Cloud environment data center operation cost dynamic accounting operation and maintenance management system and method

Abstract

The application relates to the field of resource management and discloses a cloud environment data center operation cost dynamic accounting operation and maintenance management system and method, which comprises an energy consumption cost aggregation module, a cost topology directed graph storage module and an operation and maintenance scheduling decision module, the energy consumption cost aggregation module converts real-time energy consumption data of a server node into dynamic operation cost components through a buffer queue and aggregates the dynamic operation cost components into a correction instruction; the cost topology directed graph storage module updates the correction instruction to a directed graph topology stored internally to update the weight; and the operation and maintenance scheduling decision module issues a scheduling instruction when the components exceed a threshold value, and increases a delay window to apply lag compensation when a variable variance exceeds a boundary value. The application eliminates accounting time sequence mismatch, suppresses high-frequency disturbance, guarantees the continuity of internal cost data flow conversion and optimizes resource allocation.

Description

Technical Field

[0001] This invention relates to a dynamic accounting and maintenance management system and method for cloud environment data center operation costs, belonging to the field of resource management technology. Background Technology

[0002] Currently, data centers typically collect computing resource consumption logs at fixed intervals and store historical utilization data of the central processing unit and random access memory in relational databases. After the billing cycle ends, the accounting component then uses a batch processing program to allocate the overall electricity cost and equipment depreciation cost of the data center to each business account. This method usually assumes that computing resource consumption and management costs change slowly within a cycle and that the two can be statistically analyzed separately. It has become a common processing method in resource expenditure auditing. With the application of elastic scaling in large-scale computing power clusters, computing resources may frequently expand or shrink within 10ms, and the host power consumption will fluctuate nonlinearly. At the same time, the grid electricity price will change abruptly at different times. It is difficult to keep the underlying current pulse and the upper-level cost statistics synchronized. The traditional static average expenditure coefficient cannot reflect the power mutation in a short period of time in time. Therefore, it may not be able to record the expansion of local computing power nodes during high electricity price periods, resulting in the relevant costs not being included in the accounting in time and forming an audit blind spot in the management system.

[0003] To reduce the accounting discrepancies caused by the aforementioned time asynchrony, existing solutions typically increase the log collection frequency or expand the data buffer queue in memory to receive high-frequency cost data. However, in high-density scaling scenarios, a large number of computing nodes frequently request to write to the global graph topology, which can easily lead to multi-way mutex lock contention during multi-threaded scheduling, causing blockage of internal time-series data streams and increasing processing latency. When the buffer overflows, it may also cause the loss of core accounting messages. In such large-scale computing deployments, in addition to the inherent constraints and form limitations of hardware such as computing facilities and links, the software also has shortcomings. For example, Chinese invention patent with publication number CN118363734A... The application discloses a dynamic optimization scheduling method and system for high-density services in a cloud environment, attempting to introduce a dynamic damping mechanism to achieve elastic scaling of service instances based on load prediction. However, this method has technical limitations when facing a variable cloud environment. Its underlying mechanism is based only on traditional performance indicators such as CPU utilization and response time, failing to link the underlying dynamic operating cost flow. As a result, when dealing with step fluctuations in grid electricity prices and sudden changes in regional cooling energy efficiency, it cannot perceive the marginal physical overhead differences behind resource expansion and contraction. This mismatch between the pre-existing indicators and the actual cost boundary conditions can easily lead to over-adjustment or frequent migration of ineffective services under sudden high-frequency flickering loads, making it difficult to maintain the long-term stability of resource scheduling.

[0004] Therefore, existing technologies still need to cache high-frequency updates in batches within a local data scope and adaptively adjust the control process according to the operating status at different times to reduce concurrent write lock conflicts in directed graph data structures, while ensuring that fine-grained cost changes can be continuously transmitted according to time and node relationships, and maintaining the stability of operation and maintenance scheduling. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A dynamic accounting and maintenance management system for cloud environment data center operation costs, the system comprising:

[0006] The energy consumption cost aggregation module interacts with the cost topology directed graph storage module, which in turn interacts with the operation and maintenance scheduling decision module. The energy consumption cost aggregation module includes a ring buffer storage queue, which is used to acquire real-time energy consumption data streams from server nodes and convert them into dynamic operating cost components. Through the ring buffer storage queue and a two-stage filtering rule, the dynamic operating cost components are aggregated into discrete micro-batch topology matrix correction instructions.

[0007] The cost topology directed graph storage module includes a graph data storage unit, which is used to receive discrete micro-batch topology matrix correction instructions and update the cost dynamic propagation directed graph topology in the internal storage using a single-threaded processing method to refresh the weight of the corresponding topology edge.

[0008] The operation and maintenance scheduling decision module includes a scheduling damping buffer unit, which reads the dynamic operating cost components of each server node in the cost dynamic propagation directed graph topology and issues operation and maintenance scheduling instructions when the dynamic operating cost components exceed the safety margin threshold. When the cost variation variance in a continuous sliding period exceeds the calibrated variance safety boundary, the scheduling damping buffer unit increases the scheduling activation delay window online to apply logical lag compensation for the issuance of operation and maintenance scheduling instructions.

[0009] Preferably, the energy cost aggregation module has an independent circular buffer storage queue for each server node in the memory space, so that the calculation results of dynamic operating cost components are sequentially appended to the circular buffer storage queue; the two-stage screening rule includes two-stage exclusion screening of discrete batch threshold and window time threshold, wherein the discrete batch threshold is set to 50 to 200, and the window time threshold is set to 10ms to 50ms, which is used to aggregate the continuous weight update stream into discrete micro-batch topology matrix correction instructions, and coordinate with a single write thread to update the cost dynamic propagation directed graph topology in a pulse synchronous processing mode, thereby eliminating multi-way mutex lock contention of the directed graph data structure under the load impact of high-density resource expansion and contraction.

[0010] Preferably, an asynchronous transaction processing module is provided on the flow path between the energy consumption cost aggregation module and the cost topology directed graph storage module. The asynchronous transaction processing module is used to read the asset value level label of the logical tenant to establish a mutually isolated high-priority dedicated channel and a low-priority queuing buffer, and allocate the dynamic cost accounting data packet of the high-yield core node to the high-priority dedicated channel to complete priority processing.

[0011] Preferably, the energy consumption cost aggregation module includes: the energy consumption cost aggregation module captures virtual machine container scaling events in real time, and captures the dynamic electricity price data and cooling load energy efficiency coefficient of the corresponding server node within the current accounting cycle; the energy consumption cost aggregation module integrates the dynamic electricity price data and cooling load energy efficiency coefficient into a real-time energy consumption data stream, deconstructs the marginal physical overhead of each virtual computing unit, and generates a high-frequency concurrent dynamic operating cost component.

[0012] Preferably, the cost dynamic propagation directed graph stored in the cost topology directed graph storage module includes multiple topology nodes and multiple topology edges; the topology nodes are used to represent the physical computing power cluster and the logical computing resources that establish a mapping relationship with the physical computing power cluster, and the topology edges are used to represent the network transmission loss and power supply substation loss between physical nodes. The edge weights of the topology edges are updated nonlinearly with the real-time fluctuation of the external power grid electricity price.

[0013] Preferably, the operation and maintenance scheduling decision module is equipped with a threshold arbitration unit. The threshold arbitration unit is used to read the real-time dynamic operating cost component of each topology node and perform a real-time matrix comparison with the set safety margin threshold. When the operating cost of a specific topology node exceeds the safety margin threshold, the specific topology node is locked as the source node to be migrated.

[0014] Preferably, the operation and maintenance scheduling decision module includes an operation and maintenance scheduling unit. The operation and maintenance scheduling unit is used to issue operation and maintenance scheduling instructions when the scheduling activation delay window expires, so as to migrate high-energy-consuming business loads from high-electricity-price substation areas to low-electricity-price substation areas in real time, so as to balance the resource and economic expenditure of each substation area.

[0015] Preferably, the activation threshold of the scheduling damping buffer unit is dynamically defined by the convergence state of the global directed graph topology, and the logical lag compensation output by the scheduling damping buffer unit is used to limit the instruction issuance boundary of the operation and maintenance scheduling decision module. The activation state of the scheduling damping buffer unit and the convergence state of the global directed graph topology form a bidirectional dependency.

[0016] Preferably, the system also includes an asset inventory module and an energy consumption metering module. The asset inventory module is connected to the energy consumption cost aggregation module through an internal data bus to collect data and calculate the depreciation cost of physical servers. The energy consumption metering module is connected to the energy consumption cost aggregation module through an internal data bus to collect data in real time on the ambient temperature of the computer room and the real-time wattage of computing equipment.

[0017] A method for dynamic accounting and maintenance management of cloud environment data center operating costs, implemented through a cloud environment data center dynamic accounting and maintenance management system, includes the following steps:

[0018] Step S1: Use the energy consumption cost aggregation module to obtain the real-time energy consumption data stream of the server node and convert it into dynamic operating cost components. Through the ring buffer storage queue and the two-stage filtering rules, the dynamic operating cost components are aggregated into discrete micro-batch topology matrix correction instructions.

[0019] Step S2: Receive discrete micro-batch topology matrix correction instructions using the cost topology directed graph storage module, and update the cost dynamic propagation directed graph topology stored inside the graph data storage unit using a single-threaded processing method to refresh the weights of the corresponding topology edges.

[0020] Step S3: Use the operation and maintenance scheduling decision module to read the dynamic operating cost components of each server node in the cost dynamic propagation directed graph topology. When the dynamic operating cost components exceed the safety margin threshold, issue an operation and maintenance scheduling instruction. Use the scheduling damping buffer unit to increase the scheduling activation delay window online when the cost variation variance in the continuous sliding period exceeds the calibrated variance safety boundary, and apply logical lag compensation for the issuance of the operation and maintenance scheduling instruction.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. In the dynamic accounting and maintenance management of data center operation costs in cloud environments, the energy consumption cost aggregation module incorporates the resource usage of server nodes, real-time power supply prices, and changes in cooling energy efficiency into the same accounting process. The resulting dynamic operating cost components are then updated to the cost dynamic propagation directed graph topology, enabling resource expansion and contraction and changes in operating costs to be recorded in the same time sequence. This reduces the time lag and cost omissions caused by fixed-period batch accounting, improves the timeliness and accuracy of server node operation cost accounting, and provides a more reliable basis for data center cost analysis and resource allocation.

[0023] 2. By setting up an independent circular buffer storage queue for each server node and using a two-stage filtering rule to aggregate continuously generated dynamic operating cost components into discrete micro-batch topology matrix correction instructions, and then updating the cost dynamic propagation directed graph topology by a single write thread, it is possible to avoid multi-way mutex lock contention when multiple server nodes write at the same time, reducing the risk of thread deadlock and processing delay. This processing method can also maintain the continuous flow of cost data when computing resources are frequently expanded or reduced, reducing the impact of instantaneous high-concurrency updates on the accounting process.

[0024] 3. The operation and maintenance scheduling decision module issues operation and maintenance scheduling instructions based on the dynamic operating cost components, and the scheduling damping buffer unit adjusts the scheduling activation delay window according to the cost change variance. This can apply logical lag compensation to the instruction issuance when cost fluctuations are large, so that the scheduling rhythm matches the convergence state of the directed graph topology of cost dynamic propagation. This can avoid the system from frequently migrating services due to short-term load fluctuations, reduce the repeated movement of services between different substation areas, maintain the stability of the resource scheduling process, and improve the reliability of operation and maintenance decisions under complex operating conditions. Attached Figure Description

[0025] Figure 1 This is a structural diagram of the data center cost accounting and operation and maintenance management system of the present invention;

[0026] Figure 2 This is the data center operation and maintenance scheduling state transition diagram of the present invention.

[0027] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0029] A dynamic accounting and maintenance management system for cloud environment data center operation costs, the system includes:

[0030] The energy consumption cost aggregation module interacts with the cost topology directed graph storage module, which in turn interacts with the operation and maintenance scheduling decision module. The energy consumption cost aggregation module includes a ring buffer storage queue, which is used to acquire real-time energy consumption data streams from server nodes and convert them into dynamic operating cost components. Through the ring buffer storage queue and a two-stage filtering rule, the dynamic operating cost components are aggregated into discrete micro-batch topology matrix correction instructions.

[0031] The cost topology directed graph storage module includes a graph data storage unit, which is used to receive discrete micro-batch topology matrix correction instructions and update the cost dynamic propagation directed graph topology in the internal storage using a single-threaded processing method to refresh the weight of the corresponding topology edge.

[0032] The operation and maintenance scheduling decision module includes a scheduling damping buffer unit, which reads the dynamic operating cost components of each server node in the cost dynamic propagation directed graph topology and issues operation and maintenance scheduling instructions when the dynamic operating cost components exceed the safety margin threshold. When the cost variation variance in a continuous sliding period exceeds the calibrated variance safety boundary, the scheduling damping buffer unit increases the scheduling activation delay window online to apply logical lag compensation for the issuance of operation and maintenance scheduling instructions.

[0033] Preferably, the energy cost aggregation module has an independent circular buffer storage queue for each server node in the memory space, so that the calculation results of dynamic operating cost components are sequentially appended to the circular buffer storage queue; the two-stage screening rule includes two-stage exclusion screening of discrete batch threshold and window time threshold, wherein the discrete batch threshold is set to 50 to 200, and the window time threshold is set to 10ms to 50ms, which is used to aggregate the continuous weight update stream into discrete micro-batch topology matrix correction instructions, and coordinate with a single write thread to update the cost dynamic propagation directed graph topology in a pulse synchronous processing mode, thereby eliminating multi-way mutex lock contention of the directed graph data structure under the load impact of high-density resource expansion and contraction.

[0034] Preferably, an asynchronous transaction processing module is provided on the flow path between the energy consumption cost aggregation module and the cost topology directed graph storage module. The asynchronous transaction processing module is used to read the asset value level label of the logical tenant to establish a mutually isolated high-priority dedicated channel and a low-priority queuing buffer, and allocate the dynamic cost accounting data packet of the high-yield core node to the high-priority dedicated channel to complete the exemption acceleration.

[0035] Preferably, the energy consumption cost aggregation module includes: the energy consumption cost aggregation module captures virtual machine container scaling events in real time, and captures the dynamic electricity price data and cooling load energy efficiency coefficient of the corresponding server node within the current accounting cycle; the energy consumption cost aggregation module integrates the dynamic electricity price data and cooling load energy efficiency coefficient into a real-time energy consumption data stream, deconstructs the marginal physical overhead of each virtual computing unit, and generates a high-frequency concurrent dynamic operating cost component.

[0036] Preferably, the cost dynamic propagation directed graph stored in the cost topology directed graph storage module includes multiple topology nodes and multiple topology edges; the topology nodes are used to represent the physical computing power cluster and the logical computing resources that establish a mapping relationship with the physical computing power cluster, and the topology edges are used to represent the network transmission loss and power supply substation loss between physical nodes. The edge weights of the topology edges are updated nonlinearly with the real-time fluctuation of the external power grid electricity price.

[0037] Preferably, the operation and maintenance scheduling decision module is equipped with a threshold arbitration unit. The threshold arbitration unit is used to read the real-time dynamic operating cost component of each topology node and perform a real-time matrix comparison with the set safety margin threshold. When the operating cost of a specific topology node exceeds the safety margin threshold, the specific topology node is locked as the source node to be migrated.

[0038] Preferably, the operation and maintenance scheduling decision module includes an operation and maintenance scheduling unit. The operation and maintenance scheduling unit is used to issue operation and maintenance scheduling instructions when the scheduling activation delay window expires, so as to migrate high-energy-consuming business loads from high-electricity-price substation areas to low-electricity-price substation areas in real time, so as to balance the resource and economic expenditure of each substation area.

[0039] Preferably, the activation threshold of the scheduling damping buffer unit is dynamically defined by the convergence state of the global directed graph topology, and the logical lag compensation output by the scheduling damping buffer unit is used to limit the instruction issuance boundary of the operation and maintenance scheduling decision module. The activation state of the scheduling damping buffer unit and the convergence state of the global directed graph topology form a bidirectional dependency.

[0040] Preferably, the system also includes an asset inventory module and an energy consumption metering module. The asset inventory module is connected to the energy consumption cost aggregation module through an internal data bus to collect data and calculate the depreciation cost of physical servers. The energy consumption metering module is connected to the energy consumption cost aggregation module through an internal data bus to collect data in real time on the ambient temperature of the computer room and the real-time wattage of computing equipment.

[0041] A method for dynamic accounting and maintenance management of cloud environment data center operating costs, implemented through a cloud environment data center dynamic accounting and maintenance management system, includes the following steps:

[0042] Step S1: Use the energy consumption cost aggregation module to obtain the real-time energy consumption data stream of the server node and convert it into dynamic operating cost components. Through the ring buffer storage queue and the two-stage filtering rules, the dynamic operating cost components are aggregated into discrete micro-batch topology matrix correction instructions.

[0043] Step S2: Receive discrete micro-batch topology matrix correction instructions using the cost topology directed graph storage module, and update the cost dynamic propagation directed graph topology stored inside the graph data storage unit using a single-threaded processing method to refresh the weights of the corresponding topology edges.

[0044] Step S3: Use the operation and maintenance scheduling decision module to read the dynamic operating cost components of each server node in the cost dynamic propagation directed graph topology. When the dynamic operating cost components exceed the safety margin threshold, issue an operation and maintenance scheduling instruction. Use the scheduling damping buffer unit to increase the scheduling activation delay window online when the cost variation variance in the continuous sliding period exceeds the calibrated variance safety boundary, and apply logical lag compensation for the issuance of the operation and maintenance scheduling instruction.

[0045] Example 1: A current cloud environment data center uses a distributed computing power cluster, including multiple physical computing server nodes and their respective substation areas. Each physical computing server node carries the business load of several logical tenants. The business load of a logical tenant is run by one or more virtual computing units. The virtual computing units constitute logical computing resources that establish a mapping relationship with the physical computing power cluster. The system deploys an energy consumption cost aggregation module, a cost topology directed graph storage module, and an operation and maintenance scheduling decision module. It maintains a dynamic cost propagation directed graph topology in memory space. The cost topology directed graph storage module is a non-relational database deployed on the central management server of the data center. The system consists of a graph database and cascaded graph structure computing threads, used to maintain the directed graph topology for dynamic cost propagation in a non-relational in-memory data structure. The energy consumption cost aggregation module consists of distributed lightweight resource acquisition probes deployed on each physical computing server node and a high-performance streaming data processing cluster. The lightweight resource acquisition probes convert node power signals and virtual machine container scaling events into streaming data packets via network sockets and send the streaming data packets to the streaming data processing cluster. The output of the streaming data processing cluster is connected to the graph data storage unit of the cost topology directed graph storage module through an internal data bus, thus forming a cost data flow link.

[0046] In the directed graph topology of cost dynamic propagation, topological nodes represent physical computing power clusters and logical computing resources mapped to them. A physical computing power cluster consists of one or more physical computing power server nodes, and logical computing resources correspond to virtual computing power units that carry the business load of logical tenants. Topological edges represent network transmission losses and power supply losses between physical nodes and carry dynamic transmission coefficients that change with the corresponding loss states. These dynamic transmission coefficients reflect the degree of topological coupling between the accounting data stream and resource entities. The energy consumption cost aggregation module captures virtual machine container scaling events in real time, collects load component data streams from corresponding server nodes, and captures dynamic electricity price data and cooling load energy efficiency coefficients within the current accounting cycle. The module writes dynamic electricity price data, cooling load energy efficiency coefficients, node power signals, and virtual machine container scaling events into the same data record according to the server node and sampling time, forming a real-time energy consumption data stream. Subsequently, it uses a time-series load component differential operator to extract the temporal gradient of the load component data stream and allocates node power changes to corresponding virtual computing power units based on the resource changes corresponding to virtual machine container scaling events to obtain the marginal physical overhead of each virtual computing power unit.

[0047] Real-time power estimates of nodes The calculation uses a linear relationship, and the formula is as follows: ,in, For the first Real-time processor utilization of each server node, This is the processor's baseline power consumption constant. For the first Real-time memory utilization of each server node. As a baseline power consumption constant for memory, the energy cost aggregation module will use the real-time electricity price. and transient cooling load efficiency coefficient in substation areas Real-time power estimates of nodes Combined, the dynamic operating cost components of computing nodes The formula is as follows: ,in, The real-time electricity price for the current accounting period. For the first The cooling load energy efficiency coefficient of the substation area where each server node is located is used to calculate the dynamic operating cost component, which is then written into the corresponding topological edge according to the mapping relationship between the physical computing power cluster and the logical computing resources. The cost topology directed graph storage module takes the real-time change of the external power grid electricity price as input, determines the corresponding weight adjustment amount through the current dynamic transmission coefficient of each topological edge, and adds the adjustment amount to the existing edge weight. Since the dynamic transmission coefficient changes with the current state of network transmission loss and power supply substation loss, the same electricity price change produces different weight adjustment amounts on different topological edges, thereby completing the nonlinear flow update of topological edge weights.

[0048] To handle high-concurrency writes, the energy consumption cost aggregation module configures an independent circular buffer storage queue for each physical computing server node in memory space. The calculated dynamic operating cost components are appended to the corresponding queues in the order of generation. A two-stage screening rule monitors the number of data packets accumulated in the circular buffer storage queue and the continuous residence time of the data packets: the discrete batch threshold is set to 50 to 200 packets, and the window time threshold is set to 10ms to 50ms. When either threshold reaches the current value within the set range, the sharded local topology aggregation ring is activated, and time-weighted average dimensionality reduction operation is performed on the transient cost components in the queue to generate discrete micro-batch topology matrix correction instructions. The cost topology directed graph storage module adopts a single-threaded processing mode, with a single write thread updating the discrete micro-batch topology matrix correction instructions to the cost dynamic propagation directed graph topology in a pulse-like synchronous processing mode, and refreshing the weights of the corresponding topology edges.

[0049] A single write thread maintains a double-buffered topology snapshot in memory, formed by an online buffer and an offline buffer. When discrete micro-batch topology matrix correction instructions arrive in pulse form, the single write thread performs matrix addition in the background offline buffer to update the weight values ​​of the topology edges. After the operation is completed, the memory address exchange between the online directed graph topology and the offline buffer is completed within 1ms through an atomic pointer switching operation. Before the address exchange is completed, high-frequency concurrent external read requests continue to access the original online topology snapshot. After the address exchange is completed, subsequent read requests are redirected to the updated topology snapshot. Graph structure write operations are executed serially by the single write thread according to the order of instruction arrival. Each write is based on discrete micro-batch topology matrix correction instructions that have undergone time-weighted average dimensionality reduction processing. The memory addressing time for a single pulse-style synchronous update is controlled within 0.1ms.

[0050] The operation and maintenance scheduling decision module continuously reads the dynamic operating cost components of each topology node in the cost dynamic propagation directed graph topology and forms a real-time cost matrix according to the arrangement order of the topology nodes. The safety margin thresholds corresponding to each topology node form a threshold matrix in the same order. The threshold arbitration unit performs real-time matrix comparison on the corresponding elements in the two matrices. When the dynamic operating cost component of a specific topology node exceeds its safety margin threshold, the topology node is locked as a source node to be migrated. The safety margin threshold is determined based on the long-term benchmark test results of the distributed computing power cluster under typical workload. In this embodiment, it is taken as 1.05 times the historical baseline average cost of the corresponding physical computing power server node. This threshold corresponds to the state where the current dynamic operating cost component of the node is 5% higher than the historical baseline average cost, and covers the normal energy consumption fluctuations caused by slight power supply substation ripple disturbances and uneven airflow distribution during normal operation of the computing power equipment. When the node's dynamic operating cost component is lower than this threshold, the operation and maintenance scheduling decision module maintains the original resource configuration; when the node's dynamic operating cost component continues to exceed this threshold, the threshold arbitration unit locks it as a source node to be migrated.

[0051] To address the topology edge weight oscillations caused by high-frequency flickering loads, the scheduling damping buffer unit in the operation and maintenance scheduling decision module continuously reads the cost variation variance within a continuous sliding period. The scheduling damping buffer unit includes a sliding window memory, a variance calculation unit, and a window adjustment controller. The input of the sliding window memory is connected to the weight output interface of the directed graph topology of cost dynamic propagation, storing the dynamic operating cost components of a specific server node for the most recent 20 sampling periods in a first-in-first-out order. The variance calculation unit is connected to the output of the sliding window memory, calculating the discrete variance and mean of the samples within the window. The window adjustment controller receives the variance calculation results, and its output is connected to the timer trigger of the operation and maintenance scheduling unit. It adjusts the activation countdown time threshold of the timer trigger according to the discrete variance. When the cost variation variance exceeds the calibrated variance safety boundary, the window adjustment controller increases the scheduling activation delay window online, applying logical lag compensation for the issuance of operation and maintenance scheduling instructions.

[0052] The scheduling damping buffer unit uses the convergence state of the cost dynamic propagation directed graph topology as the basis for adjusting the activation threshold. The cost topology directed graph storage module outputs the convergence state based on the changes in topology edge weights within the continuous update cycle, and the scheduling damping buffer unit adjusts the current activation threshold accordingly. After the scheduling damping buffer unit is activated, it delays the issuance of operation and maintenance scheduling instructions by increasing the scheduling activation delay window, thus delaying the migration of business load and the subsequent changes in topology edge weights. After the operation and maintenance scheduling instructions are executed, the physical resource configuration changes, and the new dynamic operating cost component is written back to the cost dynamic propagation directed graph topology, and a new convergence state is formed. Therefore, the convergence state of the cost dynamic propagation directed graph topology determines the activation threshold of the scheduling damping buffer unit, and the activation state of the scheduling damping buffer unit, in turn, affects the subsequent topology convergence process through operation and maintenance scheduling instructions. The two form a bidirectional dependency, and the logical lag compensation restricts the issuance of operation and maintenance scheduling instructions to two conditions being met simultaneously: the scheduling activation delay window has expired, and the dynamic operating cost component of the source node to be migrated still exceeds the safety margin threshold.

[0053] To avoid continuous waiting caused by bidirectional dependencies before the business load has been migrated, the system adopts a forced time-sharing decoupling mechanism based on the static period of topology updates. When the scheduling activation delay window increases, if the cost dynamic propagation directed graph topology does not receive new discrete micro-batch topology matrix correction instructions for a long time, the independent timer starts to accumulate the topology static residence time. When the topology static residence time reaches 10 seconds, the current topology state is marked as a quasi-static pseudo-convergence state. At this time, the logic switch temporarily cuts off the real-time dependency of the scheduling damping buffer unit's activation state on the topology convergence state and resets the activation threshold to the initial basic safety weight threshold. After the scheduling activation delay window expires, the operation and maintenance scheduling unit issues operation and maintenance scheduling instructions according to the current threshold comparison results, changes the physical resource configuration through actual business load migration, and makes the cost dynamic propagation directed graph topology re-enter the update process.

[0054] When communication links become congested due to extreme high load, the system deploys an asynchronous transaction processing module on the flow path between the energy consumption cost aggregation module and the cost topology directed graph storage module. The asynchronous transaction processing module reads the asset value level tags of logical tenants, identifies the nodes corresponding to logical computing resources with tags reaching the current high priority level as high-yield core nodes, and establishes mutually isolated high-priority dedicated channels and low-priority queuing buffers. The dynamic cost accounting data packets of high-yield core nodes are allocated to the high-priority dedicated channel for exemption acceleration and do not enter the low-priority queuing buffer; the dynamic cost accounting data packets of ordinary nodes are temporarily stored in the low-priority queuing buffer, and are then written into the cost dynamic propagation directed graph topology by a single write thread in the queue order. The asynchronous transaction processing module adjusts the buffer throughput window according to the real-time packet loss rate of the energy consumption cost aggregation module, and its queuing output results inversely limit the single batch synchronization size of the energy consumption cost aggregation module.

[0055] When the scheduling activation delay window expires and the instruction issuance boundary is met, the operation and maintenance scheduling unit issues an operation and maintenance scheduling instruction to migrate the high-energy-consuming business load on the source node to be migrated from the high-electricity-price substation area to the low-electricity-price substation area. After the migration is completed, the energy consumption cost aggregation module continues to collect the real-time energy consumption data stream of the migrated server node, recalculates the dynamic operating cost component, and updates the weight of the corresponding topology edge through the cost topology directed graph storage module.

[0056] Example 2: This example selects a large-scale cloud data center as the test object. The data center is divided into 3 physical partitions, with a total of 1,500 computing server nodes deployed. It carries mixed business loads such as financial transaction processing, real-time commercial logistics management and user administrative approval. The test platform deploys a high-performance cluster monitoring system. The sampling resolution of processor utilization is set to 0.1%, the acquisition accuracy of memory utilization is set to 0.5%, the instantaneous update cycle of the power supply price interface is set to 1.0s, and the monitoring accuracy of the energy efficiency coefficient of the cooling load in the substation area is set to 0.01. During the test, a business request flow conforming to the Poisson distribution is injected into the computing server nodes, so that the physical load changes dynamically between 15% and 92%.

[0057] The control group used a fixed batch processing mode, processing resource consumption data centrally after a fixed period. The experimental group used a dynamic accounting mode, where the energy consumption cost aggregation module obtained real-time energy consumption data streams from server nodes and converted them into dynamic operating cost components. Discrete micro-batch topology matrix correction instructions were formed through a circular buffer storage queue and a two-stage screening rule. The cost topology directed graph storage module then updated the cost dynamic propagation directed graph topology using a single-threaded processing method. The time difference between the collection time and the corresponding time of the accounting results was compared. The accounting time difference of the control group was 300.0s, while that of the experimental group was 0.8s. After statistically analyzing the cost calculation results of each computing power server node during load fluctuations, the cost calculation fluctuation rate of the control group was 18.5%, while that of the experimental group was 3.2%.

[0058] The operation and maintenance scheduling decision module continuously reads the dynamic operating cost components of each server node in the cost dynamic propagation directed graph topology, and issues operation and maintenance scheduling instructions when the dynamic operating cost components exceed the safety margin threshold. The control group triggers cross-substation area business migration based on fixed batch calculation results, issuing a total of 42 migration instructions. The experimental group applies logical lag compensation to the instruction issuance through the scheduling damping buffer unit, issuing a total of 18 migration instructions. The calculation results of each business load are compared with the actual resource consumption records of the corresponding time period. The business load calculation distortion rate of the control group is 12.6%, while that of the experimental group is 1.1%.

[0059] To examine the impact of high-frequency disturbances on weight updates, Gaussian white noise with a signal-to-noise ratio of 25dB was superimposed on the input stream of the computing nodes to simulate computational interference caused by sudden changes in physical load. In the control group, batch processing was performed within a fixed window, and instantaneous power changes were only included in cost accounting after the current batch ended. Therefore, the cost allocation coefficient lagged behind the load change. In the experimental group, when the... Time-series gradient of load component data stream of each computing node When the preset mutation threshold is exceeded, the energy consumption cost aggregation module triggers the weighting of the corresponding topological edge. Refresh, weight adjustment in real time Calculate using the formula: ,in, For the first Temporal gradient of the data stream of the load component of each computing node This represents the weight of the topological edge corresponding to the computing power node. This is the real-time adjustment amount for the weights. This represents the incremental change in dynamic operating cost components. The damping smoothing coefficient is set to a value range of 0.1 to 0.3. After the discrete micro-batch topology matrix correction instruction is formed, the cost topology directed graph storage module refreshes the weight of the corresponding topology edge through a single write thread. Under 25dB noise conditions, the update lag time of the dynamic operating cost component of the test group remains within 1.0s.

[0060] When the dynamic operating cost component of a computing node reaches the safety margin threshold of 1.05 relative to the historical baseline average cost, the operation and maintenance scheduling decision module locks the node as a source node to be migrated. The scheduling damping buffer unit simultaneously reads the cost variation variance within the continuous sliding period. When the cost variation variance exceeds the calibrated variance safety boundary of 0.08, the scheduling activation delay window is increased online from 5.0s to 15.0s. After the scheduling activation delay window expires, the operation and maintenance scheduling unit issues an operation and maintenance scheduling instruction based on the dynamic operating cost component at the time of expiration, migrating the high-energy-consuming business load from the high-electricity-price substation area to the low-electricity-price substation area. If the short-term load flicker falls back below the safety margin threshold within the delay window, the migration is not triggered, thereby reducing the back-and-forth movement of business loads between different substation areas.

[0061] When the load fluctuation is small, the fixed batch processing mode can maintain continuous business operation. When the cost variation variance exceeds 0.15, the control group cannot keep up with the instantaneous load change in time because the accounting results cannot keep up with the instantaneous load change. The business load accounting distortion rate increases exponentially with the increase of variance. The experimental group discretizes the continuous weight update flow by using a sharded local topology aggregation ring and adjusts the scheduling activation delay window by using a scheduling damping buffer unit. Under the same disturbance conditions, the business load accounting distortion rate is controlled within 2%.

[0062] Example 3: The cloud environment data center operation and maintenance management system in this example includes an energy consumption cost aggregation module, a cost topology directed graph storage module, and an operation and maintenance scheduling decision module. When the load of physical computing server nodes changes frequently, the system processes the real-time energy consumption data stream of each node according to the sampling time and allocates the energy consumption expenses generated by the business load to the corresponding virtual computing units. After the ambient temperature of the computer room in a specific substation area changes, the energy efficiency coefficient of the cooling load will have a transient deviation. The system incorporates this deviation, along with the electricity price and node power of the same period, into the calculation of the dynamic operating cost component.

[0063] The energy consumption metering module is connected to the energy cost aggregation module through an internal data bus. It collects the ambient temperature of the computer room and the real-time wattage of the computing equipment in real time. During the collection process, the energy consumption metering module reads the processor core frequency, voltage status and memory read / write bus throughput, and converts the raw signals into the instantaneous power flow corresponding to the real-time power estimate of the node. The energy cost aggregation module merges this power flow, dynamic electricity price data and cooling load energy efficiency coefficient into a real-time energy consumption data flow at the same sampling time. Based on the mapping relationship between virtual computing units and physical computing server nodes, it obtains the marginal physical overhead of each virtual computing unit.

[0064] In the calculation of dynamic operating cost components, the real-time power estimate of the node is in kilowatts, the real-time power supply price is in yuan per kilowatt-hour, and the energy efficiency coefficient of the cooling load is a dimensionless parameter. The energy cost aggregation module first calculates the real-time power supply price according to the sampling period, and then obtains the dynamic operating cost component expressed in yuan per second based on the calculation relationship between the determined real-time power estimate of the node, the real-time power supply price, and the energy efficiency coefficient of the cooling load.

[0065] To handle the dynamic cost accounting data packets arriving in a short period of time, the energy cost aggregation module sets up a circular buffer storage queue in memory for each physical computing server node, and adopts a two-stage filtering rule consisting of a discrete batch threshold and a window time threshold. In this operation, the number of data packets in the circular buffer storage queue first reaches the discrete batch threshold of 150. The energy cost aggregation module then performs a weighted average operation on the dynamic operating cost components in the queue, forms a discrete micro-batch topology matrix correction instruction, and sends the instruction to the cost topology directed graph storage module.

[0066] The cost topology directed graph storage module maintains the cost dynamic propagation directed graph topology through the graph data storage unit. Topology nodes are used to represent the physical computing power cluster and the logical computing resources that establish a mapping relationship with the physical computing power cluster. Topology edges are used to represent the network transmission loss and power supply substation loss between physical nodes. After receiving the discrete micro-batch topology matrix correction instruction, the graph data storage unit updates the weight of the corresponding topology edge by a single write thread according to the order of instruction arrival. Each topology edge obtains different weight adjustment amounts according to the real-time changes in the external power grid electricity price and the current loss status, thereby completing the nonlinear flow update.

[0067] During the update process, a single write thread writes the edge weights in the snapshot backup, while the read thread continues to read the current online snapshot. After the write is completed, the graph data storage unit switches to the updated snapshot, so that the read and write operations fall on different snapshots, avoiding the blocking of the query process due to lock contention. The operation and maintenance scheduling decision module reads the dynamic operating cost components of each node according to the arrangement order of the topology nodes, forming a real-time cost matrix, and compares it with the safety margin threshold matrix arranged in the same order in real time. In this embodiment, the rated energy consumption benchmark is formed based on the node's physical rated power and the regional average cooling load energy efficiency coefficient, and then converted into the dynamic operating cost baseline under the node's rated operating conditions by combining the real-time power supply price. The safety margin threshold is taken as 1.1 times the baseline. When the dynamic operating cost component of a specific topology node exceeds the corresponding safety margin threshold, the threshold arbitration unit locks the node as a source node to be migrated.

[0068] To reduce repeated migrations caused by high-frequency flicker loads, the scheduling damping buffer unit continuously monitors the cost variation variance over the most recent 20 sampling periods. The variance is calculated by the discrete deviation between the dynamic operating cost component and the mean cost of the sliding window in each sampling period. The calibration variance safety boundary is set to 0.09, which is determined based on the discrete distribution of the dynamic operating cost component under historical high-frequency flicker load conditions.

[0069] When the cost variation variance does not exceed 0.09, the dynamic operating cost component does not show high-frequency bidirectional jumps within 20 consecutive sampling periods. The current cost change is treated as a continuous high energy consumption overhead, and the scheduling activation delay window remains at the normal setting. When the cost variation variance exceeds 0.09, it indicates that the virtual machine container is frequently expanding and shrinking within the continuous sampling period. The scheduling damping buffer unit adjusts the scheduling activation delay window to 8s and applies logical lag compensation for the issuance of operation and maintenance scheduling instructions.

[0070] After the scheduling activation delay window expires, the operation and maintenance scheduling decision module reads the dynamic operating cost component of the source node to be migrated again. If this component still exceeds the safety margin threshold, the operation and maintenance scheduling unit issues an operation and maintenance scheduling instruction to migrate the high-energy-consuming business load from the high-electricity-price substation area to the low-electricity-price substation area. If the component falls back to within the safety margin threshold during the delay period, the operation and maintenance scheduling instruction will not be issued again. Through the above processing, short-term load flickering will not directly trigger the business load to switch repeatedly between different server nodes, and subsequent cost accounting will continue to be based on the physical resource status after migration.

[0071] Example 4: This example combines Figures 1 to 2 This document describes the dynamic accounting and maintenance management system and methods for cloud environment data center operations costs, such as... Figure 1As shown, the energy consumption cost aggregation module acquires real-time energy consumption data streams and converts them into operational cost components. It then sequentially appends cost calculation results through a circular buffer storage queue. After a two-stage filtering rule, the continuously updated streams are aggregated into discrete batches and output to the cost topology directed graph storage module. The cost topology directed graph storage module receives discrete micro-batch correction instruction topology matrix synchronous update messages and writes graph structure snapshots through a single thread in the graph data storage unit. This enables the cost dynamic propagation directed graph to refresh the weights of the corresponding topology edges and transfer them to the operation and maintenance scheduling decision module. The operation and maintenance scheduling decision module is triggered when the read dynamic cost components exceed the safety margin threshold. After triggering, it splits into two parallel processing branches. The first technical path is to directly transfer to the operation and maintenance scheduling instruction limit issuance boundary to smooth out disturbances. The second technical path is to monitor the variance through the scheduling damping buffer unit, apply hysteresis compensation, and then transfer to the operation and maintenance scheduling instruction limit issuance boundary to smooth out disturbances.

[0072] like Figure 2 As shown, in the original resource configuration state, when the dynamic operating cost component exceeds the safety margin threshold, the system transitions to the source node to be migrated and locked state. In the source node to be migrated and locked state, if the variance does not exceed the standard and the scheduling activation delay window expires, it returns to the original resource configuration state. If the cost variation variance within the continuous sliding period exceeds the calibrated variance safety boundary, it transitions to the logical lag compensation delay state. In the logical lag compensation delay state, if short-term load flicker falls below the threshold within the delay window, it returns to the original resource configuration state. If the topology static dwell time reaches 10 seconds, it transitions to the quasi-static pseudo-convergence state. If the round-trip delay continuously exceeds 50ms and the variation variance is continuously greater than 0.09, it transitions to the link defense mode state. In the quasi-static pseudo-convergence state, after forced time-sharing decoupling and the issuance of operation and maintenance scheduling instructions, it returns to the original resource configuration state. In the link defense mode state, when the communication link recovers stability, it returns to the source node to be migrated and locked state.

[0073] Example 5: Before the system is put into operation and maintenance, the energy consumption cost aggregation module first performs offline benchmark calibration. The operation and maintenance personnel collect instantaneous power data of each computing server node when the processor utilization rate is 0% to 5% and the memory access is idle through controlled no-load testing. The energy consumption cost aggregation module separates the static basic energy consumption generated by the rack fans and power conversion circuits from the collected data and stores it in the memory configuration library as a node-specific equipment redundancy thermal overhead correction item. When calculating the dynamic operating cost component, the correction item is first deducted from the real-time energy consumption data stream of the server node so that the calculation result corresponds to the actual power consumption generated by the business load.

[0074] During the load migration determination process, the operation and maintenance scheduling decision module reads the weight change rate of each edge in the directed graph topology of cost dynamic propagation and calculates the weight variation gradient of the target node. The scheduling damping buffer unit synchronously monitors the cost variation variance during the continuous sliding period. When the cost variation variance exceeds the calibration variance safety boundary, the online scheduling activation delay window is increased to apply logical lag compensation to the issuance of operation and maintenance scheduling instructions.

[0075] This embodiment uses a discretized buffer adjustment mechanism to determine whether the weight change is persistent. When the weight variation gradient is greater than 0.12 yuan per second for 5 consecutive sampling periods, the operation and maintenance scheduling decision module switches the original single-point threshold judgment to a lag response loop and constructs a sliding sample mean sequence in memory. When the sliding average of the dynamic operating cost component relative to the historical baseline cost mean continuously exceeds the safety margin threshold of 1.05 and the cumulative time reaches 10 sampling periods, the operation and maintenance scheduling unit issues an operation and maintenance scheduling instruction after the scheduling activation delay window expires, migrating the high-energy-consuming business load from the high-electricity-price substation area to the low-electricity-price substation area. When the sliding average falls back to within the safety margin threshold during the delay period, the operation and maintenance scheduling instruction will not be issued again.

[0076] When the system processes cross-substation area business load migration requests, it simultaneously performs cost-benefit calculations and privacy desensitization verification. The operation and maintenance scheduling decision module has a pre-set set of data flow desensitization rules. Before reporting the business migration logs of logical tenants to the management console, it removes the tenant's specific business IP address, access frequency, and exact business content, retaining only the migration event type, source end electricity price status, destination end node power consumption changes, and operation timestamp. The retained technical indicators are hashed and then written to the audit database, enabling operation and maintenance audit records to be traced by operation timestamp. At the same time, it does not retain the original business information that can be directly correlated with specific individual tenants.

[0077] Under simulated extreme fluctuation conditions, the discretized buffer adjustment mechanism segments the weight update stream to reduce conflicts caused by transient concurrent updates; the scheduling activation delay window and the sliding sample mean determination constitute a double-layer lag compensation, reducing the trigger frequency of unexpected business migration by more than 78%; the time-series reliability index of the operation and maintenance accounting data stream is statistically analyzed based on the degree of correspondence between accounting records and actual resource status changes in time, and its value has increased from the initial 0.82 to 0.98; the accounting data remains continuous during the processing period, and the accounting results are consistent with the audit records.

[0078] Example 6: Before the current system is put into operation, a controlled no-load test is performed on each computing server node. During the test, the processor's basic thermal power consumption and memory no-load power consumption are collected, and the basic static power consumption factor of the corresponding node is calculated and stored. The energy consumption cost aggregation module will As a correction item for equipment redundancy thermal overhead, it is deducted from the subsequently collected real-time power estimates of nodes, so that the dynamic operating cost component corresponds to the actual power consumption generated by the business load, and the parameters used for node cost allocation thus have a traceable test source.

[0079] After completing the no-load test, a standardized load stress test program was run on each computing server node. The load component of the computing node was increased in increments of 10%, and the real-time power estimates of the nodes were recorded simultaneously. Ten sets of load and power data were obtained for each node. The node power response curve was fitted using the least squares method to calculate the gradient of unit energy consumption growth. : ,in, For the first The computing server node is at the _ ... Load utilization in this test This is the estimated real-time power of the node corresponding to this load. and These are the arithmetic averages of the load utilization rate and the estimated real-time power of the node obtained from each test. The total number of test sampling points is set to 10 in this embodiment. The calculated... The configuration metadata of the corresponding topology node in the directed graph topology of cost dynamic propagation is written as the calculation basis when allocating costs to logical tenants.

[0080] During periods of high-frequency load fluctuations, the energy cost aggregation module adjusts the damping smoothing coefficient based on the temporal gradient of the load component data stream. When the temporal gradient for three consecutive sampling periods does not exceed 0.15, Keep it at 0.25; if the temporal gradient exceeds 0.15 for three consecutive sampling periods, then... The value was adjusted from 0.25 to 0.15. After the adjustment, the proportion of the increase in the dynamic operating cost component in the real-time adjustment of the topology edge weights decreased accordingly, and the weight update amplitude of the directed graph topology for high-frequency random disturbances was reduced.

[0081] The operation and maintenance scheduling decision module also sets an automatic reporting threshold for early warning work orders. The price will be adjusted according to the difference between the real-time electricity price and the average electricity price in the substation area: ,in, Based on the basic security weight threshold, This is the electricity price-sensitive adjustment coefficient. For real-time electricity pricing, The average electricity price within the substation area is used to generate early warning work orders based on the automatic reporting threshold. Operation and maintenance dispatch instructions are still issued based on the judgment results of the safety margin threshold, dispatch activation delay window, and logical lag compensation.

[0082] The electricity price sensitivity adjustment coefficient is determined offline based on the ratio of historical peak-valley electricity price difference to the marginal cost of data center cross-substation area business migration. In this embodiment, it is fixed at 0.5 kWh per yuan. When the real-time power supply price is higher than the average power supply price of the substation area, the correction term corresponding to the price difference is greater than zero, and the early warning judgment is adjusted according to the price change. When the real-time power supply price is lower than the average power supply price, the correction term is reduced accordingly. When the electricity price sensitivity adjustment coefficient is lower than 0.1 kWh per yuan, the response of the automatic reporting threshold of the early warning work order to the step change of the electricity price is weakened. When the coefficient is higher than 2.0 kWh per yuan, the threshold adjustment range increases, which can easily lead to early warning threshold overshoot, false early warning work orders, and erroneous resource migration.

[0083] When the physical load fluctuation of the data center exceeds the preset security boundary, the operation and maintenance scheduling decision module enters the link defense mode and reads the round-trip latency of the dynamic cost accounting data packets. and its variance ,when For more than 50ms consecutively, and When the value is consistently greater than 0.09, the scheduling damping buffer unit increases the scheduling activation delay window from 1.0s to 8.0s, delaying the issuance of operation and maintenance scheduling instructions until the communication link is restored to stability. After adopting this processing, the packet loss rate of key dynamic cost accounting data packets remained below 0.05% during the test.

[0084] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A dynamic accounting and maintenance management system for cloud environment data center operation costs, characterized in that the system... include: The energy consumption cost aggregation module interacts with the cost topology directed graph storage module, which in turn interacts with the operation and maintenance scheduling decision module. The energy consumption cost aggregation module includes a ring buffer storage queue, which is used to acquire real-time energy consumption data streams from server nodes and convert them into dynamic operating cost components. Through the ring buffer storage queue and a two-stage filtering rule, the dynamic operating cost components are aggregated into discrete micro-batch topology matrix correction instructions. The cost topology directed graph storage module includes a graph data storage unit, which is used to receive discrete micro-batch topology matrix correction instructions and update the cost dynamic propagation directed graph topology in the internal storage using a single-threaded processing method to refresh the weight of the corresponding topology edge. The operation and maintenance scheduling decision module includes a scheduling damping buffer unit, which reads the dynamic operating cost components of each server node in the cost dynamic propagation directed graph topology and issues operation and maintenance scheduling instructions when the dynamic operating cost components exceed the safety margin threshold. When the cost variation variance in a continuous sliding period exceeds the calibrated variance safety boundary, the scheduling damping buffer unit increases the scheduling activation delay window online to apply logical lag compensation for the issuance of operation and maintenance scheduling instructions.

2. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, The energy cost aggregation module has an independent circular buffer storage queue for each server node in the memory space, so that the calculation results of dynamic operating cost components are sequentially appended to the circular buffer storage queue. The two-stage filtering rules include two-stage exclusion filtering of discrete batch threshold and window time threshold. The discrete batch threshold is set to 50 to 200, and the window time threshold is set to 10ms to 50ms. It is used to aggregate the continuous weight update stream into discrete micro-batch topology matrix correction instructions, and coordinate with a single write thread to update the cost dynamic propagation directed graph topology in a pulse synchronous processing mode, eliminating the multi-way mutex lock contention of the directed graph data structure under the load impact of high-density resource expansion and contraction.

3. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, An asynchronous transaction processing module is set up on the flow path between the energy consumption cost aggregation module and the cost topology directed graph storage module. The asynchronous transaction processing module is used to read the asset value level label of the logical tenant to establish a mutually isolated high-priority dedicated channel and a low-priority queuing buffer, and allocate the dynamic cost accounting data packet of the high-yield core node to the high-priority dedicated channel to complete the exemption acceleration.

4. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, The energy consumption cost aggregation module includes: capturing virtual machine container scaling events in real time and capturing dynamic electricity price data and cooling load energy efficiency coefficient of the corresponding server node within the current accounting cycle; the energy consumption cost aggregation module integrates dynamic electricity price data and cooling load energy efficiency coefficient into a real-time energy consumption data stream, deconstructs the marginal physical overhead of each virtual computing unit, and generates a high-frequency concurrent dynamic operating cost component.

5. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, The cost topology directed graph storage module stores a cost dynamic propagation directed graph topology, which includes multiple topology nodes and multiple topology edges. Topology nodes are used to represent physical computing power clusters and logical computing resources that establish a mapping relationship with physical computing power clusters. Topology edges are used to represent network transmission losses and power supply and substation losses between physical nodes. The edge weights of the topology edges are updated nonlinearly with the real-time fluctuations of the external power grid electricity price.

6. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, The operation and maintenance scheduling decision module is equipped with a threshold arbitration unit. The threshold arbitration unit is used to read the real-time dynamic operating cost component of each topology node and compare it with the set safety margin threshold in real time. When the operating cost of a specific topology node exceeds the safety margin threshold, the specific topology node is locked as the source node to be migrated.

7. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, The operation and maintenance scheduling decision module includes an operation and maintenance scheduling unit. When the scheduling activation delay window expires, the operation and maintenance scheduling unit issues operation and maintenance scheduling instructions to migrate high-energy-consuming business loads from high-electricity-price substation areas to low-electricity-price substation areas in real time, so as to balance the resource and economic expenditure of each substation area.

8. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, The activation threshold of the scheduling damping buffer unit is dynamically defined by the convergence state of the global directed graph topology. The logical lag compensation output by the scheduling damping buffer unit is used to limit the instruction issuance boundary of the operation and maintenance scheduling decision module. The activation state of the scheduling damping buffer unit and the convergence state of the global directed graph topology form a bidirectional dependency.

9. The cloud environment data center operation cost dynamic accounting and maintenance management system according to claim 1, characterized in that, The system also includes an asset inventory module and an energy consumption metering module. The asset inventory module connects with the energy cost aggregation module via an internal data bus to calculate the depreciation cost of physical servers. The energy consumption metering module connects with the energy cost aggregation module via an internal data bus to collect data on the real-time ambient temperature of the computer room and the real-time wattage of computing equipment.

10. A method for dynamic accounting and maintenance management of cloud environment data center operating costs, which is implemented through the dynamic accounting and maintenance management system for cloud environment data center operating costs as described in claim 1, characterized in that... Includes the following steps: Step S1: Use the energy consumption cost aggregation module to obtain the real-time energy consumption data stream of the server node and convert it into dynamic operating cost components. Through the ring buffer storage queue and the two-stage filtering rules, the dynamic operating cost components are aggregated into discrete micro-batch topology matrix correction instructions. Step S2: Receive discrete micro-batch topology matrix correction instructions using the cost topology directed graph storage module, and update the cost dynamic propagation directed graph topology stored inside the graph data storage unit using a single-threaded processing method to refresh the weights of the corresponding topology edges. Step S3: Use the operation and maintenance scheduling decision module to read the dynamic operating cost components of each server node in the cost dynamic propagation directed graph topology. When the dynamic operating cost components exceed the safety margin threshold, issue an operation and maintenance scheduling instruction. Use the scheduling damping buffer unit to increase the scheduling activation delay window online when the cost variation variance in the continuous sliding period exceeds the calibrated variance safety boundary, and apply logical lag compensation for the issuance of the operation and maintenance scheduling instruction.

Images