A dynamic scaling system and method for an order dispatch service of a business opportunity platform

Abstract

The application discloses a dynamic expansion and contraction system and method for order dispatching service of a business opportunity platform, and the system comprises a monitoring module, a mode detection module, an expansion and contraction management module, a scheduling optimization module and a task scheduling module.The monitoring module collects node performance data at intervals to generate running state information.The mode detection module establishes a performance baseline model based on historical performance data and an LSDD algorithm, detects deviation of real-time data, and generates an expansion and contraction signal.The expansion and contraction management module executes expansion and contraction operations according to the expansion and contraction signal and resource priority.The scheduling optimization module dynamically adjusts resource allocation strategies in combination with an order flow prediction model and node state.The task scheduling module allocates tasks and dynamically adjusts task loads according to the resource allocation strategies.The application improves system resource utilization and task response speed through multidimensional data monitoring, intelligent expansion and contraction triggering mechanism and dynamic scheduling strategy, is suitable for high concurrency and large flow scenarios, and has a wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of computer system resource management technology, and in particular to a dynamic scaling system and method for order dispatching services on a business opportunity platform. Background Technology

[0002] With the rapid development of e-commerce and online services, order dispatch services, as a crucial foundational functional module, have become increasingly complex in their demands for resource scheduling and task allocation. Order processing on e-commerce platforms often needs to handle high concurrency, large traffic volumes, and dynamically changing order flows, which places higher demands on the system's resource management and scheduling strategies.

[0003] Existing dynamic scaling technologies typically rely on static rules or simple threshold comparisons to dynamically adjust resources, which mainly suffers from the following problems: Traditional systems often trigger scaling based on a single metric (such as CPU utilization), failing to comprehensively consider the dynamic changes in multi-dimensional performance data (such as network throughput, memory utilization, order traffic, etc.), easily leading to decision-making biases. During resource adjustment, existing technologies often ignore the impact of heterogeneous hardware performance differences on resource utilization efficiency, resulting in wasted resources on high-performance nodes or overload on low-performance nodes. Existing task scheduling schemes do not adequately differentiate the needs of different types of tasks (such as low-latency tasks and high-throughput tasks), and resource allocation strategies lack specificity, making it difficult to fully utilize system performance.

[0004] To address the aforementioned issues, there is an urgent need for a dynamic scaling system that can comprehensively perceive the system's operating status, accurately trigger scaling decisions, and optimize task scheduling by combining heterogeneous hardware characteristics. Summary of the Invention

[0005] The purpose of this invention is to provide a dynamic scaling system and method for order dispatch services on a business opportunity platform, which solves the limitations of existing technologies such as a single scaling trigger mechanism, low utilization of heterogeneous hardware resources, and inflexible task scheduling strategies. By real-time monitoring of multi-dimensional performance data, constructing a node performance baseline model based on the least squares density difference (LSDD) algorithm, and dynamically optimizing resource allocation and task scheduling strategies, the invention achieves accurate and efficient resource allocation, significantly improves the overall response speed and resource utilization efficiency of the system, and adapts to the complex needs of high-concurrency order processing scenarios.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] A dynamic scaling system for order dispatching services on a business opportunity platform, the system comprising:

[0008] The monitoring module is used to collect performance data of service nodes through a dynamically adjusted time interval via the system interface. The performance data includes CPU utilization, memory utilization, network throughput, and order traffic data, and generates node running status information.

[0009] The pattern detection module is used to establish a node performance baseline model based on historical performance data and the least squares density difference (LSDD) algorithm. It compares the real-time monitored node performance data with the performance baseline model to calculate the degree of deviation. When the deviation value exceeds the scaling threshold, a pattern drift signal is generated to trigger the scaling management module to adjust the resource allocation scheme.

[0010] The heterogeneous hardware perception module is used to collect hardware performance indicators of service nodes, including computing power, memory capacity, network bandwidth and current resource utilization. The performance indicators are normalized through dynamic standardization method and adaptive weight model to generate a resource efficiency priority list.

[0011] The scaling up / down management module is used to perform dynamic scaling up or down operations based on the mode drift signal generated by the mode detection module and the resource efficiency priority list generated by the heterogeneous hardware perception module, combined with the real-time node response time and task load level.

[0012] The scheduling optimization module is used to dynamically adjust the resource allocation strategy by combining the order quantity change trend analyzed by the order flow prediction model with the current resource pool status.

[0013] The task scheduling module is used to allocate order tasks based on the resource allocation strategy generated by the scheduling optimization module and the resource efficiency priority list generated by the heterogeneous hardware awareness module, and to dynamically adjust the task allocation scheme when the node load exceeds the limit.

[0014] As a preferred embodiment of the present invention, the monitoring module collects performance data through the following steps:

[0015] The data collection time interval is dynamically adjusted, and the time interval is calculated based on the frequency of order traffic fluctuations. The formula for the interval T is:

[0016]

[0017] Where: f 流量波动 The collection interval is automatically shortened when the order traffic fluctuation exceeds the set threshold.

[0018] Collect data on CPU utilization, memory utilization, network throughput, and order traffic of the service nodes;

[0019] The collected data is generated into a time series and stored for subsequent node performance analysis.

[0020] As a preferred embodiment of the present invention, the mode detection module generates the expansion / shrinkage trigger signal through the following steps:

[0021] Collect historical performance data of the nodes, including throughput, response time, resource utilization, and order volume;

[0022] The deviation D(p,q) between the real-time performance data distribution and the historical baseline model distribution is calculated using the least squares density difference algorithm (LSDD).

[0023] D(p,q)=∫(p(x)-q(x)) 2 dx

[0024] Where: p(x) is the distribution density function of real-time performance data; q(x) is the distribution density function of historical baseline model;

[0025] When the deviation exceeds the set threshold, an expansion or contraction trigger signal is generated.

[0026] As a preferred embodiment of the present invention, the heterogeneous hardware awareness module generates a resource efficiency priority list through the following steps:

[0027] The hardware performance indicators of the acquisition node include CPU frequency, number of available threads, memory capacity, network bandwidth, and current resource utilization.

[0028] The performance metrics are normalized using a dynamic standardization method, and the calculation formula is as follows:

[0029]

[0030] Where x is the sampled value; min(x) and max(x) are the minimum and maximum values ​​within the current sampling window, respectively;

[0031] The overall performance score S of the node is calculated based on the normalized performance metrics and the importance weights of the task requirements. node for:

[0032] S node =w1×f C ′ PU +w2×M s ′ ize +w3×B′ max +w4×U′ available

[0033] Where: w1, w2, w3, and w4 are the dynamic weights of the task requirements; f C ′ PU M s ′ ize B′max 、U′ available These are the normalized CPU frequency, memory capacity, network bandwidth, and current resource utilization, respectively.

[0034] A resource efficiency priority list is generated based on the overall performance score, from highest to lowest.

[0035] As a preferred embodiment of the present invention, the expansion / shrinkage management module performs the expansion operation through the following steps:

[0036] The expansion trigger signal generated by the receive mode detection module;

[0037] Select the idle node with the highest score based on the resource efficiency priority list;

[0038] Start the target node and add it to the order distribution task queue.

[0039] As a preferred embodiment of the present invention, the scaling-down management module performs a scaling-down operation through the following steps:

[0040] Receive the scaling-down trigger signal generated by the receiver mode detection module;

[0041] Select the node with the lowest score as the target node based on the resource efficiency priority list;

[0042] Migrate the order tasks on the target node to other nodes and shut down the target node to free up resources.

[0043] As a preferred embodiment of the present invention, the scheduling optimization module generates a resource allocation strategy through the following steps:

[0044] Generate order flow trends for future time periods based on the order flow prediction model;

[0045] Calculate resource allocation priority weights based on node load status:

[0046]

[0047] Wherein: F order (t) represents the predicted order flow; L node The node load level is calculated by weighting CPU utilization and task queue length.

[0048] Resource allocation strategies are generated based on resource allocation weights.

[0049] In a preferred embodiment of the present invention, the task scheduling module adjusts the allocation of order tasks through the following steps:

[0050] Based on the resource allocation strategy generated by the scheduling optimization module, low-latency tasks are assigned to the nodes with the shortest response times;

[0051] Assign high-throughput tasks to the nodes with the highest throughput.

[0052] When a node's load exceeds its limit, a task migration operation is triggered, and the task is assigned to the node with the lowest load.

[0053] A scaling method for a dynamic scaling system of an order dispatch service on a business opportunity platform, the method comprising the following steps:

[0054] S1: Collect performance data of service nodes, including CPU utilization, memory utilization, network throughput and order traffic data, dynamically adjust the collection time interval, generate node running status information from the performance data and store it in time series format;

[0055] S2: Utilize historical performance data and node operating status information of nodes to establish a performance baseline model. Calculate the deviation between the distribution of real-time node performance data and the distribution of the baseline model based on the least squares density difference algorithm. When the deviation exceeds the expansion threshold, an expansion trigger signal is generated; when the deviation is below the reduction threshold, a reduction trigger signal is generated.

[0056] S3: Collect hardware performance metrics of service nodes, including CPU frequency, number of available threads, memory capacity and network bandwidth, perform dynamic standardization on performance metrics to generate normalized performance data, calculate comprehensive performance scores based on task requirement weights, and sort nodes by scores to generate a resource efficiency priority list.

[0057] S4: Based on the expansion trigger signal, select the idle node with the highest overall performance score from the resource efficiency priority list, start it and add it to the order task distribution queue; based on the shrinkage trigger signal, select the node with the lowest overall performance score from the resource efficiency priority list, migrate its unfinished tasks to other nodes, shut it down and release resources;

[0058] S5: Generate order traffic change trends based on the order traffic prediction model, and generate resource allocation strategies based on the current resource pool status, prioritizing the allocation of resources to the nodes with the lowest load;

[0059] S6: Based on the resource allocation strategy, low-latency tasks are assigned to the nodes with the shortest response times, and high-throughput tasks are assigned to the nodes with the highest throughput. When the task load on a certain node exceeds the limit, a task reallocation operation is triggered, migrating some tasks to other low-load nodes.

[0060] As a preferred embodiment of the present invention, the method further includes the following steps:

[0061] Based on the historical performance data of service nodes, including throughput, response time, resource utilization and order traffic, a performance baseline model is established, and the performance baseline model is dynamically updated using a sliding window or incremental update algorithm.

[0062] The deviation between the distribution of real-time performance data of a node and the distribution of the baseline model is calculated using a performance baseline model and a least squares density difference algorithm. When the CPU utilization of a node exceeds 90% for three consecutive collection cycles and the deviation exceeds the expansion threshold, an expansion trigger signal is generated. When the resource utilization of a node is below 30% for three consecutive collection cycles and the deviation is below the shrinkage threshold and there are no pending tasks, a shrinkage trigger signal is generated.

[0063] The hardware performance indicators of service nodes are normalized, and the indicator weights are dynamically adjusted according to task requirements. The overall performance score of the nodes is calculated, and the nodes are sorted by score to generate a resource efficiency priority list. In the expansion scenario, the node with the highest overall performance score and in an idle state is selected first. In the shrinking scenario, the node with the lowest overall performance score is selected first. If there are multiple nodes with the same score, the node with the lightest task load is selected first to be shut down.

[0064] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention dynamically adjusts the time interval through a monitoring module to collect multi-dimensional performance data of service nodes in real time (including CPU utilization, memory utilization, network throughput, and order traffic). Combined with the least squares density difference (LSDD) algorithm in the pattern detection module, it constructs a node performance baseline model, which can accurately detect abnormal deviations in node operating status, thereby triggering timely scaling operations and reducing resource waste and performance insufficiency. The heterogeneous hardware awareness module dynamically standardizes the hardware performance indicators of service nodes (such as computing power, memory capacity, network bandwidth, and resource utilization), and generates a resource efficiency priority list based on the weight of task requirements. This fully considers the performance differences of heterogeneous hardware, maximizing the utilization of high-performance nodes' resources while avoiding overload operation of low-performance nodes. The scheduling optimization module combines an order traffic prediction model to analyze the trend of order quantity changes and dynamically adjusts resource allocation strategies based on real-time resource pool status. The task scheduling module further matches optimal resources based on task type (such as low-latency tasks or high-throughput tasks) and triggers task migration operations when node load exceeds limits, achieving flexible and efficient allocation of order tasks, thereby improving the overall response speed and service quality of the system. The dynamic scaling system of this invention can adapt to the dynamic changes in order traffic. Through the synergistic effect between modules, it not only achieves efficient resource scheduling but also ensures stable system operation under high concurrency and high traffic scenarios, significantly improving the user experience of the business opportunity platform. This invention comprehensively improves the resource utilization efficiency and system stability of the business opportunity platform's order dispatch service through multi-dimensional data monitoring, intelligent pattern detection, heterogeneous hardware resource optimization, and dynamic task scheduling, solving problems such as inaccurate scaling operations, low resource utilization, and inflexible task scheduling in existing technologies. It has significant practical application value and promotional significance. Attached Figure Description

[0065] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] in:

[0067] Figure 1 This is a schematic diagram of the modular structure of the system of the present invention;

[0068] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0070] like Figure 1 As shown, this is an embodiment of the present invention, which provides a dynamic scaling system for order dispatching services on a business opportunity platform, comprising:

[0071] (1) Monitoring module

[0072] The system uses a time interval that is dynamically adjusted via the system interface to collect performance data of service nodes. The performance data includes CPU utilization, memory utilization, network throughput, and order traffic data, and generates node running status information.

[0073] The monitoring module collects performance data through the following steps:

[0074] The data collection time interval is dynamically adjusted, and the time interval is calculated based on the frequency of order traffic fluctuations. The formula for the interval T is:

[0075]

[0076] Where: f 流量波动 The collection interval is automatically shortened when the order traffic fluctuation exceeds the set threshold.

[0077] Collect data on CPU utilization, memory utilization, network throughput, and order traffic of the service nodes;

[0078] The collected data is generated into a time series and stored for subsequent node performance analysis.

[0079] In one specific embodiment, the monitoring module connects to the platform's service node interface to collect various performance data in real time, including CPU utilization, memory utilization, network throughput, and order traffic data. In a specific operational scenario, the monitoring module automatically shortens the collection interval to the second level based on dynamic changes in order traffic to quickly respond to traffic peaks. After data collection is complete, the performance data is encapsulated in a time-series format and transmitted to the pattern detection module, while simultaneously being stored in a distributed database for subsequent analysis.

[0080] (2) Pattern detection module

[0081] This is used to establish a node performance baseline model based on historical performance data and combined with the least squares density difference (LSDD) algorithm. The real-time monitored node performance data is compared with the performance baseline model to calculate the degree of deviation. When the deviation value exceeds the scaling threshold, a mode drift signal is generated to trigger the scaling management module to adjust the resource allocation scheme.

[0082] The pattern detection module generates the scaling trigger signal through the following steps:

[0083] Collect historical performance data of the nodes, including throughput, response time, resource utilization, and order volume;

[0084] The deviation D(p,q) between the real-time performance data distribution and the historical baseline model distribution is calculated using the least squares density difference algorithm (LSDD).

[0085] D(p,q)=∫(p(x)-q(x)) 2 dx

[0086] Where: p(x) is the distribution density function of real-time performance data; q(x) is the distribution density function of historical baseline model;

[0087] When the deviation exceeds the set threshold, an expansion or contraction trigger signal is generated.

[0088] In one specific embodiment, the pattern detection module preprocesses historical performance data and establishes a performance baseline model using the Least Squares Density Difference (LSDD) algorithm. During actual operation, when order traffic suddenly increases, the pattern detection module detects that the distribution of real-time performance data deviates from the baseline model, with the deviation exceeding a preset scaling threshold. The module immediately generates a pattern drift signal and sends the anomaly detection result to the scaling management module, providing a basis for scaling decisions. For example, when the CPU utilization of a node continuously exceeds 90% and the response time significantly increases during system operation, the module can accurately trigger a scaling signal.

[0089] In this embodiment, the deviation threshold is set according to the following principles: the expansion threshold is used to identify peak resource usage, such as CPU utilization consistently exceeding 90% or network throughput approaching bandwidth limits; the reduction threshold is used to identify low resource usage, such as resource utilization below 30% and no pending tasks. This threshold can be dynamically adjusted based on historical performance data and business needs.

[0090] (3) Heterogeneous hardware sensing module

[0091] The hardware performance metrics of the service nodes are collected. These metrics include computing power, memory capacity, network bandwidth, and current resource utilization. The performance metrics are normalized using a dynamic standardization method and an adaptive weight model to generate a resource efficiency priority list.

[0092] The heterogeneous hardware awareness module generates a resource efficiency priority list through the following steps:

[0093] The hardware performance indicators of the acquisition node include CPU frequency, number of available threads, memory capacity, network bandwidth, and current resource utilization.

[0094] The performance metrics are normalized using a dynamic standardization method, and the calculation formula is as follows:

[0095]

[0096] Where x is the sampled value; min(x) and max(x) are the minimum and maximum values ​​within the current sampling window, respectively;

[0097] The overall performance score S of the node is calculated based on the normalized performance metrics and the importance weights of the task requirements. node for:

[0098] S node =w1×f C ′ PU +w2×M s ′ ize +w3×B′ max +w4×U′ available

[0099] Where: w1, w2, w3, and w4 are the dynamic weights of the task requirements; f C ′ PU M s ′ ize B′ max 、U′ available These are the normalized CPU frequency, memory capacity, network bandwidth, and current resource utilization, respectively.

[0100] A resource efficiency priority list is generated based on the overall performance score, from highest to lowest.

[0101] In one specific embodiment, a heterogeneous hardware awareness module is deployed in the platform resource pool to monitor the hardware performance differences of each node in real time. In a typical scenario, the module collects data on the node's CPU frequency of 2.5GHz, memory capacity of 16GB, and network bandwidth of 1Gbps, and calculates the current resource utilization. To address hardware performance differences, the module normalizes each indicator and dynamically adjusts the weights based on task type (e.g., compute-intensive or storage-intensive tasks) to generate a resource efficiency priority list. When the resource pool expands, this list prioritizes nodes with high CPU performance, while when storage demand increases, it prioritizes nodes with high memory capacity.

[0102] Specifically, when multiple nodes in the resource efficiency priority list have the same score, the selection logic is as follows:

[0103] In scaling scenarios, prioritize the node with the longest idle time;

[0104] In a scaling-down scenario, prioritize the node with the lightest current task load;

[0105] If the node load is the same, the node with the shorter response time will be selected first.

[0106] (4) Scaling Management Module

[0107] Based on the pattern drift signal generated by the pattern detection module and the resource efficiency priority list generated by the heterogeneous hardware perception module, combined with the real-time node response time and task load level, dynamic scaling up or down operations are performed.

[0108] The scaling management module performs scaling operations through the following steps:

[0109] The expansion trigger signal generated by the receive mode detection module;

[0110] Select the idle node with the highest score based on the resource efficiency priority list;

[0111] Start the target node and add it to the order distribution task queue.

[0112] The scaling management module performs a scaling-down operation through the following steps:

[0113] Receive the scaling-down trigger signal generated by the receiver mode detection module;

[0114] Select the node with the lowest score as the target node based on the resource efficiency priority list;

[0115] Migrate the order tasks on the target node to other nodes and shut down the target node to free up resources.

[0116] In one specific embodiment, the scaling management module performs resource scaling operations based on the pattern drift signal generated by the pattern detection module and the resource efficiency priority list generated by the heterogeneous hardware awareness module. For example, during a peak period, order traffic surges, and the system load exceeds a critical value. The scaling management module starts new nodes and adds them to the order dispatch queue, thereby alleviating the pressure on high-load nodes. Conversely, when order traffic declines, the module selects low-load nodes, completes the migration of unprocessed tasks, and shuts down the node to release resources and reduce operating costs.

[0117] In this embodiment: No pending tasks means that the current order task queue of the service node is empty and no new tasks to be assigned are detected within a predictable time window (such as the next 5 minutes).

[0118] (5) Scheduling optimization module

[0119] This is used to combine the order quantity change trend and the current resource pool status analyzed by the order flow prediction model to dynamically adjust the resource allocation strategy;

[0120] The scheduling optimization module generates resource allocation strategies through the following steps:

[0121] Generate order flow trends for future time periods based on the order flow prediction model;

[0122] Calculate resource allocation priority weights based on node load status:

[0123]

[0124] Wherein: F order (t) represents the predicted order flow; L node The node load level is calculated by weighting CPU utilization and task queue length.

[0125] Resource allocation strategies are generated based on resource allocation weights.

[0126] In one specific embodiment, the scheduling optimization module analyzes traffic trends over the next 5 minutes using an order traffic prediction model. For example, in actual use, if the system detects that order traffic will increase by 20% per minute, the scheduling optimization module adjusts its resource allocation strategy based on the prediction results, starting new nodes in advance to accommodate the upcoming traffic peak. This module also dynamically adjusts the resource allocation strategy based on the current operating status of the resource pool to ensure the accuracy and efficiency of resource allocation.

[0127] (6) Task scheduling module

[0128] It is used to allocate order tasks based on the resource allocation strategy generated by the scheduling optimization module and the resource efficiency priority list generated by the heterogeneous hardware awareness module, and to dynamically adjust the task allocation scheme when the node load exceeds the limit.

[0129] The task scheduling module adjusts the allocation of order tasks through the following steps:

[0130] Based on the resource allocation strategy generated by the scheduling optimization module, low-latency tasks are assigned to the nodes with the shortest response times;

[0131] Assign high-throughput tasks to the nodes with the highest throughput.

[0132] When a node's load exceeds its limit, a task migration operation is triggered, and the task is assigned to the node with the lowest load.

[0133] The definition criteria for node overload are as follows:

[0134] 1. CPU utilization exceeded 90% for three consecutive data acquisition cycles;

[0135] 2. The task queue length exceeds the node's maximum task processing capacity;

[0136] 3. The node's response time exceeds the preset maximum latency threshold (e.g., 500ms).

[0137] In one specific embodiment, the task scheduling module allocates order tasks to the most suitable nodes based on the resource allocation strategy generated by the scheduling optimization module. For example, for orders requiring rapid response, the module prioritizes the node with the shortest response time, while high-throughput tasks are prioritized for allocation to the node with the highest throughput. On a high-load node, when the task queue length reaches a set overload threshold, the module triggers a task migration operation, reallocating some tasks to low-load nodes, thereby improving the system's task processing capacity and stability.

[0138] In summary, the collaborative operation of each module in this embodiment enables dynamic scaling of the order dispatch service on the business opportunity platform, significantly improving system resource utilization and task scheduling efficiency, while ensuring system stability under high concurrency and high traffic scenarios.

[0139] like Figure 2 As shown, another embodiment of the present invention provides a scaling method for a dynamic scaling system of an order dispatch service on a business opportunity platform. The method includes the following steps:

[0140] S1: Collect performance data of service nodes, including CPU utilization, memory utilization, network throughput and order traffic data, dynamically adjust the collection time interval, generate node running status information from the performance data and store it in time series format;

[0141] S2: Utilize historical performance data and node operating status information of nodes to establish a performance baseline model. Calculate the deviation between the distribution of real-time node performance data and the distribution of the baseline model based on the least squares density difference algorithm. When the deviation exceeds the expansion threshold, an expansion trigger signal is generated; when the deviation is below the reduction threshold, a reduction trigger signal is generated.

[0142] S3: Collect hardware performance metrics of service nodes, including CPU frequency, number of available threads, memory capacity and network bandwidth, perform dynamic standardization on performance metrics to generate normalized performance data, calculate comprehensive performance scores based on task requirement weights, and sort nodes by scores to generate a resource efficiency priority list.

[0143] S4: Based on the expansion trigger signal, select the idle node with the highest overall performance score from the resource efficiency priority list, start it and add it to the order task distribution queue; based on the shrinkage trigger signal, select the node with the lowest overall performance score from the resource efficiency priority list, migrate its unfinished tasks to other nodes, shut it down and release resources;

[0144] S5: Generate order traffic change trends based on the order traffic prediction model, and generate resource allocation strategies based on the current resource pool status, prioritizing the allocation of resources to the nodes with the lowest load;

[0145] S6: Based on the resource allocation strategy, low-latency tasks are assigned to the nodes with the shortest response times, and high-throughput tasks are assigned to the nodes with the highest throughput. When the task load on a certain node exceeds the limit, a task reallocation operation is triggered, migrating some tasks to other low-load nodes.

[0146] In this embodiment, a performance baseline model is established based on the historical performance data of the service node, including throughput, response time, resource utilization and order traffic. The performance baseline model is dynamically updated using a sliding window or incremental update algorithm.

[0147] The deviation between the distribution of real-time performance data of a node and the distribution of the baseline model is calculated using a performance baseline model and a least squares density difference algorithm. When the CPU utilization of a node exceeds 90% for three consecutive collection cycles and the deviation exceeds the expansion threshold, an expansion trigger signal is generated. When the resource utilization of a node is below 30% for three consecutive collection cycles and the deviation is below the shrinkage threshold and there are no pending tasks, a shrinkage trigger signal is generated.

[0148] The hardware performance indicators of service nodes are normalized, and the indicator weights are dynamically adjusted according to task requirements. The overall performance score of the nodes is calculated, and the nodes are sorted by score to generate a resource efficiency priority list. In the expansion scenario, the node with the highest overall performance score and in an idle state is selected first. In the shrinking scenario, the node with the lowest overall performance score is selected first. If there are multiple nodes with the same score, the node with the lightest task load is selected first to be shut down.

[0149] In summary, this invention addresses the technical challenges of existing technologies, such as a single scaling / shrinking trigger mechanism, low resource allocation efficiency, and inflexible task scheduling, through real-time monitoring of multi-dimensional performance data, pattern detection based on the Least Squares Density Difference (LSDD) algorithm, dynamic optimization of heterogeneous hardware resources, and an intelligent scheduling strategy incorporating order traffic prediction. This invention innovatively combines multi-dimensional performance data with the LSDD algorithm, significantly improving the real-time performance and accuracy of scaling / shrinking operations. By using dynamic standardization and task requirement weight adjustment methods, it fully utilizes the resource performance of heterogeneous hardware, maximizing resource utilization. Furthermore, by combining an order traffic prediction model with dynamic analysis of the resource pool status, this invention further optimizes resource allocation and task scheduling, particularly in high-concurrency, high-traffic scenarios, significantly improving system stability and response speed. The technical solution of this invention is highly innovative, applicable to business opportunity platforms and other complex scenarios with dynamic resource scheduling requirements, and possesses significant practical application value and promising prospects for widespread adoption.

[0150] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0151] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0152] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic scaling system of an order dispatching service of a business opportunity platform order dispatching service, characterized in that, The system comprises: a monitoring module for collecting performance data of the service nodes at dynamically adjusted time intervals through the system interface, the performance data including CPU usage, memory usage, network throughput and order flow data, and generating node running state information; a mode detection module for establishing a node performance baseline model based on historical performance data in combination with a least square density difference (LSDD) algorithm, comparing the real-time monitored node performance data with the performance baseline model, calculating the deviation degree, and generating a mode drift signal to trigger the scale-out / in management module to adjust the resource allocation scheme when the deviation value exceeds the scale-out / in threshold; a heterogeneous hardware perception module for collecting hardware performance indicators of the service nodes, the performance indicators including computing power, memory capacity, network bandwidth and current resource utilization, normalizing the performance indicators through a dynamic standardization method and an adaptive weight model, and generating a resource efficiency priority list; a scale-out / in management module for executing dynamic scale-out or scale-in operations in combination with real-time node response time and task load level according to the mode drift signal generated by the mode detection module and the resource efficiency priority list generated by the heterogeneous hardware perception module; a scheduling optimization module for dynamically adjusting the resource allocation strategy in combination with the order quantity trend analyzed by the order flow prediction model and the current resource pool state; 2. The dynamic scaling system of the business opportunity platform order dispatching service according to claim 1, wherein, a task scheduling module for allocating order tasks according to the resource allocation strategy generated by the scheduling optimization module and the resource efficiency priority list generated by the heterogeneous hardware perception module, and dynamically adjusting the task allocation scheme when the node load exceeds the limit. The monitoring module collects performance data through the following steps: Wherein: f 流量波动 is the order flow fluctuation frequency, when the order flow fluctuation exceeds the set threshold, the collection interval is automatically shortened; the CPU usage, memory usage, network throughput and order flow data of the collection service node are collected; The time interval is dynamically adjusted, and the interval T is calculated according to the frequency of order flow fluctuation, and the interval T formula is:

3. The dynamic scaling system of the business opportunity platform order dispatching service according to claim 1, wherein, The collected data is generated into a time series and stored for subsequent node performance analysis. The mode detection module generates a scale-out / in trigger signal through the following steps: Historical performance data of the nodes are collected, including throughput, response time, resource utilization and order flow; D(p,q) = ∫(p(x) - q(x)) 2 dx The deviation degree D(p, q) between the real-time performance data distribution and the historical baseline model distribution is calculated by using the least square density difference (LSDD) algorithm as follows: Wherein: p(x) is the distribution density function of real-time performance data; q(x) is the distribution density function of the historical baseline model; 4. The dynamic scaling system of the business opportunity platform order distribution service according to claim 1, wherein, When the deviation degree exceeds the set threshold, a scale-out or scale-in trigger signal is generated. The heterogeneous hardware perception module generates a resource efficiency priority list through the following steps: Hardware performance indicators of the nodes are collected, including CPU frequency, available thread number, memory capacity, network bandwidth and current resource utilization; The performance indicators are normalized by using a dynamic standardization method, and the calculation formula is: According to the normalized performance indicators and the importance weight of the task demand, the comprehensive performance score S of the node is calculated node is: S node = w1 x f C PU + w2 x M s ize + w3 x B' max available + w4 x U'​​ wherein: w1, w2, w3, w4 are dynamic weights of task requirements; f C PU , M s ize , B′ max , U′ available are normalized CPU frequency, memory capacity, network bandwidth and current resource utilization, respectively​​ Wherein, x is the sampling value; min(x) and max(x) are the minimum and maximum values in the current sampling window; 5. The dynamic scaling system of the business opportunity platform order dispatching service according to claim 1, wherein, The resource efficiency priority list is generated from high to low according to the comprehensive performance score. The scale-out / in management module executes the scale-out operation through the following steps: Receiving the scale-out trigger signal generated by the mode detection module; Selecting the idle node with the highest score according to the resource efficiency priority list; Start the target node and add it to the order distribution task queue.

6. The dynamic scaling system of the business opportunity platform order dispatching service according to claim 1, wherein, The scale-in management module performs the scale-in operation by the following steps: The scale-in trigger signal generated by the receiving mode detection module; Select the node with the lowest score as the target node according to the resource efficiency priority list; Migrate the order tasks on the target node to other nodes and close the target node to release resources.

7. The dynamic scaling system of the business opportunity platform order dispatching service according to claim 1, wherein, The scheduling optimization module generates the resource allocation strategy by the following steps: Generate the order flow trend in the future time period according to the order flow prediction model; Calculate the resource allocation priority weight in combination with the node load state: where: F order (t) is the order flow prediction value; L node is the node load level, calculated by CPU usage and task queue length weighting; Generate the resource allocation strategy according to the resource allocation weight.

8. The dynamic scaling system of the business opportunity platform order dispatching service according to claim 1, wherein, The task scheduling module adjusts the allocation of order tasks by the following steps: According to the resource allocation strategy generated by the scheduling optimization module, allocate low-latency tasks to the node with the shortest response time; Allocate high-throughput tasks to the node with the highest throughput; When the node load exceeds the limit, trigger the task migration operation and allocate the tasks to the node with the lowest load.

9. The method of claim 1-8, wherein the method further comprises: The method comprises the following steps: S1: Collect the performance data of the service node, including CPU usage, memory usage, network throughput, and order flow data, dynamically adjust the collection time interval, generate node running state information from the performance data, and store it in time series form; S2: Use the historical performance data of the node and the node running state information to establish a performance baseline model, calculate the deviation of the real-time performance data distribution of the node from the baseline model distribution based on the least squares density difference algorithm; generate a scale-in trigger signal when the deviation value exceeds the scale-in threshold, and generate a scale-in trigger signal when the deviation value is below the scale-in threshold; S3: Collect the hardware performance indicators of the service node, including CPU frequency, available thread number, memory capacity, and network bandwidth, dynamically standardize the performance indicators, generate normalized performance data, and calculate the comprehensive performance score according to the task demand weight, sort the nodes by score to generate a resource efficiency priority list; S4: According to the scale-in trigger signal, select the idle node with the highest comprehensive performance score from the resource efficiency priority list, start it and add it to the order task distribution queue; according to the scale-in trigger signal, select the node with the lowest comprehensive performance score from the resource efficiency priority list, migrate the unfinished tasks to other nodes, then close and release resources; S5: Based on the order flow prediction model, generate the order flow trend, and generate the resource allocation strategy in combination with the current resource pool state, preferentially allocate resources to the node with the lowest load; S6: According to the resource allocation strategy, allocate low-latency tasks to the node with the shortest response time, and allocate high-throughput tasks to the node with the highest throughput; when the task load of a node exceeds the limit, trigger the task redistribution operation and migrate part of the tasks to other low-load nodes.

10. The method of claim 9, wherein the method further comprises: The method further comprises the following steps: Based on the historical performance data of the service node, including throughput, response time, resource utilization, and order flow, establish a performance baseline model, and dynamically update the performance baseline model using a sliding window or incremental update algorithm; The deviation degree of the node real-time performance data distribution and the baseline model distribution is calculated through a performance baseline model and a least square density difference algorithm, an expansion trigger signal is generated when the CPU usage of the node is more than 90% for three consecutive collection periods and the deviation value is more than the expansion threshold value; a contraction trigger signal is generated when the node resource utilization is less than 30% for three consecutive collection periods, the deviation value is less than the contraction threshold value and there is no task to be processed; The hardware performance indicators of the service nodes are normalized, the indicator weights are dynamically adjusted according to the task demand, the node comprehensive performance score is calculated, the nodes are sorted according to the score to generate a resource efficiency priority list; in the expansion scenario, the node with the highest comprehensive performance score and in the idle state is preferentially selected; in the contraction scenario, the node with the lowest comprehensive performance score is preferentially selected, and if there are multiple nodes with the same score, the node with the lightest task load is preferentially selected to be closed.

Images