Distributed request scheduling method and system based on high concurrency map service

Abstract

The application discloses a distributed request scheduling method and system based on a high-concurrency map service, belongs to the technical field of distributed request scheduling, and comprises the following steps: associated group aggregation marking, double-weight linkage adaptive intelligent regulation and control, associated request group fixed-point batch and batch aggregation processing. The application realizes real-time collection and analysis of client map service requests, multi-dimensional feature recognition and marking, and associated request group aggregation marking of batch requests in the same field of view, geographical area and zoom level. The application realizes real-time acquisition of the task processing margin of each service node, adaptive intelligent regulation and control of the margin, and determination of the execution heat linkage scheduling weight adaptive regulation and control. According to the regulation and control results, the application implements fixed-point batch binding scheduling on the associated request group, uniformly distributes the same request group to a single node for centralized processing. The application realizes efficient distributed scheduling by completing batch aggregation calculation on the associated request group received by each service node, integrating the processing results and uniformly returning the processing results to the client.

Description

Technical Field

[0001] This invention relates to the field of distributed request scheduling technology, and in particular to a distributed request scheduling method and system based on high-concurrency map services. Background Technology

[0002] Current distributed map service scheduling mostly follows the general Internet load balancing scheduling logic, relying on common indicators such as node number, hardware load, and number of connections to distribute requests. However, it does not adapt to the inherent characteristics of map service spatial hotspot clustering, tile hierarchy association, and regional request clustering.

[0003] Round-robin scheduling distributes all requests evenly, failing to distinguish between hot and ordinary requests. This can lead to a concentrated influx of hot requests onto some nodes while the rest remain idle for extended periods, resulting in severe load balancing within the cluster. Conventional weighted scheduling, on the other hand, configures weights based solely on static hardware parameters such as CPU (Central Processing Unit), memory, and bandwidth. It cannot dynamically detect changes in geographical request activity and struggles to adapt to sudden regional traffic spikes, causing hot nodes to overload, request queues to time out, service disruptions, and extremely low overall cluster resource utilization. Meanwhile, existing map service clusters are mostly deployed heterogeneously, with significant differences in computing power, read / write performance, tile rendering, and dynamic computing capabilities among nodes. Furthermore, the resource consumption of static tile requests differs considerably from that of dynamic requests such as path planning and spatial retrieval. However, existing scheduling mechanisms only collect macroscopic hardware load indicators and cannot identify the specific geographic processing computing power, cache capacity, and dynamic task availability of each node. They also cannot distinguish the resource consumption weights of different requests, easily leading to computing resource mismatch issues. Moreover, they cannot identify hidden load bottlenecks such as disk I / O (Input / Output) congestion, retrieval latency, and cache back-to-origin pressure, resulting in node processing blockage, large fluctuations in cluster response latency, and poor service stability. In addition, a single view operation by a map user triggers a batch of related requests. Current technologies use a single-request independent scheduling mode, randomly distributing related requests to multiple nodes, causing problems such as repeated loading of geographic data into the cache, a surge in I / O and query overhead, and a decrease in cache hit rate. Furthermore, the blocking of a single request can cause overall map rendering to lag. Existing scheduling strategies mostly rely on manually preset static rules, which cannot detect dynamic traffic fluctuations and hotspot migration trends in real time. Sudden traffic surges can easily lead to queue congestion or even service collapse, while low traffic periods result in idle computing power. Furthermore, the current scheduling system uses a unified first-come, first-served queuing mechanism, failing to differentiate between high-priority requests such as real-time navigation and emergency scheduling and low-priority requests such as historical queries and offline downloads. In high-concurrency scenarios, core business resources are squeezed out, latency and failure rates increase, making it difficult to meet the high-reliability, low-latency scheduling requirements of commercial and government / enterprise map services. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a distributed request scheduling method and system based on high-concurrency map services. This method enables real-time collection and parsing of client map service requests, multi-dimensional feature identification and annotation, and aggregation and labeling of batch requests with the same viewpoint, geographical region, and zoom level into related request groups. It also acquires the task processing capacity of each service node in real time, performs intelligent adjustment of adaptive computing power weight based on the capacity, and determines and performs adaptive adjustment of scheduling weight based on execution popularity. Based on the adjustment results, it implements fixed-point batch binding scheduling for related request groups, uniformly distributing the same request group to a single node for centralized processing. Finally, each service node completes batch aggregation calculations on the received related request groups, integrates the processing results, and returns them uniformly to the client, achieving efficient distributed scheduling.

[0005] This invention provides a distributed request scheduling method based on high-concurrency map services. The method includes: Step 1, collecting and parsing map service requests initiated by clients in real time, performing multi-dimensional feature recognition and classification labeling on the map service requests, and aggregating and marking batches of continuous requests under the same map view, the same geographical area, and the same zoom level into related request groups based on the results of multi-dimensional feature recognition and classification labeling; Step 2, obtaining the task processing capacity of each service node in the distributed map service cluster in real time, and determining whether to perform adaptive intelligent adjustment of the capacity computing power weight based on the task processing capacity. If so, determining whether to perform adaptive adjustment of the heat-linked scheduling weight after intelligent adjustment; if not, directly determining whether to perform adaptive adjustment of the heat-linked scheduling weight; Step 3, after determining whether to perform adaptive adjustment of the heat-linked scheduling weight, performing fixed-point batch binding scheduling of related request groups, and uniformly scheduling the same related request group to a single service node for centralized processing during the distributed distribution of requests; Step 4, receiving and processing related request groups, with each service node in the distributed map service cluster performing batch aggregation calculations on the scheduled related request groups to complete request processing, and merging and returning the results to the client.

[0006] This application also provides a distributed request scheduling system based on high-concurrency map services. This system is applied to a distributed request scheduling method based on high-concurrency map services. The system includes: an association group aggregation and marking module, a dual-weight linkage adaptive intelligent control module, an association request group fixed-point batch binding and distribution scheduling module, and an association request group batch aggregation processing module. The association group aggregation and marking module is used to collect and parse map service requests initiated by clients in real time, perform multi-dimensional feature recognition and classification annotation on the map service requests, and perform association request group aggregation and marking on batches of continuous requests under the same map view, the same geographical area, and the same zoom level based on the results of multi-dimensional feature recognition and classification annotation. The dual-weight linkage adaptive intelligent control module is used to obtain real-time information from each service node within the distributed map service cluster. The system considers the task processing margin and determines whether to perform adaptive computing power weight adjustment based on the margin. If so, it determines whether to perform adaptive adjustment of the scheduling weight based on popularity after intelligent adjustment; otherwise, it directly determines whether to perform adaptive adjustment of the scheduling weight based on popularity. The associated request group fixed-point batch binding and distribution scheduling module is used to determine whether to perform fixed-point batch binding scheduling of associated request groups after adaptive adjustment of the scheduling weight based on popularity. During the distributed distribution of requests, the same associated request group is uniformly scheduled to a single service node for centralized processing. The associated request group batch aggregation processing module is used to receive and process associated request groups. Each service node in the distributed map service cluster performs batch aggregation calculations on the scheduled associated request groups, completes request processing, and merges and returns the results to the client.

[0007] Beneficial effects The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This invention constructs an intelligent adjustment mechanism for adaptive computing power weight based on the remaining task processing capacity of nodes. Combined with a processing capacity-computing power weight mapping model, it achieves dynamic and precise matching of computing power among heterogeneous nodes, completely overcoming the technical shortcomings of traditional fixed-weight, static load scheduling, which cannot adapt to the heterogeneous computing power differences in map service clusters. This invention can dynamically increase the scheduling weight of nodes with high remaining capacity and decrease the scheduling weight of nodes with low remaining capacity based on the real-time remaining task processing capacity of each service node. This avoids the computing power mismatch problem of idle high-computing-power nodes and overloaded low-computing-power nodes, achieving a comprehensive balance between explicit and implicit loads in the cluster. This improves the overall resource utilization and computing power efficiency of the distributed map service cluster, effectively solving the problems of node load differentiation, local congestion, and limited overall throughput in high-concurrency scenarios, and enhancing the overall operational stability of the cluster.

[0008] 2. This invention introduces a geographic grid heat perception and high-concurrency request-scheduling weight mapping model to achieve adaptive dynamic adjustment of heat-linked scheduling weights. This specifically addresses the shortcomings of traditional scheduling strategies that ignore the spatial aggregation characteristics of map requests and cannot adapt to instantaneous peak traffic in hotspot areas. This invention can identify dynamic heat changes in geographic regions in real time, dynamically increasing scheduling weights for high-concurrency hotspot areas to prioritize hotspot traffic, and dynamically converging idle weight resources for ordinary steady-state areas. This achieves precise matching of hotspot traffic, dynamic traffic, and node service capabilities, effectively avoiding the risks of node overload, service timeouts, and service avalanches caused by a surge in hotspot requests. Simultaneously, it avoids redundant waste of scheduling resources in steady-state scenarios, improving the adaptive adaptability and overall resilience of high-concurrency map services to sudden traffic surges and hotspot migration scenarios.

[0009] 3. This invention completely solves the technical defects of traditional single-request independent random scheduling, such as scattered associated requests, duplicate data loading, rendering lag, and uncontrolled concurrency, by a mechanism of associated request group aggregation marking, fixed-point batch binding scheduling, and adaptive concurrency control strategy based on the maximum single-point latency. This invention unifies and distributes batch associated requests from the same viewpoint and region to a single node for centralized processing, greatly reducing the redundant overhead of repeated queries, rendering, and caching across multiple nodes, and significantly improving cluster cache hit rate and request processing efficiency. Simultaneously, it dynamically and adaptively adjusts the concurrency threshold within the group based on the maximum single-point latency, achieving fine-grained concurrency limiting, avoiding request backlog and blocking, ensuring the integrity and synchronization of batch map rendering requests, reducing front-end map rendering lag and loading delays, and comprehensively improving user experience and the real-time response of map services in high-concurrency scenarios. Attached Figure Description

[0010] Figure 1 A flowchart of a distributed request scheduling method based on a high-concurrency map service provided in this application embodiment; Figure 2 A flowchart illustrating the adaptive adjustment of hot-spot-linked scheduling weights in a distributed request scheduling method based on high-concurrency map services, provided in this application embodiment. Figure 3 This is a schematic diagram of the structure of a distributed request scheduling method system based on high-concurrency map services provided in an embodiment of this application. Detailed Implementation

[0011] like Figure 1 The diagram shown is a flowchart of a distributed request scheduling method based on a high-concurrency map service provided in this application embodiment. The method includes the following steps: Step 1: Collect and parse map service requests initiated by the client in real time, perform multi-dimensional feature recognition and classification labeling on the map service requests, and aggregate and label batches of continuous requests under the same map view, the same geographical area, and the same zoom level based on the results of multi-dimensional feature recognition and classification labeling.

[0012] It should be understood that batch aggregation computation specifically includes: uniformly loading tile data and spatial indexes of the geographic area covered by the associated request group into memory, with all requests within the group sharing the same data copy; merging spatial retrieval requests for the same spatial range within the group into a single retrieval, and distributing the results to each sub-request; merging rendering requests for adjacent tiles within the group into batch rendering tasks to reduce GPU context switching overhead; and performing layer fusion and data stitching on the server side for the returned results of each sub-request within the group, returning them to the client in a single response packet.

[0013] It should be noted that the multi-dimensional features include the request's geospatial location, the request's business type, the request's resource consumption level, the request's business priority, and the request's related relationships. Based on the geospatial location of the request, the access popularity of the geographic area is statistically analyzed to identify hot area requests and ordinary area requests. Based on the business type of the request, static tile loading requests and dynamic spatial calculation requests are distinguished. Based on the business scenario attributes, requests are divided into high-priority core business requests and low-priority ordinary business requests.

[0014] In this embodiment, the system connects to the client's full map interaction request data stream, captures and parses the message parameters and business fields of each map service request in real time, and constructs a five-dimensional fusion feature quantification and identification system covering the request's geospatial location, request business type, request resource consumption level, request business priority, and request association. A multi-feature weighted scoring algorithm is used to achieve refined classification and labeling of requests. The algorithm assigns weights to each dimension: request business priority weight 0.30, request resource consumption level weight 0.25, geographic area access popularity weight 0.25, and request association weight 0.20. The scores of each dimension are weighted and summed after Min-Max normalization to obtain a comprehensive classification score S∈[0,1]. S≥0.7 is classified as a high-priority group, 0.3≤S<0.7 is classified as a normal group, and S<0.3 is classified as a low-priority group.The specific implementation logic is as follows: First, based on the latitude and longitude coordinates and map grid encoding carried in the request, all requests are mapped to a preset uniform geographic grid unit. By statistically analyzing the request frequency, request density, and access duration of each grid unit within a fixed sliding time window, the standardized heat score of each geographic grid is calculated. Based on a preset heat threshold, the grid area is divided into high-heat hotspot areas and low-heat ordinary areas, thereby accurately identifying the corresponding hotspot area requests and ordinary area requests. Second, the request business type is determined by the request interface fields, business instruction identifiers, and data return types. Requests involving only static tile image reading, tile preloading, and static layer display are classified as low-resource-consumption static tile loading requests. Complex calculation requests involving large-scale spatial retrieval, multi-point path planning, real-time feature rendering, and batch coordinate correction are classified as high-resource-consumption dynamic spatial calculation requests. The resource consumption level of the corresponding requests is determined by combining the average CPU usage, disk input / output overhead, and processing latency parameters of the two types of requests. Third, a map business priority classification system is pre-constructed, combining the real-time requirements of the business and the importance of the service. The algorithm classifies requests for high-priority core business scenarios, such as real-time navigation monitoring, emergency geographic dispatch, vehicle trajectory tracking, and real-time government visualization, as high-priority core business requests. Requests for low-priority ordinary business scenarios, such as historical geographic data queries, offline map resource downloads, non-real-time spatial statistical analysis, and static layer playback, are classified as low-priority ordinary business requests. Differentiated scheduling tags and resource guarantee weights are assigned to requests of different priorities. Finally, based on map view parameters, geographic grid encoding, tile zoom levels, request triggering sequence, and user operation context, a request association matching algorithm is constructed. This algorithm performs association matching on batch progressive map requests triggered consecutively within a short period under the same client, map view range, geographic grid area, and zoom level. Requests are automatically aggregated into structured association request groups and assigned unique tags. The request association matching algorithm determines whether requests belong to the same association group based on the following rules: requests originate from the same session ID; the distance between the center points of the requested map view is less than 20% of the diagonal length of the view; the zoom levels are the same; and the triggering time interval is less than 500ms. Continuous requests that simultaneously meet the above four conditions are grouped into the same associated request group. This achieves structured, hierarchical, and refined feature classification of discrete map requests, providing accurate, quantifiable, and matchable pre-defined feature data support for subsequent dual-weight adaptive scheduling, targeted batch distribution of associated requests, and dynamic adjustment of concurrency thresholds. It completely solves the technical problems of traditional scheduling, such as treating all requests indiscriminately and failing to adapt to the differentiated characteristics of map requests, resulting in coarse scheduling, resource mismatch, and hotspot congestion.

[0015] Step 2: Obtain the task processing capacity of each service node in the distributed map service cluster in real time. Based on the task processing capacity, determine whether to perform intelligent adjustment of the capacity adaptive computing power weight. If yes, determine whether to perform adaptive adjustment of the heat-linked scheduling weight after intelligent adjustment. If no, directly determine whether to perform adaptive adjustment of the heat-linked scheduling weight.

[0016] It needs to be explained that the specific process for determining whether to perform intelligent adjustment of residual adaptive computing power weight is as follows: The task processing capacity of each service node is collected based on a preset time period, and the dynamic task processing capacity is divided into high capacity range, medium capacity range, and low capacity range according to the preset processing capacity threshold range. The weighted computing power score currently available for each service node is calculated by weighting the node's remaining CPU, memory, disk I / O bandwidth, and the resource consumption level of the currently queued requests. The schedulable computing power weight is used to characterize the priority and traffic allocation ratio of map service requests assigned to each service node in the corresponding distributed map service cluster. A pre-constructed processing margin-computing power weight mapping model is used to represent the correspondence between the dynamic task processing margin value and the computing power weight adjustment factor. If the task processing capacity is within the preset processing capacity threshold range, it is determined to be in the medium capacity range, and the current schedulable computing power weight remains unchanged. The preset processing capacity threshold range refers to the open interval formed by the lower bound of the preset processing capacity threshold and the upper bound of the preset processing capacity threshold.

[0017] Determining whether to perform intelligent adjustment of residual adaptive computing power weight also includes: If the task processing margin is less than or equal to the lower bound of the preset processing margin threshold, it is determined to be a low margin range, and the schedulable computing power weight of the node is reduced. If the task processing capacity is greater than or equal to the upper limit of the preset processing capacity threshold, it is determined to be a high capacity range, and the schedulable computing power weight of the node is increased. The specific process for downgrading is as follows: If it is determined to be a low margin range, the current task processing margin is input into the processing margin-computing power weight mapping model, and the computing power weight down adjustment factor is output. The computing power weight base value and the computing power weight down adjustment factor are multiplied to obtain the target schedulable computing power weight. The specific process for adjusting the price upwards is as follows: If it is determined to be a high margin range, the current task processing margin is input into the processing margin-computing power weight mapping model, and the computing power weight adjustment factor is output. The computing power weight base value and the computing power weight adjustment factor are multiplied to obtain the target schedulable computing power weight.

[0018] It should be understood that the processing margin-computing power weight mapping model outputs a unified computing power weight adjustment factor α; when in the low margin range, α∈(0,1) is used as a downward adjustment factor; when in the high margin range, α∈(1, α_max] is used as an upward adjustment factor. The computing power weight baseline value is the initial equal weight of each node 1 / N (N is the number of cluster nodes); after each adjustment, the target schedulable computing power weight of all nodes is subjected to L1 normalization to ensure ΣW_i=1.

[0019] In this embodiment, the system is pre-configured with a fixed status collection period, periodically traversing all heterogeneous service nodes within the distributed map service cluster, and collecting the dynamic task processing capacity of each node in real time. Specifically, task processing capacity is defined as the number of unused tasks available on each service node that can handle high-consumption map requests such as new dynamic spatial calculations and real-time feature rendering. This accurately quantifies the instantaneous remaining computing power resources of a node. The schedulable computing power weight is used to quantify the request distribution priority and traffic allocation ratio of each service node in the cluster; the larger the weight value, the more map service requests are allocated to the node. The calculation of task processing capacity collects multi-dimensional implicit load indicators of the nodes, including: disk I / O wait time, average spatial index retrieval latency, tile cache hit rate, and origin server frequency. After normalization, these indicators are weighted and fused with explicit indicators such as CPU / memory remaining to obtain a task processing capacity score that comprehensively reflects the explicit and implicit load status of the nodes. This invention constructs a piecewise nonlinear adaptive mapping algorithm to build a processing margin-computing power weight mapping model. This model can establish a precise mapping relationship between the continuously changing dynamic task processing margin value and the differentiated computing power weight adjustment factor. The model is implemented using a piecewise nonlinear adaptive mapping algorithm. Compared with the traditional fixed coefficient weight adjustment method, it can achieve refined margin value and nonlinear weight correction, which can significantly improve the accuracy of heterogeneous cluster computing power adaptation.In practice, the system pre-calibrates and stores globally unified lower and upper bounds of the processing margin threshold. These two thresholds together form a preset processing margin threshold range with an open interval structure, used to determine the three-level interval division of the real-time collected node task processing margin. When the node's real-time task processing margin falls within the preset open interval threshold range, the node is determined to be in the medium margin interval, indicating that the node's current task load is moderate and the remaining computing power resources are in a steady state. No weight adjustment is needed, and the node's current schedulable computing power weight is maintained constant. When the node's real-time task processing margin is less than or equal to the preset lower bound of the processing margin threshold, the node is determined to be in the low margin interval, indicating that the node has a limited number of remaining processable tasks, the load is close to saturation, and there is a risk of overload and congestion. The weight reduction mechanism is then triggered, and the accurate task processing margin value collected in real time is input into the processing margin-computing power weight mapping model. Through the model's built-in piecewise nonlinear adaptive... The algorithm dynamically matches and outputs a corresponding gradient-based reduction factor for computing power weight. The reduction factor decreases with decreasing margin value. Then, the system's preset computing power weight benchmark value is multiplied by the computing power weight reduction factor to accurately calculate the target schedulable computing power weight that matches the current load state of the node. This effectively reduces the request distribution ratio of high-load nodes and avoids continuous node overload. When the real-time task processing margin of a node is greater than or equal to the upper limit of the preset processing margin threshold, the node is determined to be in a high margin range, indicating that the node has sufficient remaining computing power resources and has been in a low-load idle state for a long time. The weight adjustment mechanism is then triggered. The current task processing margin is input into the mapping model, and the algorithm outputs a computing power weight adjustment factor that matches the current margin level. The adjustment factor increases with increasing margin value. The updated target schedulable computing power weight is obtained by multiplying the computing power weight benchmark value and the adjustment factor, which actively improves the request acceptance priority and traffic allocation ratio of high-margin, high-quality nodes. This embodiment achieves dynamic, refined, and non-linear intelligent control of the schedulable computing power weight of each node in the cluster based on real-time computing power surplus by periodically collecting node surplus, intelligently determining intervals, calculating adaptive coefficients of the model, and iteratively updating weights. This completely solves the industry pain points of traditional static weight scheduling, such as the inability to adapt to dynamic load changes of nodes, low matching degree of heterogeneous computing power, and the coexistence of idle cluster resources and local overload. It significantly improves the load balancing accuracy and overall resource utilization of the distributed map service cluster.

[0020] It should be understood that Figure 2The specific process of the distributed request scheduling method based on high-concurrency map services provided in this application embodiment is as follows: First, a high-concurrency request-scheduling weight mapping model is pre-constructed. Then, real-time access heat values ​​of each region are collected at a preset period, and the heat values ​​are compared with preset heat thresholds. If the heat value is greater than the threshold, it is determined to be a hot area with high-concurrency requests. The heat data is input into the model to output an upward adjustment coefficient. The final regional heat scheduling weight is obtained by multiplying the base weight and the upward adjustment coefficient, thereby increasing the request reception and allocation ratio of the corresponding node. If the heat value is not greater than the threshold, it is determined to be a normal steady-state area. The heat data is input into the model to output a downward adjustment coefficient. The final regional heat scheduling weight is obtained by multiplying the base weight and the downward adjustment coefficient, thereby converging idle scheduling weight resources. Finally, the regional heat scheduling weights of each node are iteratively updated, and the next round of heat collection and determination is entered, realizing the dynamic adaptive adjustment of scheduling weights with regional heat.

[0021] It should be further explained that the specific process for determining whether to perform adaptive adjustment of the scheduling weights based on popularity is as follows: Pre-define the entire geographic grid area. High-concurrency requests in hotspot areas are used to represent the massive aggregated requests for map tile access, spatial retrieval, and path planning within a unit of time in target geographic areas where the geographic spatial heat is higher than a preset heat threshold. Regional heat scheduling weight is used to characterize the request distribution priority and traffic allocation ratio of service nodes in different distributed map service clusters for each hot geographical area; A high-concurrency request-scheduling weight mapping model is pre-built. This model is used to characterize the one-to-one mapping relationship between the real-time access popularity of a geographic region and the upward and downward adjustment coefficients of the region's popularity scheduling weight. The real-time access popularity value of the target geographic area is compared with a preset popularity threshold for determination. Specifically: If the real-time access popularity value of the target geographic area is greater than the preset popularity threshold, the geographic area is determined to be a hotspot area, and there are high-concurrency requests in the hotspot area. The corresponding node is then increased in its receiving and allocation ratio for high-concurrency requests in the hotspot area. If the real-time access popularity value of the target geographic area is less than or equal to the preset popularity threshold, the geographic area is determined to be a normal steady-state area, and idle scheduling weight resources are converged.

[0022] The specific process for improving the receiving and allocation ratio of high-concurrency hot requests on the corresponding node is as follows: If there are high-concurrency requests in hot areas, the current high-concurrency requests in hot areas are input into the high-concurrency request-scheduling weight mapping model, and the scheduling weight adjustment coefficient is output. The regional hot scheduling baseline weight and the scheduling weight adjustment coefficient are multiplied to obtain the final regional hot scheduling weight. The specific process for converging idle scheduling weight resources is as follows: If it is a normal steady-state region, the high-concurrency requests of the current hot region are input into the high-concurrency request-scheduling weight mapping model, the scheduling weight reduction coefficient is output, and the regional hot scheduling base weight and the scheduling weight reduction coefficient are multiplied to obtain the final regional hot scheduling weight. Based on a preset period, the real-time request popularity of each target geographic area is collected, and the regional popularity scheduling weights corresponding to each service node are iteratively updated to achieve dynamic adaptive adjustment of the regional popularity scheduling weights according to the request popularity of the geographic area, so that high-concurrency hot point requests are prioritized and scheduled to the appropriate nodes.

[0023] In this embodiment, a uniform grid partitioning method is first used to finely divide the entire geographic space, forming a uniformly sized and fully covered geographic grid area. This enables fine-grained grid-based statistics of the geographic dimensions of map requests. Among them, high-concurrency requests in hotspot areas specifically refer to the massive amount of aggregated map service requests such as static tile access, spatial retrieval, path planning, and real-time feature rendering that accumulate within a unit of time in the target area where the real-time spatial access popularity of the geographic grid area is higher than the system's preset popularity threshold. These requests are characterized by large instantaneous traffic, high request density, and complex business types. The regional popularity scheduling weight is used to quantitatively characterize the request distribution priority and traffic allocation ratio of each service node in the cluster for different geographic hotspot areas. The weight value directly determines the resource quota and scheduling priority of the node in accepting requests for the corresponding region. This invention innovatively introduces a heat gradient normalization mapping algorithm and pre-constructs a high-concurrency request-scheduling weight mapping model. This model uses a heat gradient normalization mapping algorithm to establish a precise one-to-one quantitative mapping relationship between the continuous value of real-time access heat in a geographical area and the adjustment coefficients of the regional heat scheduling weight. It breaks the traditional coarse mode of fixed coefficient weight adjustment and can realize gradient and fine weight adaptive correction according to the heat level, which greatly improves the accuracy and adaptability of hot spot traffic scheduling. The system continuously monitors the request frequency of each geographic grid region in real time according to a preset fixed collection cycle. Combined with sliding time window smoothing filtering, it calculates a standardized real-time access heat value for each target geographic region and compares this value with a preset global heat threshold. When the real-time access heat value of a target geographic region exceeds the preset threshold, the region is identified as a hotspot with a large number of high-concurrency requests. The hotspot weight gain adjustment logic is immediately activated. The real-time heat characteristic parameters of the current region are input into the high-concurrency request-scheduling weight mapping model. Based on the heat gradient normalization mapping algorithm, a scheduling weight adjustment coefficient adapted to the heat level is matched and output. The system's preset regional heat scheduling benchmark weight is multiplied by this adjustment coefficient to accurately generate a scheduling weight that adapts to the current hotspot traffic pressure. The final regional heat scheduling weight effectively improves the acceptance authority and traffic allocation ratio of corresponding service nodes for high-concurrency hot requests, ensuring that hot traffic is preferentially diverted to the appropriate nodes and avoiding the problem of hot traffic congestion. When the real-time access heat value of the target geographical area is less than or equal to the preset heat threshold, the geographical area is determined to be a normal steady-state area with no high-concurrency traffic impact and stable overall request load. Then, the idle weight convergence and control logic is started. Similarly, the current regional heat status parameters are input into the high-concurrency request-scheduling weight mapping model. The algorithm calculates and outputs the scheduling weight reduction coefficient matching the steady-state scenario. The updated regional heat scheduling weight is obtained by multiplying the regional heat scheduling base weight and the reduction coefficient. The redundant scheduling weight of the steady-state area is actively reduced, idle scheduling resources are reclaimed, and ineffective resource occupation is avoided.This embodiment periodically and continuously collects real-time request heat data from the entire geographic grid, and iteratively updates the regional heat scheduling weights corresponding to each service node. This enables dynamic adaptive adjustment of regional heat scheduling weights as the request heat of geographic regions fluctuates and hotspot regions migrate. It achieves bidirectional intelligent regulation of hotspot traffic gain adaptation and steady-state traffic resource convergence, effectively solving the technical defects of traditional static scheduling strategies that cannot adapt to the spatiotemporal dynamic changes of map geographic heat, hotspot traffic load imbalance, and low resource utilization. It significantly improves the adaptive carrying capacity and overall load balancing performance of high-concurrency map service clusters for sudden hotspot traffic and dynamic regional traffic migration.

[0024] Step 3: After determining whether to perform adaptive adjustment of the weight of the heat-linked scheduling, perform fixed-point batch binding scheduling of the related request group. During the process of completing the distributed distribution of requests, the same related request group is uniformly scheduled to a single service node for centralized processing.

[0025] It needs to be explained that the specific process for performing fixed-point batch binding scheduling of associated request groups is as follows: Based on a preset period, the single-point processing time of all requests in each associated request group within the corresponding service node is collected, and the maximum single-point processing time of the request within the group is extracted. The maximum single-point processing time of the request within the group refers to the longest time taken for a single map service request to be distributed and processed within the same associated request group. The maximum single-point latency of requests within a group is compared with the pre-set latency warning threshold. Based on the comparison results, the concurrency threshold of requests within a group is dynamically adjusted. The concurrency threshold of requests within a group is used to characterize the maximum number of requests that a single service node is allowed to process in parallel at the same time for the same related request group. A maximum single-point latency-concurrency threshold mapping model is pre-constructed. This model is used to characterize the different intervals in which the maximum single-point latency of requests within a group is located, and outputs the corresponding concurrency threshold adjustment coefficients.

[0026] The specific process for dynamically adjusting the concurrency threshold within a group is as follows: If the maximum single-point time of a request within a group is within the pre-set time warning critical range, it is determined to be in the high time range. Then, the preset first reduction coefficient is retrieved. The pre-set time warning critical range means the open interval formed by the pre-set lower limit of the time warning critical range and the pre-set upper limit of the time warning critical range. If the maximum single-point latency of a request within a group is less than or equal to the pre-set lower bound of the latency warning threshold, it is determined to be within the normal latency range, and the current concurrency threshold of the group requests remains unchanged. If the maximum single-point time of a request within the group is greater than or equal to the upper limit of the preset time warning threshold, it is determined to be in the over-limit time interval, and the preset second reduction coefficient is retrieved. The preset first reduction factor is greater than the preset second reduction factor.

[0027] The specific steps to retrieve the preset first reduction factor are as follows: If it is determined to be in a high-consumption range, the maximum single-point consumption time of the current group request is input into the maximum single-point consumption time-concurrency threshold mapping model, and the preset first reduction coefficient is output. The preset first reduction coefficient is multiplied by the group request concurrency benchmark threshold to obtain the target group request concurrency threshold. The specific steps to retrieve the preset second reduction factor are as follows: If it is determined that it is in the excessive time consumption range, the maximum single-point time consumption of the current group request is input into the maximum single-point time consumption-concurrency threshold mapping model, and the preset second reduction coefficient is output. The preset second reduction coefficient is multiplied by the group request concurrency benchmark threshold to obtain the target group request concurrency threshold.

[0028] In this embodiment, the system performs real-time polling monitoring of the processing status of associated request groups of each service node in the distributed map service cluster according to a preset monitoring cycle. It continuously collects the end-to-end single-point processing time of each map service request distributed to the node within the same associated request group, from task distribution, data loading, calculation and processing to result return. It also extracts the maximum single-point processing time of the current associated request group by traversing statistics. This parameter is specifically defined as the maximum value of the processing time of all single requests within the same associated request group. It can accurately reflect the bottleneck time and extreme blocking state of the current node in processing the associated requests of this group, and can effectively characterize the overall processing congestion of the associated request group. At the same time, the intra-group request concurrency threshold configured in this invention is used to limit the maximum number of requests that a single service node can process in parallel for the same associated request group. It is a core control parameter for managing the granularity of batch associated request concurrency, avoiding instantaneous request accumulation, and balancing processing efficiency and system stability. This invention introduces a time-consuming gradient-based rate limiting algorithm, which pre-constructs a maximum single-point time-concurrency threshold mapping model. This model adopts a time-consuming gradient-based rate limiting algorithm, which can adaptively match and output the corresponding gradient concurrency threshold adjustment coefficient according to the different time intervals in which the maximum single-point time of requests within a group is located. This achieves dynamic adaptation of the concurrency rate limiting intensity to the degree of congestion in processing time. Compared with the traditional fixed concurrency threshold control method, it can accurately match the bottleneck characteristics of batch processing of map-related requests and effectively solve the technical defects of a single concurrency threshold being unable to adapt to dynamic congestion scenarios.In practice, the system pre-configures a lower and upper threshold for time consumption warnings. These two thresholds form an open-interval time consumption warning threshold range, used to implement a three-level interval classification for request processing time. When the maximum single-point time of a request within a group is less than or equal to the lower threshold, the current associated request group is determined to be in the normal time consumption range, indicating smooth node processing, no blocking or backlog issues, and a healthy and stable overall load. No adjustment to the concurrency granularity is needed; the concurrency threshold for requests within the current group is maintained constant. When the maximum single-point time of a request within a group falls within the open interval formed by the lower and upper thresholds, the associated request group is determined to be in a high time consumption range, indicating decreased node processing efficiency and a slight blocking trend. Mild rate limiting is required. In this case, the real-time collected maximum single-point time of requests within the group is input into the maximum single-point time - concurrency threshold mapping module. The algorithm calculates a first reduction coefficient using a time-consuming gradient-based rate limiting algorithm. This first reduction coefficient is then multiplied by a system-preset threshold for intra-group request concurrency to obtain a target intra-group request concurrency threshold suitable for the current slightly congested state. This achieves a slight reduction in the number of concurrent requests and alleviates the processing pressure on nodes. When the maximum single-point time of intra-group requests is greater than or equal to the upper bound of the time-consuming warning threshold, the associated request group is determined to be in the over-limit time-consuming range, indicating severe node processing blockage and a very high risk of request backlog. In this case, high-intensity rate limiting convergence is required. Similarly, the maximum single-point time is input into the mapping model, and the algorithm adaptively outputs a second reduction coefficient with a smaller value. The first reduction coefficient is strictly guaranteed to be greater than the second reduction coefficient. The second reduction coefficient is then multiplied by the concurrency benchmark threshold to obtain a significantly converged target intra-group request concurrency threshold, which strongly reduces the number of nodes processing in parallel and completely prevents the problem of continuous deterioration of congestion. This embodiment achieves refined, hierarchical, and adaptive concurrency limiting control based on the actual processing bottleneck of requests by periodically monitoring the bottleneck time of associated request groups in real time, classifying congestion status, outputting reduction coefficients based on the gradient of the mapping model, and dynamically iteratively updating the concurrency threshold. It can ensure the batch processing efficiency of associated request groups in steady-state scenarios, and gradually converge the concurrency granularity, eliminate task backlog and rendering stuttering in blocking scenarios. It completely solves the dual problems of low steady-state processing efficiency and avalanche blocking in congested scenarios that are easy to occur in traditional fixed concurrency mechanisms, and greatly improves the stability of batch processing of associated request groups and the smoothness of front-end map interactive rendering.

[0029] Step 4: Receive and process association request groups. Each service node in the distributed map service cluster performs batch aggregation calculations on the scheduled association request groups, completes request processing, and merges and returns the results to the client.

[0030] It should be understood that after weight adaptive adjustment and fixed-point batch binding scheduling, the corresponding target service nodes within the distributed map service cluster actively receive the complete set of association request data. They then perform integrated batch aggregation calculations on batch association requests grouped together for the same map view, the same geographic region, and the same zoom level. This avoids the inefficient mode of traditional independent processing and fragmented responses for single requests. During processing, nodes uniformly load tile resources, spatial index data, and layer parameters for the corresponding geographic region. They perform batch merging calculations, unified resource scheduling, and centralized data parsing for various sub-requests within the group, including static tile loading requests, spatial retrieval requests, path planning requests, and feature rendering requests. This effectively avoids redundant consumption of single requests, such as repeatedly loading disk data, repeatedly creating spatial indexes, and repeatedly occupying system resources. Improve the overall processing efficiency of batch requests; after completing the calculation, rendering, data matching and verification of all sub-requests within the associated request group, the service node performs structured merging, data splicing, layer fusion and format unification of the scattered processing results of each group, integrates the scattered single request response data into a complete, continuous and regular geographic data result set adapted to the client's map rendering display, and returns the finally merged complete map service data to the client, so that the client can complete the overall loading and visualization rendering of the map layer. In this way, the associated request group realizes the closed-loop processing of the entire process from batch scheduling, aggregation processing, result integration to client response, effectively solves the problems of fragmented response, asynchronous rendering and repeated loading lag caused by traditional decentralized scheduling processing, and significantly improves the integrity of map loading and the smoothness of user interaction.

[0031] like Figure 3The diagram shows the structure of a distributed request scheduling method system based on high-concurrency map services provided in this application embodiment. It includes: an association group aggregation and marking module, a dual-weight linkage adaptive intelligent control module, an association request group fixed-point batch binding and distribution scheduling module, and an association request group batch aggregation processing module. The association group aggregation and marking module is used to collect and parse map service requests initiated by clients in real time, perform multi-dimensional feature recognition and classification annotation on the map service requests, and perform association request group aggregation and marking on batch continuous requests under the same map view, the same geographical area, and the same zoom level based on the results of multi-dimensional feature recognition and classification annotation. The dual-weight linkage adaptive intelligent control module is used to obtain the task processing capacity of each service node in the distributed map service cluster in real time. The system determines whether to perform adaptive computing power weight adjustment based on the task processing margin. If so, it determines whether to perform adaptive adjustment of the scheduling weight based on popularity after intelligent adjustment. If not, it directly determines whether to perform adaptive adjustment of the scheduling weight based on popularity. The associated request group fixed-point batch binding and distribution scheduling module is used to determine whether to perform fixed-point batch binding scheduling of associated request groups after adaptive adjustment of the scheduling weight based on popularity. During the distributed distribution of requests, the same associated request group is uniformly scheduled to a single service node for centralized processing. The associated request group batch aggregation processing module is used to receive and process associated request groups. Each service node in the distributed map service cluster performs batch aggregation calculations on the scheduled associated request groups, completes request processing, and merges and returns the results to the client.

[0032] In this embodiment, the distributed map service request intelligent scheduling system of the present invention consists of four core modules working together to complete the overall scheduling work: an association group aggregation and marking module, a dual-weight linkage adaptive intelligent control module, an association request group fixed-point batch binding and distribution scheduling module, and an association request group batch aggregation processing module. In an 8-node heterogeneous cluster environment, simulating a mixed map request scenario of 10,000 concurrent users (60% tile requests + 30% spatial retrieval + 10% path planning), the test results show that the present invention reduces the average response latency of the cluster by 42% compared to the polling scheduling cluster, improves the cache hit rate by 35 percentage points compared to the weighted least connection scheduling, reduces the 99th percentile latency by 58%, and reduces the hot area request timeout rate from 12.7% to 1.3%.During operation, the association group aggregation and tagging module first continuously collects various map service requests initiated by clients across the entire network in real time. It then comprehensively analyzes and disassembles the original request messages, completing multi-dimensional feature identification and refined classification and labeling based on request geospatial location, service type, resource consumption level, service priority, and relationships. This accurately filters out batches of continuous map requests that meet the conditions of the same map view, the same geographical area, and the same map zoom level. These strongly correlated discrete requests are then uniformly collected and tagged to form structured association request groups. This achieves structured preprocessing of disordered and scattered requests, laying the data foundation for subsequent precise scheduling. After completing request aggregation and tagging... The dual-weighted adaptive intelligent control module collects the dynamic task processing capacity of all service nodes in the distributed map service cluster in real time. Based on the remaining task capacity of each node, it determines whether to trigger adaptive intelligent control of computing power weight. If control is required, it first adaptively updates the schedulable computing power weight of each node based on the node task capacity interval classification rules and a mapping model. Then, it further determines and executes adaptive control of the heat-linked scheduling weight. If no computing power weight control is required, it directly enters the adaptive determination and update process of the heat-linked scheduling weight. Through dual-weighted control of computing power capacity and geographical heat, the module achieves the scheduling priority and traffic allocation ratio of each node in the cluster. The system dynamically optimizes and adapts the data. After completing dual-weight adaptive adjustment and determining the optimal node scheduling weight ratio, the associated request group fixed-point batch binding and distribution scheduling module takes over the scheduling work. Relying on the updated node scheduling weights, it completes differentiated traffic distribution, abandoning the traditional single-request random distributed scheduling mode. It binds the entire associated request group to a single adapted service node for centralized scheduling, fundamentally avoiding problems such as data duplication, cache invalidation, and asynchronous rendering stuttering caused by the distributed processing of associated requests across nodes. Finally, the associated request group batch aggregation processing module executes the backend business processing work. The corresponding target service node receives the complete associated request group issued by the scheduler and processes it accordingly. The system performs unified resource loading, batch aggregation operations, and integrated business processing for batch related requests. It centrally completes the processing of various sub-requests such as static tile loading, spatial retrieval, path planning, and feature rendering. After all requests within the group have been processed, the fragmented processing results are fused, formatted, and structured and encapsulated. Finally, the integrated complete map visualization data is returned to the client. This achieves fully automated and intelligent closed-loop scheduling from request access, feature aggregation, intelligent weight control, precise point distribution to batch processing and unified result response. It comprehensively improves the load balancing capability, resource utilization efficiency, and user interaction experience of distributed map services in high-concurrency scenarios.

[0033] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope and intent of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A distributed request scheduling method based on high-concurrency map services, characterized in that, Includes the following steps: Step 1: Collect and parse map service requests initiated by the client in real time, perform multi-dimensional feature recognition and classification labeling on the map service requests, and aggregate and label batch continuous requests under the same map view, the same geographical area, and the same zoom level based on the results of multi-dimensional feature recognition and classification labeling. Step 2: Obtain the task processing capacity of each service node in the distributed map service cluster in real time. Based on the task processing capacity, determine whether to perform intelligent adjustment of the capacity adaptive computing power weight. If yes, determine whether to perform adaptive adjustment of the heat linkage scheduling weight after intelligent adjustment. If no, directly determine whether to perform adaptive adjustment of the heat linkage scheduling weight. Step 3: After determining whether to perform adaptive adjustment of the weight of the heat-linked scheduling, perform fixed-point batch binding scheduling of the related request group. During the process of completing the distributed distribution of requests, the same related request group is uniformly scheduled to a single service node for centralized processing. Step 4: Receive and process association request groups. Each service node in the distributed map service cluster performs batch aggregation calculations on the scheduled association request groups, completes request processing, and merges and returns the results to the client.

2. The distributed request scheduling method based on high-concurrency map services as described in claim 1, characterized in that: The multi-dimensional features include the request geospatial location, request service type, request resource consumption level, request service priority, and request association relationship; Based on the geospatial location of the request, the access popularity of the geographic area is statistically analyzed to identify hot area requests and ordinary area requests. Based on the business type of the request, static tile loading requests and dynamic spatial calculation requests are distinguished. Based on the business scenario attributes, requests are divided into high-priority core business requests and low-priority ordinary business requests.

3. The distributed request scheduling method based on high-concurrency map services as described in claim 1, characterized in that: The specific process for determining whether to perform intelligent adjustment of residual adaptive computing power weight is as follows: The task processing capacity of each service node is collected based on a preset time period, and the dynamic task processing capacity is divided into high capacity range, medium capacity range, and low capacity range according to the preset processing capacity threshold range. The task processing margin is used to characterize the weighted computing power score currently available to each service node. This score is calculated by weighting the node's remaining CPU, memory, disk I / O bandwidth, and the resource consumption level of the currently queued requests. The schedulable computing power weight is used to characterize the priority and traffic allocation ratio of the map service requests assigned to each service node in the corresponding distributed map service cluster. A pre-constructed processing margin-computing power weight mapping model is used to represent the correspondence between the dynamic task processing margin value and the computing power weight adjustment factor. If the task processing capacity is within the preset processing capacity threshold range, it is determined to be in the medium capacity range, and the current schedulable computing power weight remains unchanged. The preset processing capacity threshold range refers to the open interval formed by the lower bound of the preset processing capacity threshold and the upper bound of the preset processing capacity threshold.

4. The distributed request scheduling method based on high-concurrency map services as described in claim 3, characterized in that: The determination of whether to perform intelligent adjustment of residual adaptive computing power weight also includes: If the task processing margin is less than or equal to the lower bound of the preset processing margin threshold, it is determined to be a low margin range, and the schedulable computing power weight of the node is reduced. If the task processing capacity is greater than or equal to the upper limit of the preset processing capacity threshold, it is determined to be a high capacity range, and the schedulable computing power weight of the node is increased. The specific process for performing the downgrade is as follows: If it is determined to be a low margin range, the current task processing margin is input into the processing margin-computing power weight mapping model, and the computing power weight down adjustment factor is output. The computing power weight base value and the computing power weight down adjustment factor are multiplied to obtain the target schedulable computing power weight. The specific process for the upward adjustment is as follows: If it is determined to be a high margin range, the current task processing margin is input into the processing margin-computing power weight mapping model, and the computing power weight adjustment factor is output. The computing power weight base value and the computing power weight adjustment factor are multiplied to obtain the target schedulable computing power weight.

5. The distributed request scheduling method based on high-concurrency map services as described in claim 1, characterized in that: The specific process for determining whether to perform adaptive weight adjustment in heat-related scheduling is as follows: A pre-defined global geographic grid area is defined. The high-concurrency requests in the hotspot area are used to represent the massive aggregate requests for map tile access, spatial retrieval, and path planning within a unit of time in a target geographic area where the geographic spatial heat is higher than a preset heat threshold. The regional heat scheduling weight is used to characterize the request distribution priority and traffic allocation ratio of each hot geographical area corresponding to the service node in different distributed map service clusters. A high-concurrency request-scheduling weight mapping model is pre-constructed. The high-concurrency request-scheduling weight mapping model is used to characterize the one-to-one mapping relationship between the real-time access popularity of a geographic region and the upward and downward adjustment coefficients of the regional popularity scheduling weight. The real-time access popularity value of the target geographic area is compared with a preset popularity threshold for determination. Specifically: If the real-time access popularity value of the target geographic area is greater than the preset popularity threshold, the geographic area is determined to be a hotspot area, and there are high-concurrency requests in the hotspot area. The corresponding node is then increased in its receiving and allocation ratio for high-concurrency requests in the hotspot area. If the real-time access popularity value of the target geographic area is less than or equal to the preset popularity threshold, the geographic area is determined to be a normal steady-state area, and idle scheduling weight resources are converged.

6. The distributed request scheduling method based on high-concurrency map services as described in claim 5, characterized in that: The specific process for improving the receiving and allocation ratio of high-concurrency hotspot requests by the corresponding node is as follows: If there are high-concurrency requests in hot areas, the current high-concurrency requests in hot areas are input into the high-concurrency request-scheduling weight mapping model, and the scheduling weight adjustment coefficient is output. The regional hot scheduling baseline weight and the scheduling weight adjustment coefficient are multiplied to obtain the final regional hot scheduling weight. The specific process for converging idle scheduling weight resources is as follows: If it is a normal steady-state region, the high-concurrency requests of the current hot region are input into the high-concurrency request-scheduling weight mapping model, the scheduling weight reduction coefficient is output, and the regional hot scheduling base weight and the scheduling weight reduction coefficient are multiplied to obtain the final regional hot scheduling weight. Based on a preset period, the real-time request popularity of each target geographic area is collected, and the regional popularity scheduling weights corresponding to each service node are iteratively updated to achieve dynamic adaptive adjustment of the regional popularity scheduling weights according to the request popularity of the geographic area, so that high-concurrency hot point requests are prioritized and scheduled to the appropriate nodes.

7. The distributed request scheduling method based on high-concurrency map services as described in claim 1, characterized in that: The specific process for performing fixed-point batch binding scheduling of associated request groups is as follows: Based on a preset period, the single-point processing time of all requests in each associated request group within the corresponding service node is collected, and the maximum single-point processing time of the request within the group is extracted. The maximum single-point processing time of the request within the group refers to the longest time taken for a single map service request to be distributed and processed within the same associated request group. The maximum single-point time of requests within a group is compared with the pre-set time warning threshold. Based on the comparison results, the concurrency threshold of requests within a group is dynamically adjusted. The concurrency threshold of requests within a group is used to characterize the maximum number of requests that a single service node is allowed to process in parallel at the same time for the same associated request group. A maximum single-point latency-concurrency threshold mapping model is pre-constructed. This model is used to characterize the different intervals in which the maximum single-point latency of requests within a group is located, and outputs the corresponding concurrency threshold adjustment coefficients respectively.

8. The distributed request scheduling method based on high-concurrency map services as described in claim 7, characterized in that: The specific process for dynamically adjusting the request concurrency threshold within a group is as follows: If the maximum single-point time of a request within a group is within the preset time warning critical interval, it is determined to be in the high time interval, and then the preset first reduction coefficient is retrieved. The preset time warning critical interval represents the open interval formed by the preset lower limit of the time warning critical interval and the preset upper limit of the time warning critical interval. If the maximum single-point latency of a request within a group is less than or equal to the pre-set lower bound of the latency warning threshold, it is determined to be in the normal latency range, and the current concurrency threshold of the group requests remains unchanged. If the maximum single-point time of a request within the group is greater than or equal to the upper limit of the preset time warning threshold, it is determined to be in the over-limit time interval, and the preset second reduction coefficient is retrieved. The preset first reduction factor is greater than the preset second reduction factor.

9. The distributed request scheduling method based on high-concurrency map services as described in claim 8, characterized in that: The specific steps for retrieving the preset first reduction coefficient are as follows: If it is determined to be in a high-consumption range, the maximum single-point consumption time of the current group request is input into the maximum single-point consumption time-concurrency threshold mapping model, and the preset first reduction coefficient is output. The preset first reduction coefficient is multiplied by the group request concurrency benchmark threshold to obtain the target group request concurrency threshold. The specific steps for retrieving the preset second reduction coefficient are as follows: If it is determined that it is in the excessive time consumption range, the maximum single-point time consumption of the current group request is input into the maximum single-point time consumption-concurrency threshold mapping model, and the preset second reduction coefficient is output. The preset second reduction coefficient is multiplied by the group request concurrency benchmark threshold to obtain the target group request concurrency threshold.

10. A system applying the distributed request scheduling method based on high-concurrency map services as described in any one of claims 1-9, characterized in that, include: The module includes an associated group aggregation marking module, a dual-weight linkage adaptive intelligent control module, an associated request group fixed-point batch binding distribution scheduling module, and an associated request group batch aggregation processing module. The associated group aggregation tagging module is used to collect and parse map service requests initiated by the client in real time, perform multi-dimensional feature recognition and classification annotation on the map service requests, and perform associated request group aggregation tagging on batch continuous requests under the same map view, the same geographical area, and the same zoom level based on the results of multi-dimensional feature recognition and classification annotation. The dual-weight linkage adaptive intelligent control module is used to obtain the task processing surplus of each service node in the distributed map service cluster in real time, and determine whether to perform surplus adaptive computing power weight intelligent control based on the task processing surplus. If yes, it determines whether to perform heat linkage scheduling weight adaptive control after intelligent control. If no, it directly determines whether to perform heat linkage scheduling weight adaptive control. The associated request group fixed-point batch binding and distribution scheduling module is used to determine whether to perform hot-spot linkage scheduling weight adaptive adjustment, and then perform associated request group fixed-point batch binding scheduling. In the process of completing the distributed distribution of requests, the same associated request group is uniformly scheduled to a single service node for centralized processing. The associated request group batch aggregation processing module is used to receive and process associated request groups. Each service node in the distributed map service cluster performs batch aggregation calculations on the scheduled associated request groups, completes request processing, and merges and returns the results to the client.

Images