Server dynamic power consumption regulation method based on real-time load sensing algorithm

Abstract

This invention discloses a server dynamic power consumption control method based on a real-time load sensing algorithm, relating to the field of server control technology. The invention involves: acquiring historical burst load sensing data; obtaining a burst data network of load sensing type based on the historical burst load sensing data; constructing a burst load sensing signal model for the burst data network; collecting burst load sensing data in real time based on the burst load sensing signal model; constructing a power consumption status model for the burst sensing data; dynamically controlling the power consumption of the burst load sensing data based on the power consumption status model to obtain the corresponding urgent power consumption; and dynamically controlling the power consumption of the urgent power consumption.

Description

Technical Field

[0001] This invention relates to the field of server control technology, specifically to a method for dynamic power consumption control of servers based on a real-time load sensing algorithm. Background Technology

[0002] Load-aware algorithms refer to scheduling strategies that can perceive the load status of backend services (such as the number of connections, response time, CPU, queue depth, etc.) in real time or near real time and dynamically allocate requests accordingly. However, in existing technologies, server load is usually collected in real time according to a fixed collection period. Sudden loads may not be captured by the sampling window, resulting in a load perception blind spot. This leads to low accuracy and delayed perception of sudden loads, which in turn makes it impossible to perform dynamic power consumption control in a timely manner. Secondly, traditional solutions lack comprehensive power state modeling capabilities, cannot accurately quantify idle power consumption and overdraft power consumption gaps, and do not have a layered and progressive power compensation scheduling logic. They can only achieve single passive power consumption adjustment, which leads to insufficient power supply when faced with sudden loads. Therefore, in order to collect sudden loads in a timely manner and to schedule power consumption adequately, this invention provides a method for dynamic power consumption control of servers based on a real-time load perception algorithm. Summary of the Invention

[0003] The purpose of this invention is to provide a method for dynamic power consumption control of servers based on a real-time load awareness algorithm, so as to solve the problems in the background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a server dynamic power consumption control method based on a real-time load awareness algorithm, the method comprising the following steps: Step S1: Obtain historical burst load perception data; obtain burst data network of load perception type based on historical burst load perception data; Step S2: Construct a burst load sensing signal model for the burst data network; collect burst load sensing data in real time based on the burst load sensing signal model; Step S3: Construct a power consumption state model for burst sensing data; based on the power consumption state model, perform dynamic power consumption control on the burst load sensing data to obtain the power consumption required for emergency load sensing data. Step S4: Perform dynamic power consumption control for urgent power needs.

[0005] Furthermore, the process of acquiring historical burst load perception data includes: Obtain historical abnormal load perception data pools for different load perception types of servers within a preset historical collection time period; obtain the maximum and minimum historical abnormal standard data in the historical abnormal load perception data pool; and obtain the difference between the two, which is denoted as the historical abnormal standard data difference. Obtain the ratio of the difference between each historical abnormal load perception data and the historical abnormal standard data in the historical abnormal load perception data pool, and record it as the daily abnormal data value. Set a threshold for the mutation data value of the load sensing type, record daily abnormal data values ​​that exceed the threshold as historical sudden change values, and then mark the corresponding historical abnormal load sensing data as historical sudden load sensing data. Conversely, no action is taken.

[0006] Furthermore, the process of obtaining the load-aware type of burst data network based on historical burst load-aware data includes: Obtain other load perception types that have key correlation characteristics with each load perception type, and denot them as associated load perception types; associate the load perception types with the associated load perception types to generate an associated load data network; Obtain historical associated load perception data of each associated load perception type in the historical abnormal load perception data pool, as well as historical associated burst load perception data of the corresponding historical burst load perception data. By fusing various historical associated load sensing data, the historical daily anomaly change range of associated load sensing type is generated; similarly, by fusing various historical associated sudden load sensing data, the historical sudden anomaly change range of associated load sensing type is generated. Obtain the upper limit of the range of historical daily abnormal changes, and denot it as the upper limit of historical daily abnormal changes. Then, combine it with the range of historical sudden abnormal changes to form the range of sudden data of the corresponding associated load perception type. A corresponding burst signal is set within the burst data range, and burst signal data is formed with the corresponding burst data range; then, each burst signal data is mapped to each associated load perception type of the corresponding associated load data network to form a burst data network of the load perception type.

[0007] Furthermore, the process of constructing a burst load sensing signal model for a burst data network includes: The scope of the sudden data is evenly divided into three sudden level ranges; the sudden level weights are set according to the different criticality of the key correlation features; The range and weight of the suddenness level are uniformly mapped to the various associated load sensing types of the suddenness data network, thereby constructing a suddenness load sensing signal model.

[0008] Furthermore, the process of real-time acquisition of burst load sensing data based on the burst load sensing signal model includes: Set the acquisition period; set temporary acquisition points within the acquisition period, and then set up graded signal acquisition units and trigger acquisition units at the temporary acquisition nodes; The graded signal acquisition unit schedules the current associated load perception data collected according to the acquisition cycle based on the burst load perception signal model and the corresponding associated load perception type. Determine whether the current associated load perception data contains a burst data range corresponding to the associated load perception type; If it exists, a burst signal is triggered, thereby obtaining the burst level range of the current associated load perception data and the corresponding burst level weight; The load burst signal data of the load perception type is obtained by calculating the current associated load perception data of all corresponding associated load perception types according to the corresponding burst level weight.

[0009] Furthermore, the setup process for triggering the acquisition unit includes: The trigger acquisition unit is equipped with a level trigger signal data range mapping set, and obtains the trigger level signal of the load burst signal data according to the level trigger signal data range mapping set; and the start signal of the trigger level signal; the start signal includes yes and no; If the start signal is yes, then the temporary collection point will be directly triggered to collect burst load sensing data of the load sensing type; If the start signal is not present, the corresponding trigger level signal will be determined according to the preset range of pending trigger level signals to determine whether it is a waiting trigger level signal. If so, set the temporary collection period for the temporary collection point, and collect the load burst signal data within the collection period corresponding to the load perception type in real time according to the temporary collection period. Then determine whether the start signal triggers the temporary collection point to collect the burst load perception data of the load perception type until the next collection time point is reached according to the collection period. If not, no action will be taken.

[0010] Furthermore, the process of constructing a power state model for burst sensing data includes: Set the power consumption weight for load awareness type, and allocate the dynamic rated power consumption of the server according to the power consumption weight to obtain the power consumption of each load awareness type. Obtain the number of load-aware type connections associated with the load data network, and record it as the number of connections; then divide the power consumption of the load-aware type equally according to the number of connections, and record it as the distributed power consumption of the load-aware type in the corresponding burst data network; Based on the burst level weight allocation corresponding to the burst level of the load sensing type when acquiring burst load sensing data, the power consumption is obtained by distributing the power consumption and then the difference between the power consumption and the power consumption is obtained and recorded as idle power consumption; and the idle state corresponding to the idle power consumption is generated; the idle state includes non-idle, 0 idle and idle; and then the power consumption state model of burst sensing data is constructed.

[0011] Furthermore, the process of dynamically adjusting the power consumption of the burst load perception data based on the power consumption state model to obtain the power consumption of the urgent demand corresponding to the burst load perception data includes: Obtain the current load power consumption of the burst load sensing data, and obtain the difference between the distributed power consumption of the corresponding load sensing data and the current load power consumption based on the power consumption state model, which is recorded as the emergency demand power consumption. Determine if the power consumption for urgent needs is less than 0; If the emergency power demand is less than 0, the burst load perception data will be used to generate an emergency power demand signal and sent to the idle associated load perception type in the burst data network to obtain the corresponding idle power. All idle power consumption is scheduled to generate schedulable idle power consumption, and then the corresponding idle state is updated to 0 idle. Determine whether the schedulable idle power consumption is greater than the power consumption required by the emergency. If not, the difference between the emergency power consumption and the schedulable idle power consumption is recorded as the emergency power consumption required corresponding to the burst load perception data. If so, the difference between the schedulable idle power consumption and the emergency required power consumption is recorded as the remaining schedulable idle power consumption corresponding to the burst load perception data. If the emergency demand power consumption is greater than or equal to 0, then the emergency demand power consumption is marked as the remaining distributed power consumption.

[0012] Furthermore, the process of dynamically regulating power consumption for urgent needs includes: According to the order of the still urgent demand power consumption from smallest to largest, obtain the first burst load sensing data corresponding to the load sensing type, connect other related load data networks, and obtain the schedulable idle power consumption, remaining schedulable idle power consumption, and remaining distributed power consumption; and sum them up to record as the total schedulable power consumption. Determine whether the total schedulable power consumption is greater than or equal to the power consumption required for urgent needs. If so, the total schedulable power consumption will be allocated to the corresponding urgent power consumption, and the remaining total schedulable power consumption will be obtained to complete the power consumption scheduling for the urgent power consumption. If not, it enters the waiting area; and then continues to acquire the total schedulable power consumption corresponding to the next burst load perception data; Determine whether the total schedulable power consumption is greater than or equal to the power consumption required for urgent needs. If so, then obtain the remaining total schedulable power consumption, and then complete the power consumption scheduling for the urgent power demand. If not, it enters the waiting area; and then continues to acquire the total schedulable power consumption corresponding to the next burst load perception data; Similarly, obtain all remaining total schedulable power consumption; and sum them up to record the remaining schedulable power consumption. If the available schedulable power is greater than or equal to the power of any of the waiting regions that still require urgent power, then the available schedulable power will be scheduled sequentially to the power of any of the waiting regions that still require urgent power, thus completing the power scheduling for the power of any of the waiting regions that still require urgent power; the scheduling will stop when the available schedulable power is less than the power of any of the waiting regions that still require urgent power. Otherwise, no action will be taken. Other power scheduling will be carried out on the power consumption of the still urgent needs in the waiting area, thereby completing the power scheduling of the still urgent needs in the waiting area.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention obtains historical burst load sensing data; obtains a burst data network of load sensing type based on the historical burst load sensing data; constructs a burst load sensing signal model of the burst data network; and collects burst load sensing data in real time based on the burst load sensing signal model; effectively improving the timeliness of burst load data collection.

[0014] 2. Construct a power consumption state model for sudden load sensing data; based on the power consumption state model, dynamically adjust the power consumption of sudden load sensing data to obtain the power consumption of the emergency demand corresponding to the sudden load sensing data; dynamically adjust the power consumption of the emergency demand; effectively schedule the power consumption of sudden load data. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 As shown, a server dynamic power consumption control method based on a real-time load awareness algorithm includes the following steps: Step S1: Obtain historical burst load perception data; obtain burst data network of load perception type based on historical burst load perception data; Step S2: Construct a burst load sensing signal model for the burst data network; collect burst load sensing data in real time based on the burst load sensing signal model; Step S3: Construct a power consumption state model for burst sensing data; based on the power consumption state model, perform dynamic power consumption control on the burst load sensing data to obtain the power consumption required for emergency load sensing data. Step S4: Perform dynamic power consumption control for urgent power needs.

[0019] Further clarification is needed for step S1, specifically regarding the process of acquiring historical burst load perception data, which includes: Step S101: Obtain historical abnormal load perception data pools for different load perception types of servers within a preset historical collection time period; Step S102: Obtain the maximum and minimum historical abnormal standard data from the historical abnormal load perception data pool; and obtain the difference between the two, which is denoted as the historical abnormal standard data difference; Step S103: Obtain the ratio of the difference between each historical abnormal load perception data and the historical abnormal standard data in the historical abnormal load perception data pool, and record it as the daily abnormal data value. Step S104: Set the threshold for the mutation data value of the load sensing type, record the daily abnormal data values ​​that exceed the threshold as historical sudden change values, and then mark the corresponding historical abnormal load sensing data as historical sudden load sensing data. Step S105: Conversely, no processing is performed.

[0020] Further clarification is needed for steps S101-S105. The historical data collection period can be set to certain historical years, certain historical data collection time periods, certain historical months, certain historical days, etc. The historical data collection period can be a continuous historical data collection period or a non-continuous historical data collection period. Further, the load awareness type is used to represent the data type of awareness corresponding to all load types on the server. The load types include CPU load, memory load, IO load, network load, and business load. For example, if the load type is CPU load, the corresponding load awareness types include, but are not limited to, CPU utilization, run queue, number of interrupts, etc.; if the load type is business load, the corresponding load awareness types include, but are not limited to, QPS, number of concurrent connections, request wait queue length, 99th percentile response latency, message queue backlog, etc. Furthermore, the maximum historical anomaly standard data and the minimum historical anomaly standard data are used to represent the maximum and minimum values ​​of the historical anomaly standard data corresponding to different load sensing types during the historical data collection period, respectively. The historical anomaly standard data is used to represent the anomaly thresholds corresponding to different load sensing types during the historical data collection period. The anomaly threshold may be a single value or a range. If it is a range, the upper and lower limits of the range are split into two values ​​to obtain the maximum and minimum values. Then, the difference between the two (the difference between the maximum and minimum historical anomaly standard data) is obtained. The historical anomaly standard data difference is used to represent a benchmark value for a sudden change in the load sensing type, since the benchmark value corresponding to the same historical data collection period is the same. Furthermore, the daily anomaly data value is used to represent the ratio of historical anomaly load sensing data to the historical anomaly standard data difference as the benchmark value. Historical sudden load sensing data is obtained through the sudden change data value threshold, which can be set through specific implementation. Furthermore, anomalies are not the same as sudden anomalies. Therefore, by using the difference between historical anomaly standard data as a benchmark to determine whether historical anomaly load perception data is historical sudden load perception data, we can better process sudden load perception data.

[0021] It should be further explained that the process of obtaining the load-aware type of burst data network based on historical burst load-aware data includes: Step S106: Obtain other load perception types with key correlation characteristics for each load perception type, and denot them as associated load perception types; associate the load perception types with the associated load perception types to generate an associated load data network; Step S107: Obtain the historical associated load perception data of each associated load perception type of the associated load data network in the historical abnormal load perception data pool, as well as the historical associated burst load perception data of the corresponding historical burst load perception data. Step S108: Merge the historical associated load sensing data to generate the historical daily abnormal change range of the associated load sensing type; similarly, merge the historical associated sudden load sensing data to generate the historical sudden abnormal change range of the associated load sensing type. Step S109: Obtain the upper limit of the historical daily abnormal change range, and record it as the historical daily abnormal change upper limit. Then, combine it with the historical sudden abnormal change range to form the sudden data range of the corresponding associated load perception type. Step S110: Set corresponding burst signals in the burst data range and form burst signal data with the corresponding burst data range; then map each burst signal data to each associated load sensing type of the corresponding associated load data network to form a burst data network of load sensing type.

[0022] Further clarification is needed for steps S106-S110. The "key correlation feature" is used to represent other load perception types that have both key and correlation features with the load perception type. The correlation indicates that other load perception types change with the load perception data of the load perception type, thus establishing a causal relationship. The key feature indicates other load perception types with significant correlation. For example, when QPS, 99th percentile latency, and request queue length suddenly increase, CPU computational pressure will inevitably increase, and CPU utilization will rise significantly. Further, all historical correlation load perception data and all historical correlation mutation load perception data are obtained from the historical abnormal load perception data pool. All historical correlation load perception data are fused to obtain the historical daily abnormal change range. Similarly, all historical correlation mutation load perception data are fused to obtain the historical mutation abnormal change range. The fusion includes, but is not limited to, quantile statistical fusion, normal distribution fusion, weighted fusion, etc., which will not be elaborated here. The main purpose of the fusion is to combine the load perception data corresponding to the key correlation features of the two different abnormal types in the historical abnormal load perception data pool to obtain the corresponding change range. Furthermore, the upper limit a1 of the historical daily abnormal change range is less than or equal to the lower limit a2 of the historical sudden abnormal change range; if a1 < a2, then the sudden data range of the corresponding associated load perception type is denoted as {a1, a3}; if a1 = a2, then it is denoted as {a2, a3}, where a3 represents the upper limit of the historical sudden abnormal change range.

[0023] Further clarification is needed for step S2, which involves constructing a burst load sensing signal model for a burst data network, including: Step S201: Divide the burst data range evenly into three burst level ranges, which are respectively denoted as the first burst level range, the second burst level range, and the third burst level range; Step S202: Set the suddenness level weight corresponding to the suddenness level range according to the different keyness of the key correlation features; Step S203: Map the burst level range and burst level weight to the various associated load sensing types of the burst data network, and then construct the burst load sensing signal model.

[0024] Further limitations are needed for steps S201-S203. Different key correlation features correspond to different correlation weights, which in turn lead to different burst level weights for the burst level ranges. For example, the key correlation feature corresponding to the load perception data is the first key correlation feature; therefore, the correlation weight is set to 0.6, and the sum of the three burst level weights corresponding to the three burst level ranges is 0.6. Depending on the level, the weights for the three burst levels can be set to 0.4, 0.15, and 0.05.

[0025] It should be further explained that the process of real-time acquisition of burst load sensing data based on the burst load sensing signal model includes: Step S204: Set the acquisition period; set temporary acquisition points in the acquisition period, and then set up a graded signal acquisition unit and a trigger acquisition unit at the temporary acquisition node; Step S205: The graded signal acquisition unit schedules the current associated load perception data collected according to the acquisition cycle based on the burst load perception signal model and the associated load perception type corresponding to the load perception type. Step S206: Determine whether there is a burst data range corresponding to the load perception type in the current associated load perception data; Step S207: If it exists, trigger a burst signal to obtain the burst level range of the current associated load perception data and the corresponding burst level weight. Step S208: Calculate the load burst signal data of the load perception type by taking the current associated load perception data of all corresponding associated load perception types according to the corresponding burst level weight; The setup process for triggering the acquisition unit includes: Step S2041: The trigger acquisition unit is configured with a level trigger signal data range mapping set, and obtains the trigger level signal of the load burst signal data according to the level trigger signal data range mapping set; and the start signal of the trigger level signal; the start signal includes yes and no; Step S2042: If the start signal is yes, then the temporary acquisition point is directly triggered to collect burst load sensing data of the load sensing type; Step S2043: If the start signal is not, then determine whether the corresponding trigger level signal is a waiting trigger level signal according to the preset range of waiting level signals; Step S2044: If yes, set the temporary collection period of the temporary collection point, collect the load burst signal data within the collection period corresponding to the load perception type in real time according to the temporary collection period, and then determine whether the start signal triggers the temporary collection point to collect the burst load perception data of the load perception type until the next collection time point is reached according to the collection period. Step S2045: If not, no action is taken.

[0026] Further clarification is needed for steps S204-S208 and S2042-S2044. The scheduling process of the current associated load sensing data includes: collecting current load sensing data for all load sensing types according to the collection cycle; then scheduling the current load sensing data corresponding to the associated load sensing types based on the burst data network of the burst load sensing signal model, and recording it as the current associated load sensing data. Further, the calculation of the burst load signal data is performed using a weighted fusion algorithm, which will not be elaborated here. Further, the level trigger signal data range mapping set is used to store the load burst signal data threshold ranges corresponding to different trigger level signals and the mapped start signals; then, the load burst signal data is matched with the corresponding load burst signal data threshold range to obtain the corresponding trigger level signal, and then the corresponding start signal is obtained. Furthermore, the waiting trigger level signal indicates that the load is in a pre-burst state and is about to enter the burst data range. At this time, a temporary acquisition cycle is automatically configured for the temporary acquisition point. The temporary acquisition cycle is within the acquisition cycle and is faster than the cycle. Then, high-frequency continuous acquisition of load burst signal data is performed, constantly verifying whether the start signal meets the trigger conditions. If it does, burst load perception data is acquired; otherwise, no processing is done. This continues until one acquisition cycle is completed and the next acquisition cycle begins, repeating in this cycle. Furthermore, this process uses the acquisition cycle to determine whether the start signal is for the first acquisition of burst load perception data, improving the timeliness of burst load perception type acquisition. Then, the range of the pending occurrence level signal is used to determine whether it is a waiting trigger level signal, allowing for the second acquisition of burst load perception data within the acquisition cycle, avoiding omissions in the acquisition of burst load perception data during burst situations.

[0027] Step S3 requires further clarification. The process of constructing the power state model for burst sensing data includes: Step S301: Set the power consumption weight of the load awareness type, and obtain the power consumption of each load awareness type according to the dynamic rated power consumption of the server based on the power consumption weight. Step S302: Obtain the number of load-aware type connections associated with the load data network, and record it as the number of connections; then divide the power consumption of the load-aware type equally according to the number of connections, and record it as the distributed power consumption of the load-aware type in the corresponding burst data network; Step S303: Based on the burst level weight allocation of the burst load sensing type when acquiring burst load sensing data, obtain the corresponding distributed power consumption and usage power consumption, and then obtain the difference between distributed power consumption and usage power consumption, which is recorded as idle power consumption; and generate the corresponding idle state of idle power consumption; the idle state includes non-idle, 0 idle and idle; and then construct the power consumption state model of burst sensing data.

[0028] Further limitations are needed for steps S301-S303. The power consumption weight represents the proportion of dynamic rated power consumption occupied by the load-aware type. The setting principle can be based on the number and criticality of the burst data networks formed by associated load-aware types; and the sum of all power consumption weights is 1. The dynamic rated power consumption represents the maximum total dynamic power consumption that the entire machine can allocate under full load conditions; thus, the load-aware type power consumption of each load-aware type is obtained. Furthermore, the load-aware type can be a load-aware type associated with a load data network, or a associated load-aware type of another associated load data network; therefore, the distributed power consumption of the current corresponding burst data network is obtained based on the number of connected burst data networks; power consumption is managed according to the distribution of burst data networks. Furthermore, the corresponding distributed power consumption is obtained according to the burst level weight allocation. For example, if the distributed power consumption is 90W, the sum of the three burst level weights of the associated load perception type is 0.6, and the corresponding burst level weight is 0.4, then the corresponding power consumption is 90W ÷ 0.6 × 0.4 = 60W, and the idle power consumption is 90W - 60W = 30W. Furthermore, "non-idle" indicates that the idle power consumption is negative, the quota is overdrawn, and other idle power consumption needs to be scheduled; "0 idle" indicates that the idle power consumption is 0; "idle" indicates that the idle power consumption is positive, there is idle power consumption, and it can be scheduled.

[0029] It should be further explained that the process of obtaining the emergency demand power consumption corresponding to the burst load perception data by dynamically adjusting the power consumption based on the power consumption state model and the burst load perception data includes: Step S304: Obtain the current load power consumption of the burst load sensing data, and obtain the difference between the distributed power consumption of the corresponding load sensing data and the current load power consumption based on the power consumption state model, which is recorded as the emergency demand power consumption. Step S305: Determine whether the power consumption for emergency needs is less than 0; Step S306: If the emergency demand power consumption is less than 0, generate an emergency demand power consumption signal from the burst load sensing data and send it to the idle associated load sensing type in the burst data network to obtain the corresponding idle power consumption. Step S307: Schedule all idle power consumption to generate schedulable idle power consumption, and then update the corresponding idle state to 0 idle. Step S308: Determine whether the schedulable idle power consumption is greater than the emergency demand power consumption; Step S309: If not, then record the difference between the emergency power consumption and the schedulable idle power consumption as the emergency power consumption required corresponding to the burst load perception data. Step S310: If yes, then the difference between the schedulable idle power consumption and the emergency required power consumption is recorded as the remaining schedulable idle power consumption corresponding to the burst load perception data. Step S311: If the emergency demand power consumption is greater than or equal to 0, then mark the emergency demand power consumption as the remaining distributed power consumption.

[0030] Further clarification is needed for steps S304-S311, whereby the dynamic power consumption corresponding to the sudden load perception data is divided into remaining urgent demand power consumption, remaining schedulable idle power consumption, and remaining distributed power consumption based on the urgent demand power consumption and schedulable idle power consumption. The initial dynamic power consumption adjustment of the sudden load perception data is used to obtain the remaining urgent demand power consumption, providing data for subsequent secondary power compensation adjustment.

[0031] Step S4 requires further clarification. The process of dynamically adjusting power consumption for urgent power demands includes: Step S401: Obtain the schedulable idle power, remaining schedulable idle power, and remaining distributed power of the load sensing type corresponding to the first burst load sensing data connection to other associated load data networks in ascending order of the still urgent demand power consumption; and sum them up to record as the total schedulable power consumption. Step S402: Determine whether the total schedulable power consumption is greater than or equal to the power consumption required for urgent needs; Step S403: If yes, then allocate the total schedulable power consumption to the corresponding urgent power consumption, and obtain the remaining total schedulable power consumption, thereby completing the power consumption scheduling for the urgent power consumption. Step S404: If not, enter the waiting area; then continue to obtain the total schedulable power consumption corresponding to the next burst load sensing data; Step S405: Determine whether the total schedulable power consumption is greater than or equal to the power consumption required for urgent needs; Step S406: If yes, then obtain the remaining total schedulable power consumption, and then complete the power consumption scheduling for the urgent power demand. Step S407: If not, enter the waiting area; then continue to obtain the total schedulable power consumption corresponding to the next burst load perception data; Step S408: By analogy, obtain all remaining total schedulable power consumption; and sum them up to record as the remaining schedulable power consumption; Step S409: If the available schedulable power consumption is greater than or equal to the power consumption of any of the waiting areas that still require urgent power consumption, then the available schedulable power consumption is sequentially scheduled to the power consumption that is greater than or equal to the power consumption of any of the waiting areas that still require urgent power consumption, thereby completing the power consumption scheduling of the power consumption of any of the waiting areas that still require urgent power consumption; until the available schedulable power consumption is less than the power consumption of any of the waiting areas that still require urgent power consumption, then the scheduling stops. Step S410: Conversely, no processing is performed; Step S411: Perform other power scheduling on the still urgent power consumption in the waiting area, thereby completing the power scheduling of the still urgent power consumption in the waiting area.

[0032] Further clarification is needed for steps S401-S411, which involve compensating for the power consumption of those still requiring urgent demands, thereby completing all power control of the corresponding burst load perception data, utilizing all dynamically available idle power, and improving utilization. The power consumption of those still in the waiting area requiring urgent demands will be scheduled using other power scheduling methods; these other power scheduling methods include, but are not limited to, hardware sleep power wake-up scheduling, steady-state load power recovery scheduling, etc.

[0033] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A server dynamic power consumption control method based on a real-time load awareness algorithm, characterized in that, The method includes the following steps: Step S1: Obtain historical burst load perception data; obtain burst data network of load perception type based on historical burst load perception data; Step S2: Construct a burst load sensing signal model for the burst data network; collect burst load sensing data in real time based on the burst load sensing signal model; Step S3: Construct a power consumption state model for burst sensing data; based on the power consumption state model, perform dynamic power consumption control on the burst load sensing data to obtain the power consumption required for urgent needs corresponding to the burst load sensing data; Step S4: Perform dynamic power consumption control for urgent power needs.

2. The server dynamic power consumption control method based on real-time load awareness algorithm according to claim 1, characterized in that, The process of acquiring historical burst load awareness data includes: Obtain historical abnormal load perception data pools for different load perception types of servers within a preset historical collection time period; obtain the maximum and minimum historical abnormal standard data in the historical abnormal load perception data pool; and obtain the difference between the two, which is denoted as the historical abnormal standard data difference. Obtain the ratio of the difference between each historical abnormal load perception data and the historical abnormal standard data in the historical abnormal load perception data pool, and record it as the daily abnormal data value. Set a threshold for the mutation data value of the load sensing type, record daily abnormal data values ​​that exceed the threshold as historical sudden change values, and then mark the corresponding historical abnormal load sensing data as historical sudden load sensing data. Conversely, no action is taken.

3. The method for dynamic power consumption regulation of server based on real-time load-aware algorithm according to claim 2, characterized in that, The process of obtaining burst data networks of load-aware type based on historical burst load-aware data includes: Obtain other load perception types that have key correlation characteristics with each load perception type, and denot them as associated load perception types; associate the load perception types with the associated load perception types to generate an associated load data network; Obtain historical associated load perception data of each associated load perception type in the historical abnormal load perception data pool, as well as historical associated burst load perception data of the corresponding historical burst load perception data. By fusing various historical associated load sensing data, the historical daily anomaly change range of associated load sensing type is generated; similarly, by fusing various historical associated sudden load sensing data, the historical sudden anomaly change range of associated load sensing type is generated. Obtain the upper limit of the range of historical daily abnormal changes, and denot it as the upper limit of historical daily abnormal changes. Then, combine it with the range of historical sudden abnormal changes to form the range of sudden data of the corresponding associated load perception type. A corresponding burst signal is set within the burst data range, and burst signal data is formed with the corresponding burst data range; then, each burst signal data is mapped to each associated load perception type of the corresponding associated load data network to form a burst data network of the load perception type.

4. The method for dynamic power consumption regulation of server based on real-time load-aware algorithm according to claim 3, characterized in that, The process of constructing a burst load sensing signal model for a burst data network includes: The scope of the sudden data is evenly divided into three sudden level ranges; the sudden level weights are set according to the different criticality of the key correlation features; The range and weight of the suddenness level are uniformly mapped to the various associated load sensing types of the suddenness data network, thereby constructing a suddenness load sensing signal model.

5. The method for dynamic power consumption regulation of server based on real-time load-aware algorithm according to claim 4, characterized in that, The process of real-time acquisition of burst load sensing data based on the burst load sensing signal model includes: Set the acquisition period; set temporary acquisition points within the acquisition period, and then set up graded signal acquisition units and trigger acquisition units at the temporary acquisition nodes; The graded signal acquisition unit schedules the current associated load perception data collected according to the acquisition cycle based on the burst load perception signal model and the corresponding associated load perception type. Determine whether the current associated load perception data contains a burst data range corresponding to the associated load perception type; If it exists, a burst signal is triggered, thereby obtaining the burst level range of the current associated load perception data and the corresponding burst level weight; The load burst signal data of the load perception type is obtained by calculating the current associated load perception data of all corresponding associated load perception types according to the corresponding burst level weight.

6. The method for dynamic power consumption regulation of server based on real-time load-aware algorithm according to claim 5, wherein, The setup process for triggering the acquisition unit includes: The trigger acquisition unit is equipped with a level trigger signal data range mapping set, and obtains the trigger level signal of the load burst signal data according to the level trigger signal data range mapping set; and the start signal of the trigger level signal; the start signal includes yes and no; If the start signal is yes, then the temporary collection point will be directly triggered to collect burst load sensing data of the load sensing type; If the start signal is not present, the corresponding trigger level signal will be determined according to the preset range of pending trigger level signals to determine whether it is a waiting trigger level signal. If so, set the temporary collection period for the temporary collection point, and collect the load burst signal data within the collection period corresponding to the load perception type in real time according to the temporary collection period. Then determine whether the start signal triggers the temporary collection point to collect the burst load perception data of the load perception type until the next collection time point is reached according to the collection period. If not, no action will be taken.

7. The method for dynamic power consumption regulation of servers based on real-time load-aware algorithm according to claim 6, wherein, The process of constructing a power state model for burst sensing data includes: Set the power consumption weight for load awareness type, and allocate the dynamic rated power consumption of the server according to the power consumption weight to obtain the power consumption of each load awareness type. Obtain the number of load-aware type connections associated with the load data network, and record it as the number of connections; then divide the power consumption of the load-aware type equally according to the number of connections, and record it as the distributed power consumption of the load-aware type in the corresponding burst data network; Based on the burst level weight allocation corresponding to the burst level of the load sensing type when acquiring burst load sensing data, the power consumption is obtained by distributing the power consumption and then the difference between the power consumption and the power consumption is obtained and recorded as idle power consumption; and the idle state corresponding to the idle power consumption is generated; the idle state includes non-idle, 0 idle and idle; and then the power consumption state model of burst sensing data is constructed.

8. The method for dynamic power consumption regulation of server based on real-time load-aware algorithm according to claim 7, characterized in that, The process of dynamically adjusting power consumption based on the power consumption state model to obtain the power consumption required for urgent needs corresponding to the burst load perception data includes: Obtain the current load power consumption of the burst load sensing data, and obtain the difference between the distributed power consumption of the corresponding load sensing data and the current load power consumption based on the power consumption state model, which is recorded as the emergency demand power consumption. Determine if the power consumption for urgent needs is less than 0; If the emergency power demand is less than 0, the burst load perception data will be used to generate an emergency power demand signal and sent to the idle associated load perception type in the burst data network to obtain the corresponding idle power. All idle power consumption is scheduled to generate schedulable idle power consumption, and then the corresponding idle state is updated to 0 idle. Determine whether the schedulable idle power consumption is greater than the power consumption required by the emergency. If not, the difference between the emergency power consumption and the schedulable idle power consumption is recorded as the emergency power consumption required corresponding to the burst load perception data. If so, the difference between the schedulable idle power consumption and the emergency required power consumption is recorded as the remaining schedulable idle power consumption corresponding to the burst load perception data. If the emergency demand power consumption is greater than or equal to 0, then the emergency demand power consumption is marked as the remaining distributed power consumption.

9. The server dynamic power consumption control method based on real-time load awareness algorithm according to claim 8, characterized in that, The process of dynamically regulating power consumption for urgent needs includes: According to the order of the still urgent demand power consumption from smallest to largest, obtain the first burst load sensing data corresponding to the load sensing type, connect other related load data networks, and obtain the schedulable idle power consumption, remaining schedulable idle power consumption, and remaining distributed power consumption; and sum them up to record the total schedulable power consumption. Determine whether the total schedulable power consumption is greater than or equal to the power consumption required for urgent needs. If so, the total schedulable power consumption will be allocated to the corresponding urgent power consumption, and the remaining total schedulable power consumption will be obtained to complete the power consumption scheduling for the urgent power consumption. If not, it enters the waiting area; and then continues to acquire the total schedulable power consumption corresponding to the next burst load perception data; Determine whether the total schedulable power consumption is greater than or equal to the power consumption required for urgent needs. If so, then obtain the remaining total schedulable power consumption, and then complete the power consumption scheduling for the urgent power demand. If not, it enters the waiting area; and then continues to acquire the total schedulable power consumption corresponding to the next burst load perception data; Similarly, obtain all remaining total schedulable power consumption; and sum them up to record the remaining schedulable power consumption. If the available schedulable power is greater than or equal to the power of any of the waiting regions that still require urgent power, then the available schedulable power will be scheduled sequentially to the power of any of the waiting regions that still require urgent power, thus completing the power scheduling for the power of any of the waiting regions that require urgent power; the scheduling will stop when the available schedulable power is less than the power of any of the waiting regions that require urgent power. Otherwise, no action will be taken. Other power scheduling will be carried out on the power consumption of the still urgent needs in the waiting area, thereby completing the power scheduling of the still urgent needs in the waiting area.

Images