Server cluster energy efficiency optimization method based on dynamic resource scheduling

Abstract

The invention belongs to the field of server clusters, and particularly relates to a dynamic resource scheduling-based server cluster energy efficiency optimization method, which comprises the following steps of: obtaining a resource load capacity of a server in a historical time period and a total resource load capacity of a server cluster, and training a load prediction model based on data; predicting the total resource load amount in the future time period through the load prediction model, and generating a load curve of the corresponding time; dynamically making a plurality of resource scheduling decisions according to the load curve, wherein the decisions comprise task migration, dormancy awakening and dynamic frequency modulation among the servers; and simulating and executing an energy consumption curve of each resource scheduling decision, wherein the decision with a low fluctuation index of the energy consumption curve is a target resource scheduling decision. The total resource load amount of the server cluster is predicted, a plurality of energy consumption curves conforming to the predicted load curve are generated and compared, the energy consumption curve with a low fluctuation index is selected as a target resource scheduling decision, and the influence of fluctuation on server hardware is solved while energy efficiency optimization is performed.

Description

Technical Field

[0001] This invention relates to the field of server clusters, and more specifically to a method for optimizing the energy efficiency of server clusters based on dynamic resource scheduling. Background Technology

[0002] The primary application of server clustering technology is network load balancing. Network load balancing uses a cluster of two or more host computers working together to provide high availability and scalability for servers. Internet clients access the cluster using a single IP address or a set of addresses. Clients cannot distinguish between a cluster and a single server. Server applications do not indicate that they are running on a cluster. However, a network load balancing cluster differs significantly from a single host running a single server application because it can provide uninterrupted service even if one of the cluster hosts fails. A cluster also responds to client requests faster than a single host.

[0003] If a host fails or goes offline, network load balancing provides high availability by redirecting incoming network traffic to the working cluster hosts. Existing connections to the offline host will be lost, but internet service remains available. In most cases (e.g., in the case of a web server), client software can automatically retry failed connections, and clients experience only a few seconds' delay in receiving responses. Network load balancing provides variable performance by distributing incoming network traffic among one or more virtual IP addresses (cluster IP addresses) assigned to the network load balancing cluster. Hosts in the cluster can then respond simultaneously to different client requests, or even multiple requests from the same client. For example, a web browser can retrieve multiple images from all the individual web pages from different hosts in the network load balancing cluster. This increases processing speed and reduces the time to respond to clients.

[0004] A server cluster is a combination of multiple servers, all of which are running at high energy consumption. To reduce energy consumption, frequent starting and stopping of servers can lead to power fluctuations, which in severe cases can cause hardware damage, reduce lifespan, and bring additional maintenance costs. Currently, there is a lack of effective methods to optimize the energy efficiency of server clusters. Summary of the Invention

[0005] (a) Purpose of the invention

[0006] To address the technical problems existing in the background art, this invention proposes a server cluster energy efficiency optimization method based on dynamic resource scheduling, which features dynamic resource scheduling decisions adapted to the resource load of the server cluster.

[0007] (II) Technical Solution

[0008] To address the aforementioned technical problems, this invention provides a server cluster energy efficiency optimization method based on dynamic resource scheduling, comprising the following steps:

[0009] Obtain the resource load of the server and the total resource load of the server cluster over a historical time period, and train a load prediction model based on the data.

[0010] The total resource load in future time periods is predicted using a load forecasting model, and load curves for the corresponding time periods are generated.

[0011] Based on the load curve, several resource scheduling decisions are dynamically made, including task migration, hibernation and wake-up, and dynamic frequency adjustment between servers.

[0012] The energy consumption curves of each resource scheduling decision are simulated, and the resource scheduling decision with the lowest energy consumption fluctuation index is the target resource scheduling decision.

[0013] Preferably, the total resource load of the server cluster is the sum of the resource load index values ​​of each individual server.

[0014] Cluster total CPU utilization = Σ(single-machine CPU utilization × single-machine CPU cores) / total cluster CPU cores × 100%;

[0015] Total cluster memory usage = Σ Single machine memory usage;

[0016] Total network bandwidth usage of the cluster = Σ network egress / ingress traffic rate of a single machine.

[0017] Preferably, the training load prediction model is based on a deep learning model to predict the total resource load of the server cluster at fixed intervals within a future time period.

[0018] The model's input data is the total resource load of the server cluster over a historical time period, and the output is the predicted load value for the corresponding time period.

[0019] The load prediction model is continuously optimized. As the server cluster runs for longer, new resource load data is acquired, and the model is iterated.

[0020] Preferably, a range for future time periods and time intervals is set;

[0021] The total resource load of the server cluster is predicted with a 1-hour period and 5-minute intervals. The predicted load values ​​are for time series of 1-5, 6-10...51-55, 56-60 minutes. The predicted load values ​​at each time point within the next 1-hour period are connected to form a predicted load curve, which shows the trend of expected load changes over time.

[0022] Preferably, based on the predicted load curve of the server cluster, the resource status of the server cluster is intelligently adjusted. For the same load curve, at least one resource scheduling strategy is generated. The solution meets the prediction requirements of the load curve changing over time to plan the timing and action type of resource adjustment.

[0023] Preferably, a buffer resource pool is reserved in the server cluster to always maintain 10-15% over-provisioned computing power. Based on the peak load of the cluster, some servers are designated as hot standby server groups, which are prohibited from hibernation and are kept in standby mode with the lowest power consumption. The reserved ratio is automatically adjusted according to the volatility of the predicted load curve.

[0024] Preferably, the input for the simulation is: the total resource load of the server cluster predicted at fixed intervals in the predicted load curve over a future period, and the specific scheme strategy for the resource scheduling decision scheme.

[0025] The energy efficiency simulation engine performs the following operations based on the set time intervals:

[0026] Obtain the predicted cluster load value at the current simulation time point, derived from the server cluster predicted load curve;

[0027] Calculate the total energy consumption of the cluster at the current simulation time point;

[0028] Advance the simulation time to the next time point.

[0029] Output a simulated energy consumption curve.

[0030] Preferably, the curve corresponds to the time axis of the predicted load curve, and each time point has a simulated total power consumption value of the cluster. For each resource scheduling decision scheme A, B, and C, an energy consumption curve will be generated independently through simulation. The energy consumption curve fluctuation index is used to quantitatively evaluate the stability of the energy consumption curve.

[0031] Preferably, the energy consumption curve fluctuation indicators include variance, standard deviation, and range;

[0032] Variance is the variance of the power values ​​at all data points on the energy consumption curve. The larger the variance, the greater the fluctuation.

[0033] Standard deviation is the square root of the variance, providing a more intuitive understanding of the absolute magnitude of fluctuations.

[0034] The range is the difference between the maximum power value and the minimum power value. The larger the peak value difference, the greater the fluctuation.

[0035] The maximum power change rate is the maximum absolute value of the power change between adjacent time points, which measures the instantaneous impact intensity.

[0036] The state switching frequency or number of times is the total number of times all servers switch between running-low power-shutdown or low power-shutdown-running states throughout the entire simulation cycle. The more times the state switches, the greater the fluctuation.

[0037] The total frequency change or frequency is used to evaluate the overall magnitude or number of CPU frequency changes;

[0038] The ultimate goal is to reduce volatility, which is to minimize additional losses caused by volatility and optimize reliability / hardware lifespan. Total energy consumption is also an efficiency indicator. Ultimately, a choice must be made between these two, or weights must be assigned to them, and a corresponding strategy must be chosen.

[0039] Preferably, a visual comparison of energy consumption curves for different schemes is performed, and the simulated energy consumption curves and the original predicted load curves of different schemes A, B, and C are plotted on the same graph, with the horizontal axis representing time and the vertical axis representing load or energy consumption.

[0040] By examining trends and shapes, observe which energy consumption curve is above the load curve but smoother, with smaller fluctuations and no frequent peaks and troughs. Also, observe whether the energy consumption curve matches the load curve, indicating that the adjustment is effective, meeting demand without waste.

[0041] The above-mentioned technical solution of the present invention has the following beneficial technical effects: predicting the total resource load of the server cluster, generating multiple energy consumption curves that conform to the predicted load curve for comparison, and selecting the one with low fluctuation index as the target resource scheduling decision, thereby solving the impact of fluctuation on server hardware while optimizing energy efficiency. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the method of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0044] like Figure 1 As shown, the present invention proposes a server cluster energy efficiency optimization method based on dynamic resource scheduling, which includes the following steps:

[0045] S1. Obtain the resource load of the server and the total resource load of the server cluster during historical time periods, and train a load prediction model based on the data.

[0046] The total resource load is obtained from the resource monitoring agent software set up on each server. It periodically collects CPU utilization, memory usage, disk read / write rate and network bandwidth usage. The period includes seconds and minutes. The data for seconds is at least 5 seconds and the data for minutes is at least 1 minute.

[0047] The total resource load of the server cluster is the sum of various resource load index values ​​of the individual server resource load obtained.

[0048] Cluster total CPU utilization = Σ(single-machine CPU utilization × single-machine CPU cores) / total cluster CPU cores × 100%;

[0049] Total cluster memory usage = Σ Single machine memory usage;

[0050] Total network bandwidth usage of the cluster = Σ network egress / ingress traffic rate of a single machine.

[0051] The training load prediction model is based on a deep learning model to predict the total resource load of a server cluster within a fixed time period in the future.

[0052] The model's data is the total resource load of the server cluster over historical time periods obtained from S1, and the output is the predicted load value for the corresponding time period.

[0053] Understandably, the load prediction model is continuously optimized and can be iterated as the server cluster runs longer, acquiring new resource load data.

[0054] S2. Predict the total resource load within a future time period using a load forecasting model, and generate load curves for the corresponding time periods.

[0055] Set the range of future time periods and time intervals;

[0056] Predict the total resource load of the server cluster within the next 1-hour period at fixed 5-minute intervals;

[0057] Predict the total resource load of the server cluster in a fixed 15-minute interval over the next 12-hour period;

[0058] Predict the total resource load of the server cluster over the next 24 hours at fixed 30-minute intervals.

[0059] Taking a 1-hour period and 5-minute intervals as an example, the predicted total resource load of the server cluster includes the predicted load values ​​under the time series of 1-5, 6-10...51-55, 56-60 minutes. The predicted load values ​​at each time point within the next 1-hour period are connected to form a predicted load curve. This curve shows the trend of the expected load changing over time, such as being stable, rising, falling, and fluctuating.

[0060] S3. Dynamically formulate several resource scheduling decisions based on the load curve, including task migration, hibernation / wake-up and dynamic frequency adjustment between servers;

[0061] Based on the server cluster load curve predicted by S2, the resource status of the server cluster is intelligently adjusted to maximize energy efficiency while meeting load requirements.

[0062] Among them, task migration is virtual machine migration, which involves moving running virtual machines or containers online from one physical server to another, concentrating tasks on fewer servers to free up idle servers for shutdown, migrating tasks from high-load servers to low-load servers to achieve load balancing, and clearing tasks from servers that are about to be shut down or enter a low-power state.

[0063] Server sleep / wake-up includes shutdown, power outage, and hibernation. When the forecast shows that the overall cluster load is low enough that some server tasks can be migrated and merged to other servers, the power is turned off or the server is put into a low-power sleep state. Pausing and suspending put the server into a paused state, which saves less power than shutdown but wakes up faster. Idle state optimization reduces power consumption when the server enters the OS-level idle state, in conjunction with hardware features.

[0064] Dynamic frequency scaling dynamically reduces the processor frequency and operating voltage of an active server to reduce power consumption. Frequency reduction usually reduces the maximum computing power of a single core proportionally, and also reduces power consumption roughly proportionally. It is suitable for situations where the load is not particularly low but there is redundant computing power. On multi-core servers that support core-level power management, some idle cores are turned off.

[0065] To address the same load curve, at least one resource scheduling strategy should be generated. For example:

[0066] Option A: Aggressive strategy: Before the predicted electricity price trough arrives at 2 AM, migrate the tasks first, then shut down 60% of the servers, and then wake up the servers one by one at 6 AM when the predicted load rises.

[0067] Option B: Conservative strategy: Primarily relies on dynamic frequency adjustment and core shutdown to adjust computing power, only putting a small number of servers to sleep to reduce migration and power-on / off frequency;

[0068] Option C: Moderate strategy: combining migration, hibernation, and frequency modulation;

[0069] The above scheme meets the forecasting requirements of load curve changes over time to plan the timing and action types of resource adjustments.

[0070] In an optional embodiment, a buffer resource pool is reserved in the server cluster to always maintain 10-15% over-provisioned computing power. Based on the peak load of the cluster, some servers are designated as hot standby server groups, which are prohibited from hibernation and are kept in standby mode with the lowest power consumption. The reserved ratio is automatically adjusted according to the volatility of the predicted load curve. When the volatility is >30%, the over-provisioned computing power is increased from 10-15% to 20%. This can cope with most scenarios of sudden load increases and keep the extra energy consumption at 8%, thereby improving the redundancy and disaster recovery performance of the server cluster in the face of sudden loads.

[0071] S4. Simulate the energy consumption curves of each resource scheduling decision. The resource scheduling decision with the lowest energy consumption curve fluctuation index is the target resource scheduling decision.

[0072] The inputs for the simulation are: the total resource load of the server cluster predicted at fixed intervals over a future period in the predicted load curve; the specific scheme and strategy formulated by the resource scheduling decision scheme; each scheme defines in detail which time points on the predicted load curve, what types of resource adjustment actions to perform, task migration, sleep / wake up X servers, dynamic frequency adjustment, and shut down Y% of cores. The server cluster model includes the physical server model, quantity, CPU model, number of cores, frequency, power consumption characteristics, memory capacity, power consumption characteristics, disk, network power consumption characteristics, power values ​​of various power states such as running, idle, sleep, and shutdown, the time required for state transitions and the power consumption during the transition process, and the time for task migration.

[0073] Simulation process: The energy efficiency simulation engine performs the following operations according to the set time intervals:

[0074] Obtain the predicted cluster load value at the current simulation time point, derived from the server cluster predicted load curve;

[0075] Based on the current resource scheduling decision, check whether resource adjustment actions need to be performed at this point in time, such as migration, server shutdown, server wake-up, or frequency adjustment;

[0076] Simulating the migration process consumes additional computing resources and introduces migration time delays. After the migration, the source server's load is cleared, while the target server's load increases.

[0077] Simulate the state transition process, execute migration tasks before hibernation and shutdown, and simulate the power consumption during the transition time;

[0078] Power consumption is in a transitional state during state transitions;

[0079] Once the target state is entered, the server power consumption stabilizes at a low power level or close to zero power consumption until it is woken up.

[0080] The wake-up process simulates the peak power consumption during the wake-up time. After waking up, the server enters the running and idle states.

[0081] Dynamic frequency adjustment or core shutdown can simulate reducing the CPU frequency or shutting down the cores to improve server processing capacity and power consumption. Under the same load, power consumption is reduced, but processing capacity is decreased, which may affect task processing time and may cause processing delays under high load.

[0082] Calculate the total energy consumption of the cluster at the current simulation time point;

[0083] Move the simulation time to the next time point.

[0084] Output a simulated energy consumption curve, which corresponds to the time axis of the predicted load curve. Each time point shows the simulated total power consumption of the cluster. An independent energy consumption curve is generated for each resource scheduling decision scheme (A, B, C). The energy consumption curve fluctuation index quantitatively assesses the stability of the energy consumption curve. Frequent power consumption fluctuations pose potential risks, mainly stemming from: frequent server power on / off cycles, hibernation / wake-up, frequent switching or significant frequency adjustments of internal server cores, and instantaneous power consumption shocks caused by large-scale task migration.

[0085] Frequent power consumption fluctuations put significant pressure on thermal management. Rapid changes in power consumption lead to rapid changes in temperature, requiring frequent adjustments to the cooling system's fans and air conditioners, which also increases their own energy consumption and may even cause thermal stress damage to the hardware.

[0086] The efficiency of the power system decreases, and the power conversion efficiency may deviate from the optimal efficiency point when the load changes rapidly.

[0087] Potential hardware losses, frequent power state transitions, and temperature cycling can negatively impact the lifespan of electronic components.

[0088] Service quality risks include latency or transient performance degradation caused by state transitions and migrations.

[0089] Energy consumption curve fluctuation indicators include variance, standard deviation, and range;

[0090] Variance is the variance of the power values ​​at all data points on the energy consumption curve. The larger the variance, the greater the fluctuation.

[0091] Standard deviation is the square root of the variance, providing a more intuitive understanding of the absolute magnitude of fluctuations.

[0092] The range is the difference between the maximum and minimum power values. The larger the peak value difference, the greater the fluctuation amplitude.

[0093] The maximum power change rate is the maximum absolute value of the power change between adjacent time points, which measures the instantaneous impact intensity.

[0094] The state switching frequency or number of times is the total number of times all servers switch between running-low power-shutdown or low power-shutdown-running states throughout the entire simulation cycle. The higher the number of times, the greater the fluctuation usually is.

[0095] The total frequency change or frequency is used to evaluate the overall magnitude or number of CPU frequency changes.

[0096] In one embodiment, a visual comparison of the energy consumption curves of different schemes is performed, and the simulated energy consumption curves and the original predicted load curves of different schemes (A, B, C) are plotted on the same graph.

[0097] The horizontal axis represents time (the time interval is based on the prediction setting).

[0098] The vertical axis represents either load (e.g., CPU utilization %) or energy consumption (power value in W).

[0099] By examining trends and shapes, observe which energy consumption curve is above the load curve but smoother, with smaller fluctuations and no frequent peaks and troughs. Also, observe whether the energy consumption curve matches the load curve, indicating that the adjustment is effective, meeting demand without waste.

[0100] As an example:

[0101] Option A is aggressive, with the lowest total energy consumption of 10kWh, but it has the highest volatility, the highest variance, the most state transitions, and the energy curve has many peaks and valleys because many servers are shut down. However, it has the highest volatility due to frequent power-on and power-off cycles.

[0102] Option B is conservative, with a maximum total energy consumption of 15kWh, but the lowest volatility and variance, near-zero state switching frequency, and a relatively flat energy curve. However, it has a higher baseline and higher total energy consumption. It keeps more servers online, only adjusting and reducing frequency, but has the lowest volatility and fewer power-on / off cycles.

[0103] Option C is moderate, with moderate total energy consumption of 12kWh, moderate volatility, moderate variance, and moderate number of state transitions. The curve is smoother than that of option A, but slightly higher than that of option B and has a slight fluctuation. The total energy consumption and volatility may be between the two.

[0104] Based on the S4 rule, which aims to minimize energy consumption curve fluctuation, Option B has the lowest fluctuation, lowest variance, and fewest state transitions. Therefore, Option B will be selected as the target resource scheduling decision.

[0105] Understandably, the ultimate goal is to minimize volatility, which is to reduce additional losses caused by volatility and optimize reliability / hardware lifespan. Total energy consumption is also a very important efficiency indicator. The final choice requires a trade-off between these two, or assigning weights to both and selecting the appropriate strategy based on actual needs.

[0106] In practical applications, if the total energy consumption difference is too large, such as 12kWh for Scheme C and 15kWh for Scheme B, and the fluctuation difference is not particularly significant, Scheme C will be preferred. Choosing a scheme with acceptable fluctuation under low power consumption conditions can effectively reduce energy consumption and has a lower impact on the hardware lifespan of server equipment.

[0107] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. A method for optimizing the energy efficiency of a server cluster based on dynamic resource scheduling, characterized in that, Includes the following steps: Obtain the resource load of the server and the total resource load of the server cluster over a historical time period, and train a load prediction model based on the data. The total resource load in future time periods is predicted using a load forecasting model, and load curves for the corresponding time periods are generated. Based on the load curve, several resource scheduling decisions are dynamically made, including task migration, hibernation and wake-up, and dynamic frequency adjustment between servers. The energy consumption curves of each resource scheduling decision are simulated, and the resource scheduling decision with the lowest energy consumption fluctuation index is the target resource scheduling decision.

2. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, The total resource load of the server cluster is the sum of various resource load index values ​​of the individual server resource load obtained. Cluster total CPU utilization = Σ(single-machine CPU utilization × single-machine CPU cores) / total cluster CPU cores × 100%; Total cluster memory usage = Σ Single machine memory usage; Total network bandwidth usage of the cluster = Σ network egress / ingress traffic rate of a single machine.

3. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, The training load prediction model is based on a deep learning model to predict the total resource load of the server cluster at fixed intervals within a future time period. The model's input data is the total resource load of the server cluster over a historical time period, and the output is the predicted load value for the corresponding time period. The load prediction model is continuously optimized. As the server cluster runs for longer, new resource load data is acquired, and the model is iterated.

4. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, Set the range of future time periods and time intervals; The total resource load of the server cluster is predicted with a 1-hour period and 5-minute intervals. The predicted load values ​​are for time series of 1-5, 6-10...51-55, 56-60 minutes. The predicted load values ​​at each time point within the next 1-hour period are connected to form a predicted load curve, which shows the trend of expected load changes over time.

5. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, Based on the predicted load curve of the server cluster, the resource status of the server cluster is intelligently adjusted. To deal with the same load curve, at least one resource scheduling strategy is generated. The solution plans the timing and action type of resource adjustment to meet the prediction requirements of the load curve changing over time.

6. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, Reserve a buffer resource pool in the server cluster to always maintain 10-15% over-provisioned computing power. Calculate based on the peak load of the cluster, designate some servers as hot standby server groups, prohibit hibernation, and keep them in standby state with the lowest power consumption. The reserved ratio is automatically adjusted according to the volatility of the predicted load curve.

7. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, The inputs for the simulation are: the total resource load of the server cluster predicted at fixed intervals over a future period in the predicted load curve, and the specific scheme strategy for the resource scheduling decision-making scheme. The energy efficiency simulation engine performs the following operations based on the set time intervals: Obtain the predicted cluster load value at the current simulation time point, derived from the server cluster predicted load curve; Calculate the total energy consumption of the cluster at the current simulation time point; Advance the simulation time to the next time point. Output a simulated energy consumption curve.

8. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, The curve corresponds to the time axis of the predicted load curve. At each time point, there is a simulated total power consumption value of the cluster. For each resource scheduling decision scheme A, B, and C, an energy consumption curve will be generated independently through simulation. The energy consumption curve fluctuation index is a quantitative assessment of the stability of the energy consumption curve.

9. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, Energy consumption curve fluctuation indicators include variance, standard deviation, and range; Variance is the variance of the power values ​​at all data points on the energy consumption curve. The larger the variance, the greater the fluctuation. Standard deviation is the square root of the variance, providing a more intuitive understanding of the absolute magnitude of fluctuations. The range is the difference between the maximum power value and the minimum power value. The larger the peak value difference, the greater the fluctuation. The maximum power change rate is the maximum absolute value of the power change between adjacent time points, which measures the instantaneous impact intensity. The state switching frequency or number of times is the total number of times all servers switch between running-low power-shutdown or low power-shutdown-running states throughout the entire simulation cycle. The higher the number of times, the greater the fluctuation. The total frequency change or frequency is used to evaluate the overall magnitude or number of CPU frequency changes; The ultimate goal is to reduce volatility, which is to minimize additional losses caused by volatility and optimize reliability / hardware lifespan. Total energy consumption is also an efficiency indicator. Ultimately, a choice must be made between these two, or weights must be assigned to them, and a corresponding strategy must be chosen.

10. The server cluster energy efficiency optimization method based on dynamic resource scheduling according to claim 1, characterized in that, Visualize and compare the energy consumption curves of different schemes. Plot the simulated energy consumption curves and the original predicted load curves of different schemes A, B, and C on the same graph, with the horizontal axis representing time and the vertical axis representing load or energy consumption. By examining trends and shapes, observe which energy consumption curve is above the load curve but smoother, with smaller fluctuations and no frequent peaks and troughs. Also, observe whether the energy consumption curve matches the load curve, indicating that the adjustment is effective, meeting demand without waste.

Images