CN104423531A - Data center energy consumption scheduling method and data center energy consumption scheduling device - Google Patents

Abstract

The invention provides a data center energy consumption scheduling method and a data center energy consumption scheduling device. The method comprises the following steps of acquiring resource utilization rate and/or environment temperature of one or a plurality of cabinets of a data center; and scheduling energy consumption of the data center according to the acquired resource utilization rate and/or the environment temperature. By the data center energy consumption scheduling method and the data center energy consumption scheduling device, the problems that reduction treatment on the energy consumption of a data center is incomplete by using a correlation technique and the energy using efficiency of the data center is not high are solved, energy consumption scheduling of load distribution can be realized, and energy consumption of refrigeration equipment of the data center is reduced to a certain degree.

Description

Data center energy consumption scheduling processing method and device

technical field

The present invention relates to the communication field, in particular, to a data center energy consumption scheduling processing method and device.

Background technique

With the growth of enterprises' demand for information processing capabilities provided by cloud computing, it is conceivable that data centers will become an indispensable public infrastructure like power plants. This trend can be verified from the number of data centers that have started construction in various places. At the same time, the scale of new data centers is expanding rapidly, and the computing density and energy demand (including cooling facilities and computing equipment) are also increasing rapidly. From the operating data of existing data centers, it can be seen that energy supply expenses have gradually become one of the main expenses in the operation of data centers. How to reduce unnecessary energy expenses, improve the energy utilization efficiency of existing equipment, and reduce the overall Data center energy expenditure, thereby reducing greenhouse gas emissions and protecting the environment.

In related technologies, the research on energy usage efficiency of data centers mainly focuses on how to reduce computing energy consumption through virtualization technology. Based on the characteristics of virtual machines in terms of integration of computing resources and migration of computing tasks, the Internet Data Center (IDC) predicts that more than 50% of enterprise applications will run on virtual machines for the first time in recent years. At the same time, more than 23% of the factory-made servers will support virtualization technology every year, that is, these factory-made servers will be pre-installed with a virtual machine monitor. It can be predicted that virtualization technology will become an important basis for future enterprise applications. For data center energy-saving management, virtualization technology is also an important component.

When the research on the energy efficiency of data centers focuses on how to reduce computing energy consumption through virtualization technology, it is mainly based on the load balancing among servers. However, there are many factors affecting the heat distribution of data centers, which do not exist in related technologies. A technical treatment that takes into account the heat distribution of the data center in a unified manner.

Therefore, in the related art, there is a problem that the energy consumption reduction of the data center is not comprehensively processed, resulting in a problem that the utilization efficiency of the energy consumption of the data center is not high.

Contents of the invention

The present invention provides a data center energy consumption scheduling processing method and device to at least solve the problem in the related art that the energy consumption reduction processing of the data center is not comprehensive, resulting in low energy utilization efficiency of the data center.

According to one aspect of the present invention, a data center energy consumption scheduling processing method is provided, including: obtaining the resource utilization rate and/or ambient temperature of one or more cabinets in the data center; according to the obtained resource utilization rate and/or The ambient temperature schedules the energy consumption of the data center.

Preferably, scheduling the energy consumption of the data center according to the obtained ambient temperature includes: judging whether the highest ambient temperature in the one or more cabinets obtained within a predetermined period of time has decreased; If yes, increase the supply air temperature of the data center refrigeration equipment.

Preferably, scheduling the energy consumption of the data center according to the acquired resource utilization rate and the ambient temperature includes: determining a load distribution cabinet according to the acquired resource utilization rate and the environmental temperature; The load distribution cabinet dispatches the energy consumption of the data center.

Preferably, determining the load distribution cabinet according to the acquired resource utilization and the ambient temperature includes: determining the acquired resource utilization of the one or more cabinets and the ambient temperature as sample values; according to the The sample value is used to predict the predicted value of resource utilization and the predicted value of ambient temperature of the one or more cabinets after the new load is added to the one or more cabinets; it is determined that when the predicted value of resource utilization is not overloaded, The cabinet corresponding to the smallest predicted ambient temperature value is a load distribution cabinet.

Preferably, after it is determined that the predicted value of resource utilization is not overloaded, and the cabinet corresponding to the minimum predicted value of ambient temperature is the load allocation cabinet, the method further includes: counting the actual obtained after the new load is added. Errors between the resource utilization rate and the ambient temperature value of the one or more cabinets and the predicted resource utilization rate and the predicted ambient temperature value; updating the sample value according to the error.

According to another aspect of the present invention, a data center energy consumption scheduling processing device is provided, including: an acquisition module, used to acquire the resource utilization rate and/or ambient temperature of one or more cabinets in the data center; a scheduling module, used to Scheduling the energy consumption of the data center according to the obtained resource utilization rate and/or ambient temperature.

Preferably, the scheduling module includes: a judging unit, used to judge whether the highest ambient temperature in the one or more cabinets acquired within a predetermined time period has decreased; If the result is yes, increase the air supply temperature of the cooling equipment in the data center.

Preferably, the scheduling module includes: a determining unit, configured to determine a load distribution cabinet according to the obtained resource utilization rate and the ambient temperature; a scheduling unit, configured to assign a load distribution cabinet to the data center according to the determined Energy consumption is scheduled.

Preferably, the determination unit includes: a first determination subunit, configured to determine the obtained resource utilization rate of the one or more cabinets and the ambient temperature as sample values; a prediction subunit, configured to value, predicting the resource utilization prediction value and the ambient temperature prediction value of the one or more cabinets after the new load is added to the one or more cabinets; the second determining subunit is used to determine the resource utilization prediction in the one or more cabinets If the value is not overloaded, the cabinet corresponding to the minimum predicted ambient temperature value is a load distribution cabinet.

Preferably, the determining unit further includes: a statistical subunit, configured to count the resource utilization rate and ambient temperature value of the one or more cabinets actually obtained after the new load is added, and the predicted resource utilization rate and the An error between predicted values of ambient temperature; an updating subunit, configured to update the sample value according to the error.

Through the present invention, the resource utilization rate and/or ambient temperature of one or more cabinets in the data center are obtained; and the energy consumption of the data center is scheduled according to the obtained resource utilization rate and/or ambient temperature, which solves the problem of related technologies There is an incomplete treatment of the energy consumption reduction of the data center, which leads to the problem that the energy consumption efficiency of the data center is not high, and then achieves not only the energy consumption scheduling that can realize the load distribution, but also beneficially reduces the cooling capacity of the data center to a certain extent. The effect of equipment energy consumption.

Description of drawings

The accompanying drawings described here are used to provide a further understanding of the present invention and constitute a part of the application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention. In the attached picture:

Fig. 1 is a flowchart of a data center energy consumption scheduling processing method according to an embodiment of the present invention;

Fig. 2 is a structural block diagram of a data center energy consumption scheduling processing device according to an embodiment of the present invention;

FIG. 3 is a preferred structural block diagram 1 of the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention;

FIG. 4 is a second preferred structural block diagram of the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention;

FIG. 5 is a structural block diagram of the determination unit 42 in the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention;

6 is a preferred structural block diagram of the determining unit 42 in the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention;

Fig. 7 is a schematic diagram of the state of each cabinet in the rack according to an embodiment of the present invention.

Detailed ways

Hereinafter, the present invention will be described in detail with reference to the drawings and examples. It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

In this embodiment, a data center energy consumption scheduling processing method is provided. FIG. 1 is a flowchart of a data center energy consumption scheduling processing method according to an embodiment of the present invention. As shown in FIG. 1 , the process includes the following steps:

Step S102, obtaining the resource utilization rate and/or ambient temperature of one or more cabinets in the data center, where the ambient humidity may be the inlet temperature of the one or more cabinets;

Step S104, scheduling the energy consumption of the data center according to the acquired resource utilization rate and/or ambient temperature.

Through the above steps, the utilization of energy consumption in the data center not only considers the factors of load balancing between servers, but also considers the environmental temperature factors of the entire data center. Compared with the related technologies that only involve one-sided influencing factors, it not only solves the problem of data problems in the related technologies. The energy consumption reduction of the center is not comprehensive, which leads to the problem of low energy consumption utilization efficiency of the data center, and then achieves energy scheduling that can not only realize load distribution, but also beneficially reduce the energy consumption of data center cooling equipment to a certain extent. Effect.

When scheduling the energy consumption of the data center based on the obtained ambient temperature (taking the inlet temperature of the cabinet as an example), it is judged whether the highest ambient temperature in one or more cabinets obtained within the predetermined time period has decreased; If the result is yes, increase the air supply temperature of the refrigeration equipment in the data center, that is, by increasing the air supply temperature of the refrigeration equipment, the energy consumption of the refrigeration equipment can be effectively reduced, and the energy consumption of the data center can be reduced to a certain extent.

When scheduling data center energy consumption based on the obtained resource utilization rate and ambient temperature, it also includes the load distribution processing of each cabinet. For example, the load distribution cabinet (that is, the newly added load The corresponding cabinet); according to the determined load distribution cabinet, the energy consumption of the data center is scheduled, that is, the newly added load distribution cabinet is the cabinet with the most reasonable load distribution (that is, the cabinet with the least load), so that a new The increased load is allocated to the cabinets to be overloaded, while the cabinets with less loads are in a state of wasting resources.

Preferably, the determination of the load distribution cabinet according to the obtained resource utilization rate and ambient temperature may adopt the following process, first determine the obtained resource utilization rate and ambient temperature of one or more cabinets as sample values, that is, record the current state of the cabinet, and Use the current state as a sample for load distribution by the prediction module; based on the sample value, predict the resource utilization and ambient temperature prediction values of one or more cabinets after adding loads to one or more cabinets, that is, the prediction will be new When the load is allocated to each cabinet, the impact on the new heat distribution, that is, to obtain the possible new state of each cabinet; determine that the cabinet corresponding to the minimum predicted value of ambient temperature is the load when the predicted value of resource utilization is not overloaded Allocate the cabinet, that is, select the cabinet corresponding to the minimum maximum inlet temperature in the new state as the optimal load distribution cabinet.

In order to improve the accuracy of prediction, after determining that the cabinet corresponding to the minimum predicted value of ambient temperature is the load distribution cabinet under the condition that the predicted value of resource utilization is not overloaded, it is also necessary to collect real data feedback, that is, to obtain the new load After being assigned to the above-mentioned optimal load distribution cabinet, the actual resource utilization rate and ambient temperature of each cabinet, and then, according to the actual obtained data statistics, the resource utilization rate and environment of one or more cabinets actually obtained after the new load is added The error between the temperature value and the predicted resource utilization rate and the predicted value of the ambient temperature; and then update the above-mentioned sample values for predicting and simulating decisions based on the error.

This embodiment also provides a data center energy consumption scheduling processing device, which is used to implement the above embodiments and preferred implementation modes, and what has already been described will not be repeated. As used below, the term "module" may be a combination of software and/or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementations in hardware, or a combination of software and hardware are also possible and contemplated.

Fig. 2 is a structural block diagram of a data center energy consumption scheduling processing device according to an embodiment of the present invention. As shown in Fig. 2, the device includes an acquisition module 22 and a scheduling module 24, and the device will be described below.

The obtaining module 22 is used to obtain the resource utilization rate and/or ambient temperature of one or more cabinets in the data center; the scheduling module 24 is connected to the above-mentioned obtaining module 22 and is used to process the data according to the obtained resource utilization rate and/or ambient temperature Central energy consumption is scheduled.

Fig. 3 is a preferred structural block diagram 1 of the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention. As shown in Fig. 3, the scheduling module 24 includes a judging unit 32 and an improving unit 34, the scheduling Module 24 is described.

The judging unit 32 is used to judge whether the highest ambient temperature in one or more cabinets acquired within a predetermined period of time has decreased; the raising unit 34 is connected to the judging unit 32 for judging the result of the judging unit 32 If yes, increase the supply air temperature of the data center cooling equipment.

Fig. 4 is a preferred structural block diagram 2 of the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention. As shown in Fig. 4, the scheduling module 24 includes a determining unit 42 and a scheduling unit 44, and the scheduling Module 24 is described.

The determination unit 42 is configured to determine the load distribution cabinet according to the obtained resource utilization rate and the ambient temperature; the scheduling unit 44 is connected to the determination unit 42 and used to schedule the energy consumption of the data center according to the determined load distribution cabinet.

Fig. 5 is a structural block diagram of the determining unit 42 in the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention. As shown in Fig. 5, the determining unit 42 includes a first determining subunit 52 and a predicting subunit 54 and the second determination subunit 56, the determination unit 42 will be described below.

The first determining subunit 52 is configured to determine the acquired resource utilization rate and ambient temperature of one or more cabinets as sample values; the predicting subunit 54 is connected to the above-mentioned first determining sub-unit 52 and is used to, based on the above-mentioned sample values, Predicting the predicted value of resource utilization and the predicted value of ambient temperature of one or more cabinets after the new load is added to one or more cabinets; the second determination subunit 56 is connected to the above-mentioned prediction subunit 54 for determining the resource utilization rate If the predicted value is not overloaded, the cabinet corresponding to the minimum predicted ambient temperature is the load distribution cabinet.

Fig. 6 is a preferred structural block diagram of the determining unit 42 in the scheduling module 24 in the data center energy consumption scheduling processing device according to an embodiment of the present invention. As shown in Fig. 6, the determining unit 42 includes all the modules shown in Fig. 5, It also includes a statistics subunit 62 and an update subunit 64, and the determination unit 42 will be described below.

The statistical subunit 62 is connected to the above-mentioned second determination subunit 56, and is used to count the difference between the resource utilization rate and ambient temperature value of one or more cabinets actually obtained after the new load is added, and the predicted resource utilization rate and ambient temperature prediction value. The error between them; the update subunit 64 is connected to the statistics subunit 62, and is used to update the sample value according to the above error.

Based on the related technology, the distribution of data center cabinets is the layout of cold and hot corridors, and the cold air provided by the refrigeration equipment is blown up from the underground of the "cold corridor" through the floor with seams (generally, there will be a gap between the floor and the ground of the data center). There is a separate layer for the circulation of cold air), enters from the entrance of the front cover of the cabinet, mixes with the hot air in the cabinet, and then discharges from the rear of the cabinet to take away the corresponding heat. Therefore, the corridor behind the cabinet is "hot air". Most of these "hot air" will be sucked from the top by the cooling equipment, but a small part will flow into the "cold corridor", which will affect the inlet temperature of the workstations in the cabinet. Generally speaking, the temperature of the workstation should be lower than a certain temperature specified by the equipment supplier in order to be in a safe working environment. For example, for the small data center used, the temperature can be 32 degrees Celsius. Different workstation equipment will There are different safe operating temperatures. At the same time, the inlet temperature of the workstation is also related to its current workload. The greater the workload, the more heat it will generate, and the heat that "cold air" can take away is certain; Thermal properties.

Therefore, the inlet temperature of each rack and even each cabinet in the data center will be different, and the previous cloud computing resource scheduling management system only considered the load balancing problem among servers, or only reduced the cloud from the perspective of virtual machine integration. The energy consumption of the data center is ignored, while the waste of cooling energy consumption caused by the unbalanced heat distribution of the data center is ignored. In this embodiment, through the intelligent scheduling of the data center load, the heat distribution of the data center is balanced, so that the air supply temperature of the refrigeration equipment can be increased, thereby reducing the cost of cooling energy consumption. Because at present, in order to ensure the safe operation of all equipment in the data center, the air supply temperature of the cooling equipment is related to the current maximum cabinet inlet temperature of the data center, so when the maximum cabinet inlet temperature of the data center decreases, the air supply temperature can be appropriately increased , thereby reducing the energy consumption of refrigeration equipment.

Aiming at the problem in related technologies that unbalanced heat distribution in the data center leads to excessive cooling energy consumption, a load scheduling method based on heat perception is provided in this embodiment, so that the heat distribution in the data center is as balanced as possible. As a result, the air supply temperature of the refrigeration equipment can be appropriately increased, thereby reducing the cost of refrigeration energy consumption.

The heat-sensing-based energy-saving method for a cloud data center includes the following steps:

Step S1, monitoring the physical environment of the data center: for example, the distribution of heat and the operation of refrigeration equipment. Monitor the inlet temperature of each rack inlet and each workstation in the cabinet in the data center. If the monitoring data shows that the indoor temperature has risen to a predetermined alarm value, it will automatically send reminders and alarm information to relevant personnel through the network, so that they can take timely measures to prevent computing equipment from working in a high temperature environment for a long time and affect the service life; and management The operator can quickly judge whether the equipment is running normally through the operation data of the refrigeration equipment. In addition, the collected information will be stored in a centralized network and an open interface will be provided for these data;

Step S2, data center layer load scheduling strategy based on heat perception: Based on the neural network algorithm model, through learning and modeling historical data, find the relationship between the total load of each rack and its inlet temperature, and then use the model to predict the new load The heat distribution that will be brought by the workstations assigned to different racks, by comparing the results of each heat distribution, select the optimal heat distribution results to determine which rack the new load is assigned to;

Step S3, load scheduling strategy based on heat perception at the blade cabinet layer: it is also a model based on neural network and reinforcement learning algorithm. Through learning and modeling historical data, find the relationship between the load of each cabinet in the rack and its inlet temperature, and then pass This model can predict the heat distribution that the new load will bring to different cabinets in the rack. By comparing the results of each heat distribution, the optimal heat distribution result is selected to determine which cabinet the new load is distributed to.

In step S1, in order to monitor the physical environment of the data center, such as the distribution of heat and the operation of refrigeration equipment, etc., temperature sensors are deployed in front of each rack and each cabinet, which can sense the temperature in each rack and cabinet. Inlet temperature of each workstation; In addition, sensors are placed in the refrigeration equipment of the data center to measure the temperature of the air-conditioning supply, the speed of the air conditioner, and the setting of the safe temperature. If the indoor temperature rises to a predetermined alarm value, it can automatically send reminders and alarm information to relevant personnel through the network, so that they can take timely measures to avoid affecting the service life of computing equipment working in a high temperature environment for a long time; the operating data of refrigeration equipment It allows the manager to quickly judge whether the device is running normally. In addition, the collected information will be stored in a centralized network and an open interface will be provided for these data.

The model prediction algorithm based on the neural network algorithm involved in step S2 and step S3 may include the following steps:

First of all, it is necessary to collect certain sample data, including the resource utilization rate of each cabinet at regular intervals, and the inlet temperature of the cabinet, and use <U _i (t), T _i (t)> to represent the state. Based on the neural network prediction model, it can quickly predict the state-behavior of the current environment,

$\left\{\begin{array}{c}<span\; class="notranslate">\; \{</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span><span\; class="notranslate">\; \}</span>\stackrel{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}}{<span\; class="notranslate">\; \&Right\; Arrow;</span>}\\ <span\; class="notranslate">\; \{</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ></span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ></span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ></span><span\; class="notranslate">\; \}</span>\end{array}\right.$

(1)

In the above formula (1), we can see how the neural network model predicts the new arrival load distribution behavior. The left side of the arrow is the state of the system environment at time t, and U _i (t) represents the resource utilization rate of workstation i at time t , T _i (t) represents the inlet temperature of workstation i at time t. The right side of the arrow is the predicted state of the system after the new load is distributed to workstation i, where T _i (t+1)′ is calculated by U _i (t+1)′ through the neural network prediction model, U _i (t+ 1)' can be considered equal to U _i (t) at the beginning of the t+1 period, and then updated according to the actual observation value at the end of the t+1 period. With these prediction results, it is convenient to find the highest value of the inlet temperature, which is found by comparing the results of various load distribution behaviors, so that it can be determined which workstation is the most suitable load receiving node. For the load migration behavior of the node with too high inlet temperature, the receiving node with the most suitable load can also be found through the same method. After making behavioral decisions and executions based on the prediction results, it is also necessary to collect real system feedback, that is, the maximum value of the real system environment inlet temperature, and add the collected values to the sample space, and then based on the new sample space at regular intervals To update the neural network prediction model to improve the accuracy of prediction.

Through the above embodiments and preferred implementation modes, not only the formulation of intelligent load distribution strategies based on the utilization rate of cabinet resources and inlet temperature is realized; but also the combination of neural network model and closed-loop control theory is realized, and the neural network prediction model is continuously updated to achieve Improve the prediction accuracy of heat distribution. Make the heat distribution of the data center as balanced as possible, thereby reducing the cost of cooling energy consumption, so as to achieve the purpose of energy saving.

Preferred embodiments of the present invention will be described below in conjunction with the accompanying drawings.

The energy-saving method of cloud data center based on heat perception includes the following steps:

Step S1, monitoring the resource utilization rate and inlet temperature of each cabinet: Real-time monitoring of the resource utilization rate and inlet temperature of each cabinet through the sensors deployed in the data center and the monitoring module running on each server, every once in a while, Calculate and record the current cabinet resource utilization rate and inlet temperature value <U _i (t), T _i (t)>. After all these information records are completed, they will be sent to the neural network prediction module for decision-making of load distribution;

Step S2, the formulation of heat-aware intelligent load distribution strategy: the neural network prediction module predicts the impact of the new load on the new heat distribution after the new load is distributed to each different cabinet according to the current state of each cabinet, that is, each The possible new state of the cabinet <U _i (t+1), T _i (t+1)>, compare and analyze the prediction results generated by different distribution options, and select the optimal load distribution, which is the highest inlet temperature in the new state The smallest option, FIG. 7 is a schematic diagram of the state of each cabinet in the rack according to an embodiment of the present invention, that is, the cabinet 3 is selected in FIG. 7 ;

Step S3, load distribution execution and post-execution state monitoring and collection: execute load distribution according to the decision result of step S2, and monitor and collect cabinet state information after load distribution at the same time. According to the observation results and the actual prediction results, the error statistics are carried out, and finally the sample data when the follow-up neural network prediction model is updated.

Step S4, adjust the air supply temperature of the refrigeration equipment: judge whether the air supply temperature of the refrigeration equipment can be properly adjusted according to the collected results of state monitoring, if the monitoring results show that the maximum inlet temperature has decreased, the air supply temperature of the refrigeration equipment can be appropriately increased , so that the energy consumption of cooling equipment can be reduced, and the energy consumption of the entire data center can be saved.

Wherein, the cabinet resource utilization rate in step S1 can be mainly represented by the usage of CPU resources;

The neural network prediction model in step S2 mainly can find the relationship between the resource utilization rate of each cabinet and the inlet temperature of the cabinet by analyzing and modeling the collected sample space, that is, <U ₁ (t), U ₂ (t ), U ₃ (t)……, U _i (t),…> and <T ₁ (t), T ₂ (t), T ₃ (t)…, T _i (t),…> The mapping relationship between, and then the selection process in step S2 is as follows:

① First, quantify the newly arrived load, such as how much CPU resources are needed. This process requires resource demand statistics for various types of loads.

②Predict the load distribution to each different cabinet i, <U ₁ (t), U ₂ (t), U ₃ (t)...U _i (t)+△u, ...> the possible heat distribution corresponding to< T ₁ (t+1), T ₂ (t+1), T ₃ (t+1), ..., T _i (t+1), ...>, and find the maximum inlet temperature MaxT _i (t+1) .

③According to the maximum inlet temperature {MaxT _i (t+1)} found by assigning to different cabinet i, and then find the minimum value Min{MaxT _i (t+1)}, the i corresponding to the minimum value is the final cabinet of choice.

What is used in step S3 is the online reinforcement learning technology, which improves the accuracy of the model through continuous learning.

The measure of adjusting and controlling the air supply temperature of the refrigeration equipment in step S4 is mainly related to the interface provided by the equipment supplier, and this control process can be completed through software control.

Obviously, those skilled in the art should understand that each module or each step of the above-mentioned present invention can be realized by a general-purpose computing device, and they can be concentrated on a single computing device, or distributed in a network formed by multiple computing devices Alternatively, they may be implemented in program code executable by a computing device so that they may be stored in a storage device to be executed by a computing device, and in some cases in an order different from that shown here The steps shown or described are carried out, or they are separately fabricated into individual integrated circuit modules, or multiple modules or steps among them are fabricated into a single integrated circuit module for implementation. As such, the present invention is not limited to any specific combination of hardware and software.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. A data center energy consumption scheduling processing method, characterized in that, comprising: Obtain resource utilization and/or ambient temperature for one or more cabinets in the data center; Scheduling the energy consumption of the data center according to the obtained resource utilization rate and/or ambient temperature. 2. The method according to claim 1, wherein scheduling the energy consumption of the data center according to the acquired ambient temperature comprises: judging whether the highest ambient temperature in the one or more cabinets obtained within a predetermined period of time has decreased; If the judgment result is yes, increase the air supply temperature of the data center refrigeration equipment. 3. The method according to claim 1, wherein scheduling the energy consumption of the data center according to the acquired resource utilization rate and the ambient temperature comprises: Determine the load distribution cabinet according to the acquired resource utilization rate and the ambient temperature; The energy consumption of the data center is scheduled according to the determined load distribution cabinets. 4. The method according to claim 1, wherein determining the load distribution cabinet according to the acquired resource utilization rate and the ambient temperature comprises: Determining the acquired resource utilization rate of the one or more cabinets and the ambient temperature as sample values; According to the sample value, predicting the predicted value of the resource utilization rate and the predicted value of the ambient temperature of the one or more cabinets after the new load is added to the one or more cabinets; It is determined that in a case where the predicted value of resource utilization is not overloaded, the cabinet corresponding to the minimum predicted value of ambient temperature is a load distribution cabinet. 5. The method according to claim 4, wherein after it is determined that the cabinet corresponding to the minimum predicted ambient temperature value is the load distribution cabinet when the predicted value of resource utilization is not overloaded, further include: Counting the errors between the resource utilization rate and the ambient temperature value of the one or more cabinets actually obtained after adding the load and the predicted resource utilization rate and the predicted value of the ambient temperature; The sample value is updated according to the error. 6. A data center energy consumption scheduling processing device, characterized in that it comprises: An acquisition module, configured to acquire resource utilization and/or ambient temperature of one or more cabinets in the data center; A scheduling module, configured to schedule the energy consumption of the data center according to the obtained resource utilization rate and/or ambient temperature. 7. The device according to claim 6, wherein the scheduling module comprises: a judging unit, configured to judge whether the highest ambient temperature in the one or more cabinets obtained within a predetermined period of time has decreased; The increasing unit is configured to increase the air supply temperature of the data center refrigeration equipment when the judgment result of the above judgment unit is yes. 8. The device according to claim 6, wherein the scheduling module comprises: A determining unit, configured to determine a load distribution cabinet according to the acquired resource utilization rate and the ambient temperature; A scheduling unit, configured to schedule the energy consumption of the data center according to the determined load distribution cabinet. 9. The device according to claim 6, wherein the determining unit comprises: A first determining subunit, configured to determine the acquired resource utilization rate of the one or more cabinets and the ambient temperature as sample values; The prediction subunit is used to predict the predicted resource utilization value and the predicted value of the ambient temperature of the one or more cabinets after the new load is added to the one or more cabinets according to the sample value; The second determining subunit is configured to determine that the cabinet corresponding to the minimum predicted ambient temperature value is a load distribution cabinet when the predicted value of resource utilization is not overloaded. 10. The device according to claim 9, wherein the determining unit further comprises: The statistical subunit is used to count the error between the resource utilization rate and the ambient temperature value of the one or more cabinets actually obtained after adding the load and the predicted resource utilization rate and the predicted value of the ambient temperature; An updating subunit, configured to update the sample value according to the error.