Power supply system, power supply management method, computer program product, and storage medium

Abstract

The invention discloses a power supply system, a power supply management method, a computer program product and a storage medium, and is applied to the technical field of servers, the power supply system comprises a plurality of power supply modules for supplying power to a server cluster and a power supply management device, the power supply management device comprises a sensing device for monitoring the load state of each server in the server cluster; the control device is used for determining a load prediction value of each server after the first duration; adjusting the power distribution of the power supply system to enable the total error of the power distribution to be lower than a preset first threshold value, and enabling each server in the server cluster to at least receive power supplied by K power supply modules; the total power distribution error represents a sum of power distribution errors of each server in the server cluster. By means of the scheme, power management can be effectively achieved, the energy utilization efficiency is improved, and the stability of server operation is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of servers, and in particular to a power supply system, a power supply management method, a computer program product and a storage medium. Background Art

[0002] With the continuous development of information technology, the status of servers is becoming increasingly important. In data centers and other places, servers often need to run uninterruptedly for a long time to support the processing and storage of massive data, so the performance of the power system is crucial.

[0003] At present, common server power supply systems mainly adopt a fixed power distribution method. Specifically, the power distribution plan of the server cluster will be designed in advance based on the maximum power consumption of each server in the server cluster. This static power distribution mode is simple and easy to implement. However, in actual applications, the load of the server will change dynamically with the different running tasks, resulting in energy waste in many cases with this fixed power distribution method, especially when the server cluster as a whole is in a low-load state, the preset power supply is often much higher, resulting in the ineffective use of excess power. In addition, in the current solution, during some peak load periods, it may not be able to meet the needs of the server, resulting in a decline in system performance and even affecting the stability of server operation.

[0004] In summary, how to effectively implement power management, improve energy utilization efficiency, and ensure the stability of server operation is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] The purpose of the present invention is to provide a power supply system, a power supply management method, a computer program product and a storage medium to effectively implement power supply management, improve energy utilization efficiency and ensure the stability of server operation.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a power supply system, comprising: a plurality of power supply modules for supplying power to a server cluster, and a power supply management device connected to each of the power supply modules; the power supply management device comprises:

[0008] A sensor device for monitoring the load status of each server in the server cluster;

[0009] A control device connected to the sensor device is used to determine the load prediction value of each server after a first period of time based on the load state of each server; adjust the power distribution of the power supply system so that the total power distribution error is lower than a preset first threshold value, and each server in the server cluster receives power from at least K power modules;

[0010] Among them, the starting time of the first time period is the time when the load status monitoring of each server is completed; K is a positive integer not less than 2, the total power allocation error represents the sum of the power allocation errors of each server in the server cluster, and the power allocation error of a single server represents the error between the total power obtained by the server from the power supply system and the load prediction value of the server.

[0011] On the other hand, it also includes:

[0012] A fault monitoring device, used to monitor the fault status of each power module in the power system and send it to the control device;

[0013] The control device is connected to the fault monitoring device, and is also used to execute a preset fault response operation when the fault monitoring device detects a fault in the power module.

[0014] On the other hand, the power supply system also includes a backup power supply module;

[0015] Accordingly, the control device executes a preset fault response operation, including:

[0016] Determining the number of the power modules that have failed;

[0017] From the standby power modules, an equal number of standby power modules are selected to take over the work of the failed power modules, and the power distribution of the power system is adjusted.

[0018] On the other hand, the control device adjusts the power distribution of the power supply system, including:

[0019] The control device adjusts the power distribution of the power supply system according to the preset constraint rules;

[0020] The constraint rule includes: for any power module in the power system, the total power provided by the power module shall not exceed the maximum power output limit value of the power module.

[0021] On the other hand, the constraint rules also include:

[0022] The total power obtained by a single server from the power supply system is greater than A times the load prediction value of the server; wherein A is a preset multiplier value and A is greater than 1.

[0023] On the other hand, it also includes a communication device, and the communication device is connected to the control device;

[0024] The control device is also used to: receive a power distribution instruction through the communication device, and adjust the power distribution of the power supply system according to the power distribution instruction.

[0025] On the other hand, the sensor device is specifically used for:

[0026] Monitor the load status of each server in the server cluster within the most recent second period of time;

[0027] Correspondingly, the control device determines the load prediction value of each server after the first time period based on the load status of each server, including:

[0028] The control device determines a load prediction value of each of the servers after the first time period based on the load status of each of the servers within the most recent second time period.

[0029] On the other hand, the sensor device is specifically used for:

[0030] Monitor each server in the server cluster, and for any one of the servers, determine a central processing unit utilization sequence for reflecting changes in central processing unit utilization of the server within a recent second time period, a memory utilization sequence for reflecting changes in memory occupancy of the server within a recent second time period, a request quantity sequence for reflecting changes in the number of requests of the server within a recent second time period, and a network bandwidth sequence for reflecting changes in the network bandwidth of the server within a recent second time period, and use the central processing unit utilization sequence, the memory utilization sequence, the request quantity sequence and the network bandwidth sequence as the monitored load status of the server within the recent second time period.

[0031] On the other hand, the control device determines the load prediction value of each of the servers after the first time period based on the load state of each of the servers in the most recent second time period, including:

[0032] For any one of the servers in the server cluster, the load state of the server is input into a pre-trained first model to obtain a load prediction result output by the first model that reflects the load state of the server after a first period of time; wherein the load prediction result includes the CPU utilization, memory occupancy, number of requests and network bandwidth of the server after the first period of time;

[0033] The load prediction result is input into a pre-trained second model to obtain a load prediction value output by the second model as the determined load prediction value of the server after the first period of time.

[0034] On the other hand, the control device determines the load prediction value of each server after the first time period based on the load state of each server in the most recent second time period, including:

[0035] For any one of the servers in the server cluster, based on the load state of the server in the most recent second time period, N pending prediction values ​​of the server after the first time period are determined by N different calculation methods; wherein N is a positive integer not less than 2;

[0036] Based on the N pending prediction values, a load prediction value of the server after a first period of time is determined.

[0037] On the other hand, based on the N pending prediction values, determining the load prediction value of the server after the first duration includes:

[0038] Based on N pending prediction values ​​and respective weight parameters of N different calculation methods, the load prediction value of the server after the first time period is obtained by weighted superposition of the N pending prediction values.

[0039] On the other hand, the control device is also used for:

[0040] For any server, determining the input power value currently required by the server;

[0041] Determining whether an error between the input power value and the load prediction value of the server exceeds a preset error range;

[0042] If yes, the input power value currently required by the server is used to replace the load prediction value of the server, and an operation of adjusting the power distribution of the power supply system that supplies power to the server cluster is performed.

[0043] In a second aspect, the present invention provides a power management method, comprising:

[0044] Monitor the load status of each server in the server cluster;

[0045] Based on the load status of each of the servers, a load prediction value of each of the servers after a first time period is determined; wherein the start time of the first time period is the time when the load status monitoring of each server is completed;

[0046] Adjusting power distribution of a power supply system that supplies power to the server cluster so that a total power distribution error is lower than a preset first threshold and each of the servers in the server cluster receives power from at least K power modules;

[0047] In which, the power supply system includes multiple power supply modules, K is a positive integer not less than 2, the total power distribution error represents the sum of the power distribution errors of each server in the server cluster, and the power distribution error of a single server represents the error between the total power obtained by the server from the power supply system and the load prediction value of the server.

[0048] In a third aspect, the present invention provides a computer program product, comprising a computer program / instruction, which implements the steps of the power management method as described above when executed by a processor.

[0049] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the power management method as described above are implemented.

[0050] The technical solution provided by the embodiment of the present invention is applied, considering that the traditional fixed power distribution method is prone to energy waste and will affect the stability of server operation. In this regard, in the solution of the present application, dynamic adjustment of power distribution will be performed. Specifically, in the solution of the present application, it is necessary to monitor the load status of each server in the server cluster, and then determine the load prediction value of each server after the first time period at the current moment based on the load status of each server. This is to take into account that the load will change, and determine the load prediction value of each server after the first time period, thereby realizing dynamic power distribution, which is conducive to ensuring the rationality of power distribution, and then conducive to ensuring the stability of server operation.

[0051] When adjusting the power distribution of the power supply system that supplies power to the server cluster, it is necessary to make the total power distribution error lower than the preset first threshold. The total power distribution error represents the sum of the power distribution errors of each server in the server cluster. In other words, making the total power distribution error lower than the preset first threshold means that the power distribution of each server is consistent with the load forecast value of the server. This means that when it is predicted that the future load of the server will increase, the power supply can be increased in advance, thereby avoiding voltage instability or system performance degradation caused by the increase in load. Conversely, when it is predicted that the future load of the server will decrease, the power supply can be reduced in advance to prevent power waste and improve energy efficiency.

[0052] Furthermore, in order to ensure the stability of power supply, each server in the server cluster needs to receive power from at least K power modules, where K is a positive integer not less than 2. This means that each server will receive power from 2 or more power modules. In this way, when a single power module supplying power to the server fails, the server will not immediately lose all power supply, which is also beneficial to ensure the stability of server operation to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A flowchart of a power management method according to a specific embodiment of the present invention;

[0055] Figure 2 A schematic diagram of the structure of a power supply system provided in a specific embodiment of the present invention;

[0056] Figure 3 A schematic diagram of the structure of a device provided by a specific embodiment of the present invention;

[0057] Figure 4 The present invention is a schematic diagram of the structure of a computer-readable storage medium. DETAILED DESCRIPTION

[0058] The core of the present invention is to provide a power supply system, a power supply management method, a computer program product and a storage medium, which can effectively realize power supply management, improve energy utilization efficiency and ensure the stability of server operation.

[0059] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] Please refer to Figure 1 and Figure 2 , Figure 1 A flowchart of a power management method according to a specific embodiment of the present invention is provided. Figure 2 A schematic diagram of the structure of a power supply system provided in a specific embodiment of the present invention, the power supply system may include: a plurality of power supply modules for supplying power to a server cluster, and a power supply management device connected to each power supply module; the power supply management device includes:

[0061] A sensor device for monitoring the load status of each server in the server cluster;

[0062] A control device connected to the sensor device is used to determine the load prediction value of each server after a first period of time based on the load state of each server; adjust the power distribution of the power supply system so that the total power distribution error is lower than a preset first threshold value, and each server in the server cluster receives power from at least K power modules;

[0063] Among them, the starting time of the first time period is the time when the load status monitoring of each server is completed; K is a positive integer not less than 2, the total power distribution error represents the sum of the power distribution errors of each server in the server cluster, and the power distribution error of a single server represents the error between the total power obtained by the server from the power supply system and the load prediction value of the server.

[0064] Specifically, the sensor device can monitor the load status of each server in the server cluster. The load status of the server can be measured by multiple indicators. The specific measurement method can be set and adjusted according to actual needs. In addition, the load status of a certain server can be measured by multiple indicators at the current moment, or based on multiple indicators within a certain period of time. It does not affect the implementation of the present invention, and can be set and selected according to actual needs.

[0065] In a specific implementation of the present invention, the sensor device is specifically used to monitor the load status of each server in the server cluster within the most recent second time period.

[0066] This implementation takes into account that monitoring the load status of each server in the server cluster is to be able to determine the load prediction value of each server after the first time period at the current moment. Therefore, if the load status of each server in the recent period of time is monitored, it will be beneficial to obtain accurate load prediction values ​​later. Therefore, in this implementation, what is specifically monitored is the load status of each server in the server cluster in the most recent second time period.

[0067] Further, in a specific implementation of the present invention, monitoring the load status of each server in the server cluster within the most recent second period of time may specifically include:

[0068] Monitor each server in the server cluster, and for any server, determine a central processing unit utilization sequence for reflecting changes in central processing unit utilization of the server within a recent second time period, a memory utilization sequence for reflecting changes in memory occupancy of the server within a recent second time period, a request quantity sequence for reflecting changes in the number of requests of the server within a recent second time period, and a network bandwidth sequence for reflecting changes in the network bandwidth of the server within a recent second time period, and use the central processing unit utilization sequence, memory utilization sequence, request quantity sequence, and network bandwidth sequence as the load status of the monitored server within the recent second time period.

[0069] This implementation method takes into account that, for a server, the CPU (Central Processing Unit), memory, I / O (Input / Output) request quantity and network bandwidth will all affect the load of the server. In other words, the load of the server will be reflected in the CPU, memory, IO quantity and network bandwidth. Therefore, in this implementation method, the central processing unit utilization sequence, memory occupancy sequence, request quantity sequence and network bandwidth sequence will be used as the load status of the monitored server in the most recent second time period.

[0070] The CPU utilization sequence reflects the CPU utilization change of the server in the most recent second duration. For example, the CPU utilization of the server can be periodically monitored and arranged in chronological order to obtain a CPU utilization sequence corresponding to the most recent second duration. Similarly, the memory occupancy of the server can be periodically monitored and arranged in chronological order to obtain a memory occupancy sequence corresponding to the most recent second duration. The IO quantity of the server, that is, the number of requests, can be periodically monitored and arranged in chronological order to obtain a request quantity sequence corresponding to the most recent second duration. The network bandwidth of the server can be periodically monitored and arranged in chronological order to obtain a network bandwidth sequence corresponding to the most recent second duration.

[0071] In addition, it can be seen that in this implementation, the load changes of the server in the most recent second time period can be effectively reflected in the form of sequences, namely, the CPU utilization sequence, the memory occupancy sequence, the request quantity sequence and the network bandwidth sequence. When these sequences are used as the load status of the server in the most recent second time period, a more accurate load prediction value can be obtained subsequently.

[0072] The control device connected to the sensor device can determine the load prediction value of each server after the first time period based on the load status of each server. The load prediction value reflects the power required by the server after the first time period at the current moment. The current moment described here is Figure 1 At the triggering moment of step S102, the monitoring of the load status of each server in the server cluster has been completed. Therefore, based on the monitoring result, that is, based on the load status of each server, the load forecast value of each server at a certain moment in the future (after the first period of time at the current moment) can be predicted.

[0073] In one embodiment of the above, when monitoring the load status of each server in the server cluster, what is monitored is the load status of each server in the server cluster within the second time period most recent to the current moment. Therefore, in this embodiment, the control device can determine the load prediction value of each server after the first time period at the current moment based on the load status of each server within the second time period most recent to the current moment.

[0074] In a specific implementation of the present invention, the control device determines the load prediction value of each server after the first time period based on the load state of each server in the most recent second time period, which may specifically include:

[0075] For any server in the server cluster, the load state of the server is input into the pre-trained first model to obtain a load prediction result output by the first model that reflects the load state of the server after a first period of time; wherein the load prediction result includes the CPU utilization rate, memory occupancy rate, number of requests and network bandwidth of the server after the first period of time;

[0076] The load prediction result is input into a pre-trained second model to obtain a load prediction value output by the second model as the determined load prediction value of the server after the first period of time.

[0077] This implementation method takes into account that first, Figure 1 The load status of each server obtained in step S101 is used to predict the load status of each server after the first period of time, and then the load prediction value of each server is obtained according to the predicted load status of each server after the first period of time, that is, the power required by each server is obtained.

[0078] In this regard, in this implementation, the first model is a model that can predict the load state of the server at a certain moment in the future based on the load state of the server over a period of time in the past. Therefore, for any server in the server cluster, the load state of the server can be first input into the pre-trained first model. Considering that the load state of the server input into the first model can usually include the CPU status, memory status, IO quantity and network bandwidth of the server in the most recent second time period, therefore, in this implementation, the load prediction result output by the first model can include the CPU utilization, memory occupancy, number of requests and network bandwidth of the server after the first time period.

[0079] The second model is a model that can predict the power required by the server based on the load prediction results. Therefore, after obtaining the load prediction result output by the first model, it is input into the pre-trained second model to obtain the load prediction value output by the second model as the determined load prediction value of the server after the first period of time.

[0080] In practical applications, considering that machine learning models can accurately predict load changes, the first model and the second model used in this implementation are usually prediction models based on machine learning. By learning from historical load data, the first model and the second model can meet the use requirements of this application. Commonly used machine learning models include long short-term memory network models, random forest regression models, etc. The specific model type can be set and adjusted according to actual needs.

[0081] The first model and the second model need to be trained in advance. Taking the first model as an example, the training samples are the historical load status of the server, including CPU utilization, memory occupancy, number of IO requests and network bandwidth, and they are arranged in time to obtain the corresponding sequence. The label is the load situation at a certain point in the future corresponding to the corresponding sample, specifically including CPU utilization, memory occupancy, number of IO requests and network bandwidth at a certain point in the future. The loss function can, for example, specifically use the mean square error, which can minimize the gap between the predicted value and the actual load value, and gradually optimize the performance of the model. Through multiple rounds of training, a first model that can effectively predict the load status of the server at a certain moment in the future can be obtained. The training of the second model is the same. Through effective training, a second model that can effectively predict the power required by the server based on the load status of the server can be obtained.

[0082] Further, in a specific implementation of the present invention, the control device determines the load prediction value of each server after the first time period based on the load state of each server in the most recent second time period, which may include:

[0083] For any server in the server cluster, based on the load state of the server in the most recent second time period, N pending prediction values ​​of the server after the first time period are determined by N different calculation methods; wherein N is a positive integer not less than 2;

[0084] Based on the N pending prediction values, a load prediction value of the server after the first time period is determined.

[0085] This implementation method takes into account that in addition to obtaining the load prediction value by the machine learning model, it is also possible to combine multiple analysis methods, such as an autoregressive moving average model that can effectively capture the periodic fluctuations and trend changes in the server load, to comprehensively obtain the load prediction value of the server after the first period of time.

[0086] In this regard, in this implementation, for any server in the server cluster, based on the load status of the server in the most recent second time period, N different calculation methods are used to determine N pending prediction values ​​of the server after the first time period.

[0087] For example, one of the calculation methods is the machine learning-based method described above, that is, through the first model and the second model described above, a pending prediction value of the server after the first time period is determined. Since each method can obtain one pending prediction value, a total of N pending prediction values ​​can be obtained in this implementation. For example, another calculation method uses an autoregressive moving average model. The autoregressive moving average model also needs to be trained, and the samples used can also be the historical load data of the server, but more attention can be paid to the time series characteristics in the data, such as periodic changes in load status, long-term trends, etc. In addition, the autoregressive and moving average parameters of the autoregressive moving average model can be adjusted according to actual needs, so that it can continuously optimize its prediction ability, which is particularly suitable for the prediction of long-term trends.

[0088] In other implementations, the specific implementation of N different calculation methods can be set and adjusted according to actual needs. In some cases, in order to simplify the calculation, N is set to a minimum value of 2.

[0089] When determining the load prediction value of the server after the first period of time based on N pending prediction values, there can be many specific implementation methods. For example, a simple implementation method is to directly take the average value of the N pending prediction values ​​as the determined load prediction value of the server after the first period of time.

[0090] In a specific implementation of the present invention, determining the load prediction value of the server after the first duration based on the N pending prediction values ​​may specifically include:

[0091] Based on the N pending prediction values ​​and the respective weight parameters of N different calculation methods, the load prediction value of the server after the first time period is obtained by weighted superposition of the N pending prediction values.

[0092] This implementation takes into account that respective weight parameters can be set for N different calculation methods, so that the N different calculation methods each have different influences on the load prediction value, thereby improving the flexibility of the solution. It is also understandable that if it is found that the undetermined prediction value obtained by a certain calculation method is more accurate, the weight parameter of the calculation method can be appropriately increased to further improve the accuracy of the obtained load prediction value.

[0093] Taking N=2 as an example, the load prediction value of the server after the first period can be expressed as P final=αP1+βP2. Here, P1 and P2 respectively represent the pending prediction value determined by the first calculation method and the pending prediction value determined by the second calculation method, and α and β respectively represent the weight coefficients of the two calculation methods, and α+β=1. The specific values ​​of α and β can be dynamically adjusted based on the specific performance of the two calculation methods, which ensures the flexibility of the present application scheme and is also conducive to improving the accuracy of the obtained load prediction value.

[0094] In addition to weighted superposition, the load prediction value of the server after the first time period can be determined based on other methods using N pending prediction values, such as voting method, weighted majority method, verification data set method, etc.

[0095] After the control device determines the load prediction value of each server after the first period of time, it can adjust the power distribution of the power supply system that supplies power to the server cluster so that the total power distribution error is lower than the preset first threshold and each server in the server cluster receives power from at least K power modules.

[0096] In the present application, it is necessary to dynamically distribute power to the power supply system to ensure the rationality of power distribution, which is conducive to ensuring the stability of server operation. When adjusting the power distribution of the power supply system, it can be adjusted in real time or periodically. In practical applications, considering that the server load will fluctuate, it usually does not fluctuate frequently or violently in a short period of time. Therefore, it is usually possible to adjust the power distribution of the power supply system once every period of time to avoid excessive adjustment.

[0097] Furthermore, when adjusting the power distribution of the power supply system, it is necessary to make the total power distribution error lower than the preset first threshold. The total power distribution error represents the sum of the power distribution errors of each server in the server cluster, while the power distribution error of a single server represents the error between the total power obtained by the server from the power supply system and the load prediction value of the server. In other words, dynamically distributing power to the power supply system so that the total power distribution error is lower than the preset first threshold means that the power distribution of each server can meet the load prediction value of the server. This means that for any server, if it is predicted that the load of the server will increase in the future, the power supply can be increased in advance, thereby avoiding the situation where the load increase causes voltage instability or system performance degradation. Conversely, if it is predicted that the load of the server will decrease in the future, the power supply can be reduced in advance to prevent power waste and improve energy efficiency.

[0098] In practical applications, the power distribution error of a single server can be To express, you can also use The specific form can be set and adjusted according to actual needs, and can effectively reflect the error between the total power obtained by the server from the power system and the load prediction value of the server. represents the power distribution error of a single server, where x ij It represents the power output of power module j to server i, and M is the total number of power modules currently working in the power system. is the load prediction value of server i.

[0099] If you use represents the power allocation error of a single server, then the total power allocation error can be specifically expressed as , where N represents the total number of servers. That is, in this specific implementation, after power distribution, it is required is lower than a preset first threshold. It is understandable that, ideally, Should be equal to 0.

[0100] Furthermore, in the solution of the present application, after power distribution is performed, each server in the server cluster is required to receive power from at least K power modules, where K is a positive integer not less than 2. This is because if each server receives power from two or more power modules, even if a single power module supplying power to the server fails, the server will not immediately lose all power supply, which is also beneficial to ensure the stability of server operation to a certain extent and achieves the purpose of redundancy.

[0101] In a specific implementation of the present invention, the control device adjusting the power distribution of the power system may specifically include: the control device adjusting the power distribution of the power system according to a preset constraint rule.

[0102] In this implementation, when adjusting the power distribution of the power system, the constraint rules are also required to be met. The constraint rules specifically include: for any power module in the power system, the total power provided by the power module does not exceed the maximum power output limit of the power module.

[0103] This is because in some cases, after power distribution, some power modules will provide a lot of power, while some power modules will provide less power, or even some power modules will provide power beyond the limit of the power module. Therefore, in this implementation, it is required that the power distribution is adjusted to meet the constraint rules, so that after the power distribution is completed, for any power module in the power system, the total power provided by the power module does not exceed the maximum power output limit of the power module.

[0104] For each power module, its maximum power output limit value can be determined in advance, for example, the maximum output power set when the power module leaves the factory can be directly used as the maximum power output limit value of the power module in this implementation. Of course, in some occasions, the power module may be adjusted, for example, a power amplifier is added to the output end of the power module to allow the power module to output more power, and the maximum power output limit value of each power module can be adjusted according to the actual situation.

[0105] Since in this implementation manner, the total power provided by the power modules does not exceed the maximum power output limit value of the power modules, the power supply security of each power module can be effectively guaranteed.

[0106] Furthermore, in a specific implementation of the present invention, the constraint rules may also include:

[0107] The total power obtained by a single server from the power supply system is greater than A times the load forecast value of the server; where A is a preset multiplier value and A is greater than 1.

[0108] This implementation method takes into account that in the present application, the current power distribution is based on the load prediction value of each server, that is, the current power distribution of each server is based on the future load conditions of each server. Therefore, in order to further ensure the stability of power supply, in this implementation method, the total power obtained by a single server from the power supply system is greater than A times the load prediction value of the server, so that when the load suddenly increases to a certain extent, the power supply stability of the server will not be destroyed. For example, if A is set to 120%, the constraint rules of this implementation method can be expressed as . is the load prediction value of server i, x ij It indicates the power output by power module j to server i.

[0109] For example, in a specific scenario, the power supply system is equipped with three power modules, the maximum power output limit of power module 1 is 1500W, the maximum power output limit of power module 2 is 800W, and the maximum power output limit of power module 3 is 1000W. For example, the load prediction value of server 1 is 800W, 800×1.2=960W. The load prediction value of server 2 is 500W, 500×1.2=600W. The load prediction value of server 3 is 700W, 700×1.2=840W. The load prediction value of server 4 is 300W, 300×1.2=360W.

[0110] After adjusting the power distribution of the power supply system for powering the server cluster, for example, in this embodiment, server 1 obtains 600W from power module 1 and 360W from power module 2, for a total of 960W. Server 2 obtains 500W from power module 1 and 100W from power module 2, for a total of 600W. Server 3 obtains 300W from power module 1 and 540W from power module 3, for a total of 840W. Server 4 obtains 200W from power module 2 and 160W from power module 3, for a total of 360W.

[0111] In this example, for power module 1, the total power provided is 600W+500W+300W=1400W, which does not exceed its maximum power output limit of 1500W. For power module 2, the total power provided is 360W+100W+200W=660W, which does not exceed its maximum power output limit of 800W. For power module 3, the total power provided is 540W+160W=700W, which does not exceed its maximum power output limit of 800W.

[0112] It can also be seen that, in this example, after the power distribution of the power supply system is adjusted, the total power distribution error reaches the minimum value, and each server receives power from at least two power modules, and the above constraint rules are also satisfied.

[0113] In the present application scheme, it is necessary to be able to adjust the power distribution of the power supply system. There are many specific implementation methods. For example, in a specific implementation, for any power module in the power supply system, the power supply module can have multiple output interfaces, and each output interface can be used to connect to the server to power the corresponding server. For different output interfaces of the power supply module, different power outputs can be provided. For example, multiple groups of adjustable windings can be set on the secondary side of the transformer in the power supply module, and the power adjustment of the corresponding output interface can be achieved by adjusting each group of adjustable windings. For example, for different output interfaces of the power supply module, the power control of the output interface can also be achieved by PWM pulse width modulation and other methods. There are many specific implementation methods, which do not affect the implementation of the present invention. Since the power supply module can have multiple output interfaces for connecting different servers, the output power of each output interface is adjustable. Therefore, by adjusting each power module in the power supply system, the power distribution of the power supply system is adjusted, and after the adjustment, it can meet the requirements described above.

[0114] In a specific embodiment of the present invention, a fault monitoring device may be further included to monitor the fault status of each power module in the power supply system, and when a power module fault is detected, a preset fault response operation is executed.

[0115] This implementation takes into account that the fault status of each power module in the power supply system can also be monitored by a fault monitoring device. If a power module failure occurs, the control device can execute preset fault response operations, thereby ensuring the reliability of the power supply system, which is also beneficial to ensuring the operating stability of the server.

[0116] The specific content of the fault response operation can be set and adjusted according to actual needs. For example, in a specific embodiment of the present invention, the power supply system also includes a backup power supply module. Accordingly, the control device performs a preset fault response operation, which can specifically include:

[0117] Determine the number of failed power modules;

[0118] From the standby power supply modules, an equal number of standby power supply modules are selected to take over the work of the failed power supply modules, and the operation of adjusting the power distribution of the power supply system that supplies power to the server cluster is returned.

[0119] This implementation takes into account that a certain number of backup power modules can be pre-set. If a power module in the power supply system fails, the same number of backup power modules can be selected based on the number of failures to replace the failed power modules. In actual applications, when a power module fails, the backup power module can be immediately started to replace the power module, and the power switching can be completed within milliseconds to ensure that the power supply of the server is not interrupted.

[0120] This implementation method also takes into account that after selecting the same number of backup power modules to take over the work of the faulty power modules, since the power that the backup power modules can provide may not be completely consistent with the original power modules, the power of the power system can be redistributed immediately to ensure the stability of server operation.

[0121] There are many specific implementations for determining the power module failure, for example, when abnormal current / voltage, overheating, etc. are detected, the power module failure can be determined. In addition, it can be understood that in actual applications, after the power module is repaired, it can be reconnected to the power system, and by redistributing the power of the power system, the power module can supply power to the server again.

[0122] In a specific embodiment of the present invention, a communication device may also be included, and the communication device is connected to the control device:

[0123] The control device is also used to: receive power distribution instructions through the communication device, and adjust the power distribution of the power supply system that supplies power to the server cluster according to the power distribution instructions.

[0124] This embodiment takes into account that a communication device may be provided in the power supply system, and the power distribution of the power supply system may be adjusted based on the power distribution instruction sent by the host computer received by the communication device. The communication device may support a variety of industrial standard communication protocols, including Ethernet, CAN bus, RS-485, etc., to ensure compatibility.

[0125] In this implementation, the power distribution of the power supply system can be adjusted directly according to the power distribution instruction, ensuring the flexibility of the solution. It should be noted that since the power distribution of the power supply system is adjusted according to the power distribution instruction, the relevant constraint rules can be ignored. Of course, generally speaking, the power distribution instructions are relatively reasonable power distribution instructions sent by the staff according to actual needs. After the power distribution of the power supply system is adjusted according to the power distribution instruction, the power supply of each server will usually meet the requirements of the relevant constraint rules.

[0126] In a specific embodiment of the present invention, the control device can also be used for:

[0127] For any server, determine the current input power value required by the server;

[0128] Determine whether the error between the input power value and the load prediction value of the server exceeds a preset error range;

[0129] If yes, the input power value currently required by the server is used to replace the load prediction value of the server, and an operation of adjusting the power distribution of the power supply system that supplies power to the server cluster is performed.

[0130] As can be seen from the above description, in the present application scheme, the power distribution of the power supply system is dynamically performed so that the total power distribution error is lower than the preset first threshold value, which means that the power distribution of each server can meet the load prediction value of the server, which means that for any server, if the future load of the server is predicted to increase, the power supply can be increased in advance, thereby avoiding the situation where the load increase causes voltage instability or system performance degradation. On the contrary, if the future load of the server is predicted to decrease, the power supply can be reduced in advance to prevent power waste and improve energy efficiency. In other words, in the present application scheme, the current power distribution of the power supply system is actually adjusted based on the future load of each server. This method usually has a good effect, but it may affect the stability of the server operation in some special scenarios. Such a special scenario is that the server load has an unexpected mutation. At this time, the server should not be powered according to the predicted load prediction value of the server, but should be powered immediately according to the current load of the server.

[0131] In this regard, in this implementation, for any server, the input power value currently required by the server can be determined. For example, due to a sudden computing task, the input power value required by a server suddenly increases to 1000W, far exceeding its load prediction value of 600W. Therefore, it can be determined that the error between the input power value of the server and the load prediction value of the server exceeds the preset error range. At this time, the load prediction value of 600W can be ignored, and the current input power value of 1000W required by the server is directly used as the load prediction value of the server, that is, the input power value of 1000W required by the server is used to replace the original load prediction value of 600W of the server, and an electric power distribution adjustment is immediately performed so that the server can be allocated enough power. For another example, due to the sudden stop of a task on a server, the required input power value suddenly drops to 500W, far below its load prediction value of 800W. Therefore, it can be determined that the error between the input power value of the server and the load prediction value of the server exceeds the preset error range. At this time, the load prediction value of 800W can be ignored, and the current input power value of 500W required by the server can be directly used as the load prediction value of the server, and a power distribution adjustment is immediately performed to avoid power waste.

[0132] The specific value of the error range can be set and adjusted as needed. There are also multiple specific implementation methods for determining the input power value currently required by the server, such as determining it based on the current business of the server, or determining it based on the CPU status, memory status, IO quantity, etc. of the server.

[0133] The technical solution provided by the embodiment of the present invention is applied, considering that the traditional fixed power distribution method is prone to energy waste and will affect the stability of server operation. In this regard, in the solution of the present application, dynamic adjustment of power distribution will be performed. Specifically, in the solution of the present application, it is necessary to monitor the load status of each server in the server cluster, and then determine the load prediction value of each server after the first time period at the current moment based on the load status of each server. This is to take into account that the load will change, and determine the load prediction value of each server after the first time period, thereby realizing dynamic power distribution, which is conducive to ensuring the rationality of power distribution, and then conducive to ensuring the stability of server operation.

[0134] When adjusting the power distribution of the power supply system that supplies power to the server cluster, it is necessary to make the total power distribution error lower than the preset first threshold. The total power distribution error represents the sum of the power distribution errors of each server in the server cluster. In other words, making the total power distribution error lower than the preset first threshold means that the power distribution of each server is consistent with the load forecast value of the server. This means that when it is predicted that the future load of the server will increase, the power supply can be increased in advance, thereby avoiding voltage instability or system performance degradation caused by the increase in load. Conversely, when it is predicted that the future load of the server will decrease, the power supply can be reduced in advance to prevent power waste and improve energy efficiency.

[0135] Furthermore, in order to ensure the stability of power supply, each server in the server cluster needs to receive power from at least K power modules, where K is a positive integer not less than 2. This means that each server will receive power from 2 or more power modules. In this way, when a single power module supplying power to the server fails, the server will not immediately lose all power supply, which is also beneficial to ensure the stability of server operation to a certain extent.

[0136] In summary, the solution of the present application can effectively implement power management, improve energy utilization efficiency, and ensure the stability of server operation.

[0137] Corresponding to the above power system embodiment, the embodiment of the present invention further provides a power management method, which can be referred to in correspondence with the above.

[0138] See also Figure 1 As shown, the power management method can be applied to a power management device, comprising the following steps:

[0139] Step S101: monitoring the load status of each server in the server cluster;

[0140] Step S102: based on the load status of each server, determining the load prediction value of each server after the first time period; wherein the start time of the first time period is the time when the load status monitoring of each server is completed;

[0141] Step S103: adjusting the power distribution of the power supply system for the server cluster so that the total power distribution error is lower than a preset first threshold value, and each server in the server cluster receives power from at least K power modules;

[0142] Among them, the power supply system includes multiple power supply modules, K is a positive integer not less than 2, the total power distribution error represents the sum of the power distribution errors of each server in the server cluster, and the power distribution error of a single server represents the error between the total power obtained by the server from the power supply system and the load prediction value of the server.

[0143] Corresponding to the above embodiments, the embodiments of the present invention further provide a power management device, a computer-readable storage medium, and a computer program product, which can be referred to in correspondence with the above.

[0144] See also Figure 3 As shown, the device may include:

[0145] Memory 301, used for storing computer programs;

[0146] The processor 302 is configured to execute a computer program to implement the steps of the power management method in any of the above embodiments.

[0147] The computer program product includes a computer program / instruction, and when the computer program / instruction is executed by a processor, the steps of the power management method in any of the above embodiments are implemented.

[0148] See also Figure 4 The computer readable storage medium 40 stores a computer program 41, and when the computer program 41 is executed by the processor, the steps of the power management method in any of the above embodiments are implemented. The computer readable storage medium 40 mentioned here includes RAM (Random Access Memory), memory, ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), register, hard disk, removable disk, or any other form of storage medium known in the technical field.

[0149] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0150] Professionals may further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention, and the description of the above embodiments is only used to help understand the technical solution and its core idea of ​​the present invention. It should be pointed out that for ordinary technicians in the technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. A power supply system, characterized in that: include: A plurality of power supply modules for supplying power to the server cluster, and a power management device connected to each of the power supply modules; The power management device comprises: A sensor device for monitoring the load status of each server in the server cluster; A control device connected to the sensor device is used to determine a load prediction value of each server after a first period of time based on a load state of each server; adjust the power distribution of the power supply system so that a total power distribution error is lower than a preset first threshold value, and each server in the server cluster receives power from at least K power modules; Among them, the starting time of the first time period is the time when the load status monitoring of each server is completed; K is a positive integer not less than 2, the total power allocation error represents the sum of the power allocation errors of each server in the server cluster, and the power allocation error of a single server represents the error between the total power obtained by the server from the power supply system and the load prediction value of the server.

2. The power supply system according to claim 1, characterized in that: Also includes: A fault monitoring device, used to monitor the fault status of each power module in the power system and send it to the control device; The control device is connected to the fault monitoring device, and is also used to execute a preset fault response operation when the fault monitoring device detects a fault in the power module.

3. The power supply system according to claim 2, characterized in that: The power supply system also includes a backup power supply module; Accordingly, the control device executes a preset fault response operation, including: Determining the number of the power modules that have failed; From the standby power modules, an equal number of standby power modules are selected to take over the work of the failed power modules, and the power distribution of the power system is adjusted.

4. The power supply system according to claim 1, characterized in that: The control device adjusts the power distribution of the power supply system, including: The control device adjusts the power distribution of the power supply system according to the preset constraint rules; The constraint rule includes: for any power module in the power system, the total power provided by the power module shall not exceed the maximum power output limit value of the power module.

5. The power supply system according to claim 4, characterized in that: The constraint rules also include: The total power obtained by a single server from the power supply system is greater than A times the load prediction value of the server; wherein A is a preset multiplier value and A is greater than 1.

6. The power supply system according to claim 1, characterized in that: Also includes a communication device, and the communication device is connected to the control device; The control device is also used to: receive a power distribution instruction through the communication device, and adjust the power distribution of the power supply system according to the power distribution instruction.

7. The power supply system according to claim 1, characterized in that: The sensor device is specifically used for: Monitor the load status of each server in the server cluster within the most recent second period of time; Accordingly, the control device determines the load prediction value of each server after the first time period based on the load status of each server, including: The control device determines a load prediction value of each of the servers after the first time period based on the load status of each of the servers within the most recent second time period.

8. The power supply system according to claim 7, characterized in that: The sensor device is specifically used for: Monitor each server in the server cluster, and for any one of the servers, determine a central processing unit utilization sequence for reflecting changes in central processing unit utilization of the server within a recent second time period, a memory utilization sequence for reflecting changes in memory occupancy of the server within a recent second time period, a request quantity sequence for reflecting changes in the number of requests of the server within a recent second time period, and a network bandwidth sequence for reflecting changes in the network bandwidth of the server within a recent second time period, and use the central processing unit utilization sequence, the memory utilization sequence, the request quantity sequence and the network bandwidth sequence as the monitored load status of the server within the recent second time period.

9. The power supply system according to claim 8, characterized in that: The control device determines the load prediction value of each server after the first time period based on the load state of each server in the most recent second time period, including: For any one of the servers in the server cluster, the load state of the server is input into a pre-trained first model to obtain a load prediction result output by the first model that reflects the load state of the server after a first period of time; wherein the load prediction result includes the CPU utilization, memory occupancy, number of requests and network bandwidth of the server after the first period of time; The load prediction result is input into a pre-trained second model to obtain a load prediction value output by the second model as the determined load prediction value of the server after the first period of time.

10. The power supply system according to claim 7, characterized in that: The control device determines the load prediction value of each server after the first time period based on the load state of each server in the most recent second time period, including: For any one of the servers in the server cluster, based on the load state of the server in the most recent second time period, N pending prediction values ​​of the server after the first time period are determined by N different calculation methods; wherein N is a positive integer not less than 2; Based on the N pending prediction values, a load prediction value of the server after a first period of time is determined.

11. The power supply system according to claim 10, characterized in that: Determining a load prediction value of the server after a first time period based on the N pending prediction values ​​includes: Based on N pending prediction values ​​and respective weight parameters of N different calculation methods, the load prediction value of the server after the first time period is obtained by weighted superposition of the N pending prediction values.

12. The power supply system according to any one of claims 1 to 11, characterized in that: The control device is also used for: For any server, determining the input power value currently required by the server; Determining whether an error between the input power value and the load prediction value of the server exceeds a preset error range; If yes, the input power value currently required by the server is used to replace the load prediction value of the server, and an operation of adjusting the power distribution of the power supply system that supplies power to the server cluster is performed.

13. A power management method, characterized in that: include: Monitor the load status of each server in the server cluster; Based on the load status of each of the servers, a load prediction value of each of the servers after a first time period is determined; wherein the start time of the first time period is the time when the load status monitoring of each server is completed; Adjusting power distribution of a power supply system that supplies power to the server cluster so that a total power distribution error is lower than a preset first threshold and each of the servers in the server cluster receives power from at least K power modules; In which, the power supply system includes multiple power supply modules, K is a positive integer not less than 2, the total power distribution error represents the sum of the power distribution errors of each server in the server cluster, and the power distribution error of a single server represents the error between the total power obtained by the server from the power supply system and the load prediction value of the server.

14. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the power management method as claimed in claim 13 are implemented.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the power management method according to claim 13 are implemented.

Images