix. L Specifically, it is expressed as: ; in, Indicates the distance from the i-th user to the node. Real-time network latency; Set a latency tolerance threshold for each online user. The details are as follows: ； in: For the first The game level of each online user; The lowest game level among all online users; The highest game level among all online users; For online users The historical average latency is as follows: ； in, For the traversal sequence number, For online users Historical login count, For online users In the The actual delay during the first login; For game type The average tolerable delay is calculated as follows: ； in, For the traversal sequence number, For game type The total number of online users For game type The Middle The actual latency for each online user; Constructing a Delayed Demand Matrix D The details are as follows: ； in, Indicates the delay tolerance threshold and The difference between them; like , then it represents a node Satisfying online users The demand; Otherwise, node Not satisfied with online users The demand; Construct a node load evolution model to predict future loads and calculate schedulable resources based on historical load sequences; Predicting user online behavior and clustering them by location yields a set of user clusters; For each cluster, select nodes with acceptable latency and construct a candidate node set based on schedulable resources; Dynamic load scheduling is performed based on the candidate node set to allocate node resource shares to users within the cluster and perform corrections. Node load is monitored in real time, and load migration is triggered when the load exceeds the threshold to achieve scheduling adjustment.

2. The game service node scheduling method based on edge computing according to claim 1, characterized in that: The process of acquiring online user location distribution information and edge node resource information, and calculating the comprehensive value of remaining resources for each node, includes: The edge node resource information includes CPU utilization. Memory usage and network bandwidth ; For each node The current total value of remaining resources is calculated as follows: ； in: Represents a node Current total value of remaining resources; Representing nodes respectively Current , and Maximum available value.

3. The game service node scheduling method based on edge computing according to claim 2, characterized in that: The construction of the node load evolution model, which predicts future load and calculates schedulable resources based on historical load sequences, includes: Get each node The historical load time series is denoted as ; in, Indicates the node at time t The total load is denoted as The total load is the sum of CPU, memory, and bandwidth usage. The node load evolution model is constructed as follows: ； in: Indicates the prediction node The overall load at the next time point t; P Indicates a time-based sliding window; Indicates time Next node The overall load; Indicates time Next node The overall load; pass and Calculate the current node schedulable resources The specific expression is: .

4. The game service node scheduling method based on edge computing according to claim 3, characterized in that: The process of predicting user online behavior and clustering them by location to obtain a set of user clusters includes: Statistics on each user's online frequency at different time periods and input frequency sets The online frequency acquisition process is as follows: Divide a day into O equal time slots, with each time slot lasting for a duration of... ; Select the closest to the day fre_day The sky is used as the data collection point; Construct the frequency generation function as follows: ； in, fd For the traversal sequence number, Indicates online users In the fd Heavenly time Total online time for each time slot; Divide all online users into M clusters based on their location; Online users are clustered into groups as follows: ； in, This represents the m-th cluster.

5. The game service node scheduling method based on edge computing according to claim 4, characterized in that: For each cluster, the process of filtering nodes with acceptable latency and constructing a candidate node set based on schedulable resources includes: For each cluster Select nodes with acceptable latency, as follows: ； in, Cluster The corresponding set of acceptable nodes; Will Internal elements are categorized according to schedulable resources Sort from highest to lowest, as follows: ； in: Indicates schedulable resources Sort by height from highest to lowest , denoted as cluster The set of candidate nodes; This indicates a descending order operation.

Images