CN109800066B - Energy-saving scheduling method and system for data center - Google Patents
Energy-saving scheduling method and system for data center Download PDFInfo
- Publication number
- CN109800066B CN109800066B CN201811524080.5A CN201811524080A CN109800066B CN 109800066 B CN109800066 B CN 109800066B CN 201811524080 A CN201811524080 A CN 201811524080A CN 109800066 B CN109800066 B CN 109800066B
- Authority
- CN
- China
- Prior art keywords
- data center
- server
- scheduled
- power consumption
- servers
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Air Conditioning Control Device (AREA)
Abstract
The invention provides a data center energy-saving scheduling method and a system, which are characterized in that the total power consumption of a data center and the setting parameters of a machine room air conditioner after a task to be scheduled is deployed on any one server are predicted by acquiring the resource utilization rate of all servers in the current data center, the resource demand of the task to be scheduled in a task queue and the current external environment parameters through a preset prediction model; and determining a data center energy-saving scheduling scheme meeting preset conditions according to the prediction result, and scheduling the total energy consumption of the data center according to the scheme. The energy consumption of the server system and the machine room air conditioning system is jointly scheduled by a machine learning method, the problem of low energy saving efficiency caused by single-layer optimization, adoption of an inaccurate energy consumption model and CFD (computational fluid dynamics) based simulation scheduling in the related technology is solved, and the effect of cross-layer unified optimization of the total energy consumption of the data center is achieved.
Description
Technical Field
The invention relates to the field of data centers, in particular to a data center energy-saving scheduling method and system for realizing cross-layer unified energy consumption optimization of a data center.
Background
With the rapid development of cloud computing technology, as a physical platform of cloud computing, data centers around the world have also been developed unprecedentedly. The rapidly increasing number of data centers also causes significant energy consumption overhead for operators. For example, the total amount of electrical energy consumed by data centers in the united states in 2014 has accounted for 1.8% of the total annual energy consumption throughout the united states, and this value has also remained increasing year by year. In order to reduce the consumption of electric energy in data, a series of energy-saving schemes are provided in academia and industry. Since the server system and the machine room air conditioning system occupy more than about 80% of the power consumption of the data center in total, most of the research is conducted on the two systems. However, there is a trade-off between the energy saving goals for the server system and the air conditioning system. For example, an energy saving scheme for a server system may aggregate IT load onto a few servers to shut down as many servers as possible for energy saving purposes. However, due to the super-linear functional relationship between the fan speed and the energy consumption of the air conditioning system, the energy saving scheme for the air conditioning system can share the IT load among all the servers as much as possible. Therefore, the purpose of saving the overall energy consumption of the data center cannot be achieved by saving the energy consumption of any system in a unilateral manner.
At present, energy-saving schemes for cross-layer unification of a data center server and an air conditioning system are generally divided into three categories. The first type is that a specific mathematical function relation exists between the two systems and the IT load, and a scheduling algorithm is designed on the assumption to carry out overall energy-saving optimization design. However, because of the complex interaction and feedback loops among various parameters affecting the energy consumption of the data center, modeling the energy consumption of the data center by using the traditional engineering formula method is very inaccurate, and therefore, the scheme has poor effect in practice. And the second method is that a temperature sensor is arranged on a server, and the parameters of the cooling system are set according to temperature data sent back by the sensor, and the method cannot predict the total energy consumption of the data center and is difficult to guide the optimization of an energy-saving scheduling algorithm. The last method is a method based on a numerical analysis model to simulate a thermodynamic distribution diagram of the data center, and the method has too high calculation cost and cannot accurately reflect the complexity of heat dissipation of the data center, so that the aim of minimizing the total energy consumption of the data center by performing real-time scheduling on IT loads and setting system parameters cannot be fulfilled.
Disclosure of Invention
The invention provides a data center energy-saving scheduling method and system, which are used for combining energy consumption of a scheduling server system and a machine room air conditioning system through a machine learning method, and solving the problem of low energy-saving efficiency caused by single-level optimization, adoption of an inaccurate energy consumption model and CFD (computational fluid dynamics) based simulation scheduling in the related technology.
According to one aspect of the invention, an energy-saving scheduling method for a data center is provided, wherein the data center comprises a server system comprising at least one server, a machine room air conditioning system, a task queue and an external environment parameter monitoring system, and the method comprises the following steps:
acquiring the current resource utilization rate of all servers in a data center, the resource demand of tasks to be scheduled in a task queue and current external environment parameters;
predicting a result through a preset prediction model generated by a neural network based on a machine learning method according to the obtained current resource utilization rate and current external environment parameters of all the servers, wherein the prediction result comprises the corresponding data center total power consumption and the corresponding machine room air conditioner setting parameters after the task to be scheduled is deployed on any one server, and the data center total power consumption comprises the server system total power consumption and the machine room air conditioner system total power consumption;
according to the prediction result, determining a data center energy-saving scheduling scheme meeting preset conditions, namely traversing all servers capable of accommodating tasks to be scheduled, selecting the servers meeting the preset conditions after the tasks to be scheduled are deployed, marking the servers as the servers to be scheduled, and predicting the total power consumption of the data center corresponding to the tasks to be scheduled after the tasks to be scheduled are deployed on the servers to be scheduled and the corresponding air conditioner setting parameters of the machine room according to a prediction model;
and scheduling the total energy consumption of the data center according to the energy-saving scheduling scheme.
Optionally, the prediction model is obtained by taking the resource utilization rate of all servers in the data center and historical or experimental data of corresponding external environment parameters as input of a neural network, taking a set of machine room air conditioner setting parameters meeting the cooling requirements of the servers and the historical or experimental data of the total power consumption of the data center as output, and training by means of the nonlinear processing capacity of the neural network. The reason why the resource utilization rates of all the servers are adopted instead of the total resource utilization rate of all the servers is taken as input is that the distribution of the relative geographical positions of the servers and the air conditioning system of the machine room greatly influences the total energy consumption of the data center.
Optionally, the preset conditions include: minimizing the overall power consumption of the data center.
Optionally, the scheduling the total energy consumption of the data center includes: and deploying the tasks to be scheduled to the servers to be scheduled, adjusting the working state of the servers to realize power consumption scheduling of the server system, and adjusting the set parameters of the air conditioners in the machine room to realize power consumption scheduling of the air conditioning system in the machine room.
Optionally, the scheduling of power consumption of the air conditioning system in the machine room further includes: whether the current air conditioner setting meets the cooling requirement of the server is detected, and if not, the air conditioner parameters are finely adjusted until the cooling requirement of the server is met.
According to another aspect of the present invention, there is provided an energy-saving scheduling system for a data center including a server system including at least one server, a room air conditioning system, a task queue, and an external environment parameter monitoring system, the system including:
the acquisition device comprises a server resource utilization rate acquisition device, a task resource demand acquisition device to be scheduled, an external environment parameter acquisition device, a machine room air conditioning system parameter acquisition device and a data center total power consumption acquisition device, and is respectively responsible for acquiring the resource utilization rates of all servers in the data center, the resource demands of tasks to be scheduled in a task queue, the external environment parameters, machine room air conditioning setting parameters and the data center total power consumption;
the prediction device is responsible for training a neural network according to a machine learning method through the data acquired by the acquisition device to generate a prediction model for outputting the corresponding total power consumption of the data center and the corresponding air conditioner setting parameters of the machine room after the task to be scheduled is deployed on any one server;
the energy-saving scheduling scheme generating device is responsible for generating a data center energy-saving scheduling scheme meeting preset conditions according to the prediction model generated by the prediction device, the current resource utilization rate of the server acquired by the acquiring device, the resource requirements of the tasks to be scheduled in the task queue and the external environment parameters;
and the setting device is responsible for deploying the energy-saving scheduling scheme generated by the energy-saving scheduling scheme generating device to the data center.
Optionally, the data center total power consumption obtaining device includes a server power consumption obtaining device and a machine room air conditioning system power consumption obtaining device, and is respectively responsible for obtaining the server system total power consumption and the air conditioning machine room system total power consumption.
Optionally, the prediction device includes a training device, which is responsible for training to obtain the prediction model by means of the nonlinear processing capability of the neural network, with the resource utilization rate of all servers in the data center and the historical or experimental data corresponding to the external environment parameters as the input of the neural network, and with the corresponding machine room air conditioner setting parameters and the historical or experimental data of the total power consumption of the data center as the output.
Optionally, the energy-saving scheduling scheme generating device includes:
the server determining device is responsible for traversing all servers capable of accommodating the tasks to be scheduled, selecting the servers meeting preset conditions after the tasks to be scheduled are deployed, and marking the servers as the servers to be scheduled;
and the air conditioner determining device is responsible for predicting the total power consumption of the data center corresponding to the task to be scheduled after the task to be scheduled is deployed in the server to be scheduled and the corresponding air conditioner setting parameters of the machine room according to the prediction model.
Optionally, meeting the preset condition includes: minimizing the overall power consumption of the data center.
Optionally, the setting means comprises:
the task deployment device to be scheduled is responsible for deploying the task to be scheduled to the server to be scheduled;
the server working state setting device is responsible for adjusting the working state of the server to realize power consumption scheduling of the server system;
and the parameter setting device of the machine room air conditioning system is responsible for adjusting the setting parameters of the machine room air conditioners so as to realize power consumption scheduling of the machine room air conditioning system.
Optionally, the device for setting parameters of the air conditioning system in the machine room further comprises:
the detection device is responsible for detecting whether the current air conditioner setting meets the cooling requirement of the server;
and the fine-tuning device is used for fine-tuning the air conditioner parameters until the cooling requirement of the server is met if the detection result of the detection device is not up to the standard.
According to the invention, the resource utilization rate of all servers in the current data center, the resource demand of the task to be scheduled in the task queue and the current external environment parameter are obtained; according to the acquired server resource utilization rate and the current environment parameters, predicting the corresponding data center total power consumption and the machine room air conditioner setting parameters of the tasks to be scheduled after being deployed on any one server according to a preset prediction model generated based on a machine learning method; the data center energy-saving scheduling scheme meeting the preset conditions is determined according to the prediction result, and the total energy consumption of the data center is scheduled according to the scheme, so that the problem of low energy-saving efficiency caused by single-level optimization, adoption of an inaccurate energy consumption model and CFD (computational fluid dynamics) simulation scheduling in the related technology is solved, and the effect of cross-layer unified optimization of the total energy consumption of the data center is achieved.
Drawings
Fig. 1 is a flowchart of a data center energy-saving scheduling method according to an embodiment of the present invention;
FIG. 2 is a block diagram of a data center energy-saving scheduling system according to an embodiment of the present invention;
fig. 3 is a block diagram of a data center energy-saving scheduling system obtaining apparatus 201 according to an embodiment of the present invention;
fig. 4 is a block diagram of a preferred structure of the data center energy-saving dispatching system predicting device 202 according to an embodiment of the present invention;
fig. 5 is a preferred structural framework diagram of the energy-saving scheduling scheme generating device 203 of the energy-saving scheduling system of the data center according to the embodiment of the present invention;
fig. 6 is a preferred structural framework diagram of the data center energy-saving dispatching system setting device 204 according to the embodiment of the invention.
Detailed Description
For a further understanding of the technical aspects of the present invention, reference will now be made in detail to the embodiments of the present invention with reference to the accompanying drawings.
In this embodiment, an energy-saving scheduling method for a data center is provided, where the data center at least includes one or more servers, a machine room air conditioning system, a task queue, and an external environment parameter monitoring system, and fig. 1 is a flowchart of the energy-saving scheduling method for a data center according to an embodiment of the present invention, where the flowchart includes the following steps:
and 103, determining a data center energy-saving scheduling scheme meeting preset conditions according to the prediction result, and scheduling the total energy consumption of the data center according to the energy-saving scheduling scheme.
Through the steps, the problem of low overall energy-saving efficiency of the data center caused by optimizing only aiming at the energy consumption of a single layer of the data center in the related technology is solved through the energy consumption of the combined dispatching server system and the machine room air conditioning system, in addition, the problem that an energy-saving strategy is invalid due to the fact that task dispatching is carried out based on an inaccurate energy consumption model in the related technology and the problem that the task dispatching strategy based on CFD simulation cannot meet the real-time dispatching requirement of the data center due to too large computing overhead is solved through the adoption of a machine learning-based method, and the energy consumption saving effect of the data center is further improved.
The preset prediction model can be generated by adopting the following method: the resource utilization rate of all servers in the data center and historical or experimental data corresponding to external environment parameters are acquired at certain time intervals and used as input of a neural network, and a set of machine room air conditioner setting parameters meeting the cooling requirements of the servers and the historical or experimental data of the total power consumption of the data center are acquired as output training. The neural network has strong nonlinear processing capacity, is very suitable for processing complex nonlinear relations in the data center, and is short in time spent when the model trained by the neural network is used for prediction, so that the neural network is suitable for real-time online scheduling of the data center. As known to those skilled in the art, the room air conditioning parameters mainly include temperature and wind speed (wind speed determines the amount of air delivered). The setting of air conditioning parameters in a data center requires the condition of server cooling to be satisfied, for example, the american society of heating, cooling and air conditioning engineers (ASHRAE) set the condition of server intake air to 18-27 ℃ in 2008. Therefore, under the condition that the resource utilization rate of the server and the external environment parameters are kept unchanged, a series of air conditioner parameter settings which accord with the refrigeration condition of the server can be obtained by traversing the air conditioner setting parameters, and one group of air conditioner setting parameters corresponding to the data center with the minimum total power consumption can be selected to be used as the output of the neural network.
There is a trade-off between the energy saving goals for the server system and the energy saving goals for the air conditioning system. For example, an energy saving scheme for a server system may aggregate IT load onto a few servers to shut down as many servers as possible for energy saving purposes. However, due to the super-linear functional relationship between the fan speed and the energy consumption of the air conditioning system, the energy saving scheme for the air conditioning system can share the IT load among all the servers as much as possible. Therefore, the purpose of saving the overall energy consumption of the data center cannot be achieved by saving the energy consumption of any system in a unilateral manner. In order to save the energy consumption of the data center on the whole, the total power consumption of the data center comprises two parts, namely the total power consumption of a server system and the total power consumption of a machine room air conditioning system.
The data center energy-saving scheduling scheme meeting the preset conditions can be determined by traversing all servers capable of containing the tasks to be scheduled, selecting the servers meeting the preset conditions after the tasks are deployed, and marking the servers as the servers to be scheduled. The condition that the server can accommodate the task to be scheduled is that the remaining resources of the server are not less than the resource demand of the task to be scheduled. Supposing that N servers are provided, calculating the resource utilization rate of the server after a task to be scheduled is deployed in a server I (1< ═ I < ═ N), inputting the resource utilization rate of the server, the resource utilization rates of all other servers and the current external environment parameters into a preset prediction model, recording (or storing) the output of the model, and selecting the servers meeting the preset conditions in N outputs and marking as I; and then, predicting the corresponding total power consumption of the data center and the corresponding air conditioner setting parameters of the machine room after the task is deployed on the server to be scheduled according to a preset prediction model, and if the corresponding output of the task deployed on the server I is stored in the process of determining the server to be scheduled, directly calling and recording to obtain the corresponding total power consumption of the data center and the corresponding air conditioner setting parameters of the machine room.
Meeting the preset conditions comprises the following steps: minimizing the overall power consumption of the data center.
The scheduling of the total energy consumption of the data center according to the energy-saving scheme comprises three aspects: deploying the task to the determined server to be scheduled; as tasks in the data center dynamically arrive and leave along with time, when a new task is deployed, a phenomenon that the utilization rate of part of servers is reduced, even part of servers are in an idle state because all tasks carried by the servers are processed, is likely to occur, and at this time, the servers in the idle state waste a large amount of energy consumption, so that the working state of the servers per se needs to be adjusted to realize power consumption scheduling of a server system, for example, the servers in the idle state are switched to a low-power-consumption sleep mode (even turned off), and the working voltage and frequency of the servers with low utilization rate are reduced; and adjusting the parameter setting group of the air conditioning system of the machine room to realize power consumption scheduling of the air conditioning system of the machine room. Due to the changes in the server load level and the operating state, the air conditioning room system needs to be adjusted accordingly to meet the new server refrigeration demand, because if the server system refrigeration demand decreases, the air conditioning system needs to correspondingly decrease the cooling capacity to reduce the energy consumption of the data center, and if the server system refrigeration demand increases, the air conditioning system needs to correspondingly increase the cooling capacity to meet the server refrigeration demand.
Because there is a certain possibility that the parameter setting of the air conditioning system of the machine room predicted by the prediction model cannot meet the cooling requirement of the current server, it is necessary to detect whether the parameter setting of the current air conditioning parameter reaches the cooling requirement of the server after adjusting the parameter setting group of the air conditioning system of the machine room, and as known to those skilled in the art, whether the parameter setting of the current air conditioning parameter reaches the cooling requirement of the server can be judged by detecting whether the temperature of the top inlet of the cabinet meets the specification (for example, is between 18 and 27 ℃); if the temperature is not reached, the air conditioner parameters can be finely adjusted until the cooling requirement of the server is reached, for example, if the temperature is higher than 27 ℃, the temperature and the air speed of the air conditioner can be gradually increased until the temperature reaches the standard, which indicates that the cooling capacity of the air conditioner is insufficient at the moment; if the temperature is lower than 18 ℃, which indicates that the cooling capacity of the air conditioner is excessive, the temperature and the wind speed of the air conditioner can be gradually reduced until the temperature reaches the standard.
In this embodiment, a data center energy-saving scheduling system is provided, and this system is used to implement the foregoing embodiments, and therefore, the description already made will not be repeated herein. All of the following means can be implemented by hardware or software, or a combination of hardware and software. Fig. 2 is a block diagram of a data center energy-saving scheduling system according to an embodiment of the present invention, and as shown in fig. 2, the system includes an obtaining device 201, a predicting device 202, an energy-saving scheduling scheme generating device 203, and a setting device 204, where:
the acquiring device 201 is responsible for acquiring resource utilization rates of all servers in a data center, resource demands of tasks to be scheduled in a task queue, external environment parameters, machine room air conditioner parameters and total data center power consumption, fig. 3 is a structural block diagram of the acquiring device 201, and as shown in fig. 3, the acquiring device 201 comprises a server resource utilization rate acquiring device 205, a machine room air conditioner system parameter acquiring device 206, a task resource demand acquiring device 207 to be scheduled, an external environment parameter acquiring device 208 and a total data center power consumption acquiring device 209;
the prediction device 202 is responsible for the resource utilization rates of all servers in the data center, the resource requirements of the tasks to be scheduled in the task queue, the external environment parameters, the machine room air conditioning parameter setting and the total power consumption of the data center, which are acquired by the acquisition device, and generates a prediction model capable of predicting the corresponding machine room air conditioning parameters and the total power consumption of the data center according to the resource utilization rates of all servers in the data center, the resource requirements of the tasks to be scheduled in the task queue and the external environment parameters by a machine learning method;
the energy-saving scheduling scheme generating device 203 is responsible for generating a data center energy-saving scheduling scheme meeting preset conditions according to the prediction model generated by the prediction device, the resource utilization rate of the current server acquired by the acquiring device, the resource demand of the tasks to be scheduled in the task queue and the external environment parameters;
the setting device 204 is responsible for deploying the energy-saving scheduling scheme of the data center generated by the energy-saving scheduling scheme generating device to the data center.
Preferably, the data center total power consumption obtaining device 209 includes a server system power consumption obtaining device and a machine room air conditioning system power consumption obtaining device.
Fig. 4 is a frame diagram of a preferred structure of the prediction apparatus 202 in the energy-saving scheduling system of the data center according to an embodiment of the present invention, as shown in fig. 4, the prediction apparatus 202 may generate a prediction model through a built-in training apparatus 210, and the training apparatus 210 is responsible for training the prediction model by using the resource utilization rate of all servers in the data center and the history or experimental data of corresponding external environment parameters as inputs of a neural network, and using the corresponding machine room air conditioner setting parameters and the history or experimental data of total power consumption of the data center as outputs, and using the strong nonlinear processing capability of the neural network. As known to those skilled in the art, neural networks may be implemented by software.
Fig. 5 is a preferred structural framework diagram of the energy-saving scheduling scheme generating device 203 in the energy-saving scheduling system of the data center according to the embodiment of the present invention, and as shown in fig. 5, the device 203 includes a server determining device 211 and an air conditioner determining device 212, which are respectively described as follows:
the server determining device 211 is responsible for traversing all servers capable of accommodating the tasks to be scheduled, selecting the servers meeting the preset conditions after the tasks are deployed, and marking the servers as the servers to be scheduled;
and the air conditioner determining device 212 is responsible for predicting the total power consumption of the data center corresponding to the task after the task is deployed on the server to be scheduled and the corresponding air conditioner setting parameters of the machine room according to a preset prediction model.
Fig. 6 is a frame diagram of a preferred structure of a setting device 204 in an energy-saving dispatching system of a data center according to an embodiment of the present invention, and as shown in fig. 6, the device 204 includes a task deployment device 213 to be dispatched, a server operating state setting device 214, and a machine room air conditioning system parameter setting device 215, which are respectively described as follows:
a task deployment device 213 to be scheduled, which is responsible for deploying the task to the server to be scheduled;
the server working state setting device 214 is responsible for adjusting the working state of the server to realize power consumption scheduling of the server system;
and the machine room air conditioning system parameter setting device 215 is responsible for adjusting the machine room air conditioning system parameter setting group so as to realize power consumption scheduling of the machine room air conditioning system.
Preferably, the equipment room air conditioning system parameter setting device 215 further includes:
the detection device is responsible for detecting whether the current air conditioner setting meets the cooling requirement of the server;
and the fine-tuning device is used for fine-tuning the air conditioner parameters until the cooling requirement of the server is met if the detection result of the detection device is not up to the standard.
In recent years, data centers have been developed in a standardized and large-scale manner. A large number of standardized cabinets are arranged in a large cloud computing data center in order, and a plurality of standardized servers are arranged in the cabinets. Based on the related technology, the cabinets of the existing data center are generally arranged in a manner of separating cold channels from hot channels, namely, cold air is blown upwards from small holes on the floor between two rows of cabinets and is sucked by the cabinets for refrigeration of servers in the cabinets, and hot air is discharged from the rear of the cabinets and is sucked away from the top of a house by refrigeration equipment. The servers can only work safely when the temperature of the servers cannot exceed the temperature specified by the manufacturer, and therefore, the parameter setting of the air conditioner of the machine room must be capable of ensuring that the temperature of all the servers cannot be higher than the safety range of the servers. Due to the influence of factors such as airflow, load distribution, environment and the like, the temperature of cold air at the inlets of different servers in the data center is greatly different, so that a large amount of energy consumption of the data center is wasted. The factors influencing the energy consumption of the data center are many, the relation is complex and is difficult to accurately depict by using the numerical relation, the scale of the cloud computing data center is large generally, so that some energy-saving schemes cannot meet the real-time scheduling requirement due to too large computing overhead, and the factors hinder the improvement of the energy-saving effect of the data center.
In view of the above problems, the present embodiment provides a load scheduling method based on energy consumption perception, and the method can make the distribution of loads meet the energy saving target of a data center as much as possible, so as to reduce the overall energy consumption of the data center. The method comprises the following steps:
301, acquiring resource utilization rates of all servers in the current data center, resource requirements of tasks to be scheduled and current external environment parameters by a data center monitoring platform;
step 302, according to the acquired resource utilization rate of the server and the current environmental parameters, predicting the corresponding total power consumption of the data center and the corresponding setting parameters of the air conditioner of the machine room after the task to be scheduled is deployed in any one machine cabinet according to a preset machine cabinet prediction model, wherein the prediction model is generated based on a machine learning method, and the machine cabinet with the minimum total power consumption of the corresponding data center after deployment is selected as the machine cabinet to be scheduled;
step 303, acquiring resource utilization rates of all servers in the cabinet to be scheduled, and predicting the total power consumption of the corresponding cabinet after the task to be scheduled is deployed in any one server according to a preset server prediction model, wherein the prediction model is generated based on a machine learning method, and the server with the minimum total power consumption of the corresponding cabinet after deployment is selected as the server to be scheduled;
and step 304, deploying the task to be scheduled to a server to be scheduled.
The prediction models in the steps 302 and 303 are generated by a machine learning method, and because the neural network has strong processing capability on complex relations and is very suitable for the energy-saving problem of a data center, the neural network can be adopted for model training. As known to those skilled in the art, neural networks are divided into three parts: the input layer, the hidden layer and the output layer can be realized by software. The cabinet prediction model based on the neural network in step 302 may be generated by the following method: the method comprises the steps of obtaining the resource utilization rate of all cabinets in a data center and history or experimental data of external environment parameters (mainly temperature and pressure) as input of a neural network, obtaining the corresponding total power consumption of the data center and the history or experimental data of machine room air conditioner setting parameters meeting the cooling requirements of a server as output of the neural network for training, and setting certain sampling time for collecting the data. The neural network based server prediction model in step 302 may be generated by: and acquiring history or experimental data of resource utilization rates of all servers in a given cabinet as input of a neural network at certain sampling time intervals, and acquiring history or experimental data of corresponding cabinet power consumption as output of the neural network for training. The power consumption data can be actually measured by a power meter or collected by simulation software.
Under the condition that all cabinets in the machine room are uniform and homogeneous, the model trained in step 303 can be applied to all cabinets, so that a large amount of training workload can be saved compared with a method for constructing a data center by load deployment and directly predicting total power consumption by using a neural network. Obviously, the above embodiment is also applicable to the scenario that the cabinets in the machine room are not uniform and homogeneous. By the embodiment, load scheduling based on energy consumption can be realized, the method can solve the problem of low overall energy-saving efficiency of the data center caused by inaccurate energy consumption depiction in the prior energy-saving technology, in addition, input and output parameters of the neural network are greatly reduced by decoupling in the step 302 and the step 303, the time for calculating and outputting the input data through the trained neural network is also greatly reduced, and the method can be used for realizing real-time energy-saving load scheduling.
The above embodiments are only intended to illustrate the technical solution of the present invention and not to limit the same, and a person skilled in the art can modify the technical solution of the present invention or substitute the same without departing from the spirit and scope of the present invention, and the scope of the present invention should be determined by the claims.
Claims (10)
1. A data center energy-saving dispatching method is characterized in that the data center comprises a server system comprising at least one server, a machine room air conditioning system, a task queue and an external environment parameter monitoring system, and the method comprises the following steps:
acquiring the current resource utilization rate of all servers in a data center, the resource demand of tasks to be scheduled in a task queue and current external environment parameters;
calculating the resource utilization rate of a server after the task to be scheduled is deployed on the server according to the obtained current resource utilization rates of all the servers and the resource requirements of the task to be scheduled in the task queue, predicting a result by a preset prediction model generated by a neural network based on a machine learning method together with the resource utilization rates of all other servers and current external environment parameters, wherein the prediction result comprises the corresponding total power consumption of a data center and the corresponding air conditioner setting parameters of a machine room after the task to be scheduled is deployed on any one server, and the total power consumption of the data center comprises the total power consumption of a server system and the total power consumption of the air conditioner system of the machine room;
traversing all servers capable of containing tasks to be scheduled according to the prediction result, selecting the servers meeting preset conditions after the tasks to be scheduled are deployed, marking the servers as the servers to be scheduled, predicting the corresponding total power consumption of the data center and the corresponding air conditioner setting parameters of the machine room after the tasks to be scheduled are deployed on the servers to be scheduled according to the prediction model, and determining the energy-saving scheduling scheme of the data center meeting the preset conditions;
and scheduling the total energy consumption of the data center according to the energy-saving scheduling scheme.
2. The method of claim 1, wherein the prediction model is obtained by training with the aid of nonlinear processing capability of a neural network, with resource utilization of all servers in the data center and historical or experimental data of corresponding external environment parameters as inputs of the neural network, and with a set of machine room air conditioner setting parameters meeting cooling requirements of the servers and historical or experimental data of total power consumption of the data center as outputs.
3. The method of claim 1, wherein the preset condition comprises minimizing a total power consumption of the data center.
4. The method of claim 1, wherein scheduling total energy consumption for a data center comprises: deploying the tasks to be scheduled to a server to be scheduled, adjusting the working state of the server to realize power consumption scheduling of the server system, and adjusting the set parameters of the air conditioner of the machine room to realize power consumption scheduling of the air conditioning system of the machine room;
the power consumption scheduling of the machine room air conditioning system further comprises: whether the current air conditioner setting meets the cooling requirement of the server is detected, and if not, the air conditioner parameters are finely adjusted until the cooling requirement of the server is met.
5. An energy-saving dispatching system of a data center, wherein the data center comprises a server system comprising at least one server, a machine room air conditioning system, a task queue and an external environment parameter monitoring system, and the system is characterized by comprising:
the acquisition device comprises a server resource utilization rate acquisition device, a task resource demand acquisition device to be scheduled, an external environment parameter acquisition device, a machine room air conditioning system parameter acquisition device and a data center total power consumption acquisition device, and is respectively responsible for acquiring the resource utilization rates of all servers in the data center, the resource demands of tasks to be scheduled in a task queue, the external environment parameters, machine room air conditioning setting parameters and the data center total power consumption;
the prediction device is responsible for training a neural network according to a machine learning method through the data acquired by the acquisition device to generate a prediction model for outputting the corresponding total power consumption of the data center and the corresponding air conditioner setting parameters of the machine room after the task to be scheduled is deployed on any one server;
the energy-saving scheduling scheme generating device is responsible for calculating the resource utilization rate of a server after the task to be scheduled is deployed in the server according to the current resource utilization rate of the server acquired by the acquiring device and the resource demand of the task to be scheduled in the task queue, inputting the resource utilization rate of the server together with the resource utilization rates of all other servers and the current external environment parameters acquired by the acquiring device into the prediction model generated by the predicting device, and generating a data center energy-saving scheduling scheme meeting preset conditions;
and the setting device is responsible for deploying the energy-saving scheduling scheme generated by the energy-saving scheduling scheme generating device to the data center.
6. The system of claim 5, wherein the data center total power consumption obtaining device comprises a server power consumption obtaining device and a machine room air conditioning system power consumption obtaining device which are respectively responsible for obtaining the server system total power consumption and the air conditioning machine room system total power consumption.
7. The system of claim 5, wherein the prediction device comprises a training device which takes the resource utilization rate of all servers in the data center and the historical or experimental data of the corresponding external environment parameters as the input of the neural network, takes the corresponding machine room air conditioner setting parameters and the historical or experimental data of the total power consumption of the data center as the output, and obtains the prediction model by means of the nonlinear processing capability training of the neural network.
8. The system of claim 5, wherein the power-saving scheduling scheme generating means comprises:
the server determining device is responsible for traversing all servers capable of accommodating the tasks to be scheduled, selecting the servers meeting preset conditions after the tasks to be scheduled are deployed, and marking the servers as the servers to be scheduled;
and the air conditioner determining device is responsible for predicting the total power consumption of the data center corresponding to the task to be scheduled after the task to be scheduled is deployed in the server to be scheduled and the corresponding air conditioner setting parameters of the machine room according to the prediction model.
9. The system of claim 5, wherein the means for setting comprises:
the task deployment device to be scheduled is responsible for deploying the task to be scheduled to the server to be scheduled;
the server working state setting device is responsible for adjusting the working state of the server to realize power consumption scheduling of the server system;
and the parameter setting device of the machine room air conditioning system is responsible for adjusting the setting parameters of the machine room air conditioners so as to realize power consumption scheduling of the machine room air conditioning system.
10. The system of claim 5, wherein the room air conditioning system parameter setting means further comprises:
the detection device is responsible for detecting whether the current air conditioner setting meets the cooling requirement of the server;
and the fine-tuning device is used for fine-tuning the air conditioner parameters until the cooling requirement of the server is met if the detection result of the detection device is not up to the standard.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811524080.5A CN109800066B (en) | 2018-12-13 | 2018-12-13 | Energy-saving scheduling method and system for data center |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811524080.5A CN109800066B (en) | 2018-12-13 | 2018-12-13 | Energy-saving scheduling method and system for data center |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109800066A CN109800066A (en) | 2019-05-24 |
| CN109800066B true CN109800066B (en) | 2020-11-03 |
Family
ID=66556684
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811524080.5A Active CN109800066B (en) | 2018-12-13 | 2018-12-13 | Energy-saving scheduling method and system for data center |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109800066B (en) |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110376896A (en) * | 2019-07-30 | 2019-10-25 | 浙江大学 | It is a kind of that refrigerating method is optimized based on deep learning and the single heat source air-conditioning of fuzzy control |
| WO2021046774A1 (en) * | 2019-09-11 | 2021-03-18 | 阿里巴巴集团控股有限公司 | Resource scheduling method and information prediction method, device, system, and storage medium |
| CN110781125A (en) * | 2019-09-12 | 2020-02-11 | 华东计算技术研究所（中国电子科技集团公司第三十二研究所） | Management method and system for complete cabinet super-fusion server |
| CN111174375B (en) * | 2019-12-11 | 2021-02-02 | 西安交通大学 | Data center energy consumption minimization-oriented job scheduling and machine room air conditioner regulation and control method |
| CN112070353B (en) * | 2020-08-04 | 2023-09-29 | 中国科学院信息工程研究所 | Method and system for accurately detecting energy efficiency of data center |
| CN112308734B (en) * | 2020-10-27 | 2024-01-05 | 中国科学院信息工程研究所 | IT equipment non-IT energy consumption metering and cost sharing method and electronic device |
| CN112612305B (en) * | 2020-12-04 | 2022-04-08 | 格力电器（武汉）有限公司 | Temperature adjusting method, device, equipment, storage medium and air conditioning system |
| CN112888268B (en) * | 2021-02-04 | 2022-08-09 | 中国工商银行股份有限公司 | Energy-saving control method, device and equipment for data center machine room and storage medium |
| US11442516B1 (en) * | 2021-03-18 | 2022-09-13 | Baidu Usa Llc | Data center control hierarchy for neural networks integration |
| CN115129141A (en) * | 2021-03-25 | 2022-09-30 | 华为技术有限公司 | Energy-saving method, device and system applied to machine room |
| CN113515150B (en) * | 2021-04-12 | 2022-05-03 | 天津大学 | Variable-temperature control method for realizing instant photovoltaic power generation consumption of data center |
| CN113835460B (en) * | 2021-11-23 | 2022-05-03 | 中铁建设集团有限公司 | Data computer lab intelligence environmental control system based on hydrodynamics analog digital twin |
| CN114063545B (en) * | 2022-01-14 | 2022-06-07 | 宁波亮控信息科技有限公司 | Data center energy consumption control system and method fusing edge calculation and controller |
| CN114595851B (en) * | 2022-02-20 | 2022-09-30 | 特斯联科技集团有限公司 | Air conditioner room power consumption analysis device using neural network |
| CN115907138B (en) * | 2022-11-18 | 2023-06-30 | 安华数据(东莞)有限公司 | Method, system and medium for predicting PUE value of data center |
| CN116193819B (en) * | 2023-01-19 | 2024-02-02 | 中国长江三峡集团有限公司 | Energy-saving control method, system and device for data center machine room and electronic equipment |
| CN116527416B (en) * | 2023-07-03 | 2023-09-01 | 深圳市立湾科技有限公司 | Intelligent AI energy-saving control system and method applied to data center |
| CN117421131B (en) * | 2023-12-18 | 2024-03-26 | 武汉泽塔云科技股份有限公司 | Intelligent scheduling method and system for monitoring power consumption load of server |
| CN117472167B (en) * | 2023-12-28 | 2024-03-22 | 苏州元脑智能科技有限公司 | Method and device for adjusting energy consumption of server, computer equipment and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8457938B2 (en) * | 2007-12-05 | 2013-06-04 | International Business Machines Corporation | Apparatus and method for simulating one or more operational characteristics of an electronics rack |
| CN102620378B (en) * | 2011-01-27 | 2014-01-15 | 国际商业机器公司 | Method and system for data center energy saving controlling |
| CN102445980A (en) * | 2011-09-19 | 2012-05-09 | 浪潮电子信息产业股份有限公司 | Energy-saving control system based on back-propagation (BP) neural network |
| CN104423531A (en) * | 2013-09-05 | 2015-03-18 | 中兴通讯股份有限公司 | Data center energy consumption scheduling method and data center energy consumption scheduling device |
| CN104317654A (en) * | 2014-10-09 | 2015-01-28 | 南京大学镇江高新技术研究院 | Data center task scheduling method based on dynamic temperature prediction model |
-
2018
- 2018-12-13 CN CN201811524080.5A patent/CN109800066B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109800066A (en) | 2019-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109800066B (en) | Energy-saving scheduling method and system for data center | |
| CN109375994B (en) | Data center task temperature prediction and scheduling method based on RBF neural network | |
| Tang et al. | Sensor-based fast thermal evaluation model for energy efficient high-performance datacenters | |
| Fang et al. | Thermal-aware energy management of an HPC data center via two-time-scale control | |
| CN104698843B (en) | A kind of data center's energy-saving control method based on Model Predictive Control | |
| US8155793B2 (en) | System and method for controlling air conditioning facilities, and system and method for power management of computer room | |
| Wang et al. | Thermal aware workload placement with task-temperature profiles in a data center | |
| CN111174375B (en) | Data center energy consumption minimization-oriented job scheduling and machine room air conditioner regulation and control method | |
| Fang et al. | Qos-driven power management of data centers via model predictive control | |
| CN107036238B (en) | Intelligent energy-saving control method for dynamically predicting external air and load | |
| US20130191676A1 (en) | Operation management method of information processing system | |
| CN107143981B (en) | A kind of controlling system of central air conditioner and method | |
| CN104049716A (en) | Computer energy-saving method and system combined with temperature sensing | |
| US11906180B1 (en) | Data center management systems and methods for compute density efficiency measurements | |
| CN114063545B (en) | Data center energy consumption control system and method fusing edge calculation and controller | |
| CN114901057B (en) | Multi-point energy consumption detection and dynamic regulation system in data center machine room | |
| CN112070353A (en) | Method and system for accurately detecting energy efficiency of data center | |
| CN113225994B (en) | Intelligent air conditioner control method facing data center | |
| Conficoni et al. | Hpc cooling: A flexible modeling tool for effective design and management | |
| CN111083201B (en) | Energy-saving resource allocation method for data-driven manufacturing service in industrial Internet of things | |
| CN101893306A (en) | Control the equipment and the method for the amount of opening of a plurality of air transfer grids | |
| Fang et al. | Using model predictive control in data centers for dynamic server provisioning | |
| WO2020176080A1 (en) | Data center management systems and methods for compute density efficiency measurements | |
| EP2575003B1 (en) | Method for determining assignment of loads of data center and information processing system | |
| Bortot et al. | Data centers are a software development challenge |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |