WO2021046774A1

WO2021046774A1 - Resource scheduling method and information prediction method, device, system, and storage medium

Info

Publication number: WO2021046774A1
Application number: PCT/CN2019/105455
Authority: WO
Inventors: 奉有泉; 卢毅军; 李栈; 陶原; 赵旭; 陈钢; 宋军
Original assignee: 阿里巴巴集团控股有限公司
Priority date: 2019-09-11
Filing date: 2019-09-11
Publication date: 2021-03-18
Also published as: CN113826078B; CN113826078A

Abstract

Embodiments of the present application provide a resource scheduling method and an information prediction method, a device, a system, and a storage medium. In the embodiments of the present application, the prediction of costs is combined with resource scheduling, the resource scheduling based on the costs is achieved by predicting the costs of a schedulable unit, priority scheduling of resources having low costs can be achieved, and the probability of occurrence of local hot spots can also be reduced, facilitating reducing the overall power consumption. In addition, during prediction of the costs, the contribution of the power consumption of an infrastructure on which the schedulable unit depends in the costs of the schedulable unit is considered, the predicted costs can be more reasonable and accurate, thereby improving the reasonability of the resource scheduling based on the costs.

Description

Resource scheduling and information prediction method, equipment, system and storage medium

Technical field

This application relates to the technical field of computer room systems or data center systems, and in particular to a method, equipment, system and storage medium for resource scheduling and information prediction.

Background technique

With the development of cloud computing technology, various computer room systems or data center systems (DC) are constantly being deployed. The pressure on computer room systems or data center systems continues to increase, and the requirements for service quality are getting higher and higher. As a result, the energy consumption of computer room systems or data center systems has become increasingly prominent.

On the issue of how to reduce the energy consumption of the computer room system or data center system, one solution is to configure a power consumption control strategy for the server of the computer room system or data center system. Based on the generalization of power consumption control strategies, it is necessary to provide new solutions.

Summary of the invention

Various aspects of the present application provide a resource scheduling and information prediction method, equipment, system, and storage medium to reduce the power consumption of a data center system or a computer room system.

An embodiment of the present application provides a resource scheduling method, including: acquiring current data corresponding to a cost impact factor of at least one schedulable unit, the current data including the at least one schedulable unit and the infrastructure on which it depends According to the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price, predict the The cost of at least one schedulable unit; and perform resource scheduling on the at least one schedulable unit according to the cost of the at least one schedulable unit.

An embodiment of the present application also provides an information prediction method, including: acquiring current data corresponding to a cost impact factor of a target schedulable unit, the current data including information about the target schedulable unit and the infrastructure on which it depends Current power consumption data; according to the current power consumption data of the target schedulable unit and the infrastructure on which it depends, and the influence relationship between the pre-trained cost cost influencing factor and the cost cost, predict that the target can be The cost of the dispatch unit.

An embodiment of the present application also provides a resource scheduling device, including: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program for use For: acquiring current data corresponding to the cost impact factor of at least one schedulable unit, where the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends; according to the at least one schedulable unit; The current power consumption data of the schedulable unit and the infrastructure on which it depends, as well as the influence relationship between the pre-trained cost cost influencing factor and the cost cost, predict the cost cost of the at least one schedulable unit; According to the cost of the at least one schedulable unit, resource scheduling is performed on the at least one schedulable unit.

An embodiment of the present application also provides an information prediction device, including: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program for use In: obtaining current data corresponding to the cost impact factor of the target schedulable unit, the current data including current power consumption data of the target schedulable unit and the infrastructure on which it depends; according to the target schedulable unit The current power consumption data of the infrastructure and the infrastructure on which it depends, as well as the impact relationship between the pre-trained cost cost influencing factor and the cost cost, predict the cost cost of the target schedulable unit.

The embodiment of the present application also provides a computer room system, including: several schedulable units, several infrastructures, and resource scheduling equipment; the several infrastructures provide basic services for the several schedulable units; the resource scheduling A device for acquiring current data corresponding to the cost impact factor of at least one schedulable unit, where the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends; according to the Current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, and the pre-trained cost impact factor and the impact relationship between the cost price, and predict the cost price of the at least one schedulable unit; Perform resource scheduling on the at least one schedulable unit according to the cost of the at least one schedulable unit; wherein the at least one schedulable unit comes from the several schedulable units.

The embodiment of the present application also provides another computer room system, including: several schedulable units, several infrastructures, information prediction equipment, and resource scheduling equipment; the several infrastructures provide basic services for the several schedulable units The information prediction device is used to obtain current data corresponding to the cost impact factor of at least one schedulable unit, the current data including the current power consumption of the at least one schedulable unit and the infrastructure on which it depends Data; predict the at least one schedulable unit based on the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price The cost of the unit; the cost of the at least one schedulable unit is provided to the resource scheduling device; the resource scheduling device is used to predict the cost of the at least one schedulable unit provided by the device according to the information , Perform resource scheduling on the at least one schedulable unit; wherein the at least one schedulable unit comes from the several schedulable units.

The embodiment of the present application also provides a data center system, including: at least one computer room and resource scheduling equipment; each computer room includes a number of schedulable units and a number of infrastructures, and the number of infrastructures is the number of schedulable The unit provides basic services; the resource scheduling device is used to obtain current data corresponding to the cost impact factor of at least one schedulable unit, and the current data includes the at least one schedulable unit and the infrastructure on which it depends According to the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price, predict the The cost of at least one schedulable unit; according to the cost of the at least one schedulable unit, perform resource scheduling on the at least one schedulable unit; wherein the at least one schedulable unit comes from the at least one computer room The schedulable unit.

The embodiment of the present application also provides another data center system, including: at least one computer room, information prediction equipment, and resource scheduling equipment; each computer room includes a number of schedulable units and a number of infrastructures. The several schedulable units provide basic services; the information prediction device is used to obtain the current data corresponding to the cost impact factor of at least one schedulable unit, and the current data includes the at least one schedulable unit and its Current power consumption data of the dependent infrastructure; according to the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, and the pre-trained cost impact factor and the impact of the cost price Relationship, predict the cost of the at least one schedulable unit; provide the cost of the at least one schedulable unit to the resource scheduling device; the resource scheduling device is used to predict the cost of the at least one schedulable unit based on the information According to the cost of the at least one schedulable unit, resource scheduling is performed on the at least one schedulable unit; wherein the at least one schedulable unit comes from the schedulable unit in the at least one schedulable unit.

An embodiment of the present application also provides an edge computing system, including: several edge computing nodes, several infrastructures and servers; the several infrastructures provide basic services for the several edge computing nodes;

The server is configured to obtain current data corresponding to the cost impact factor of at least one edge computing node, where the current data includes current power consumption data of the at least one edge computing node and the infrastructure on which it depends; Describe the current power consumption data of the at least one edge computing node and the infrastructure on which it depends, and the pre-trained cost impact factor and the impact relationship between the cost price, and predict the cost price of the at least one edge computing node; According to the cost of the at least one edge computing node, resource scheduling is performed on the at least one edge computing node; wherein the at least one edge computing node comes from the plurality of edge computing nodes.

The embodiments of the present application also provide a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor is caused to implement the steps in the method embodiments of the present application.

In the embodiment of the present application, the cost cost prediction is combined with resource scheduling. By predicting the cost cost of the schedulable unit, the cost cost-based resource scheduling is realized, which can prioritize the scheduling of resources with lower cost costs and reduce occurrences. The probability of local hotspots is conducive to reducing the overall power consumption; in addition, in the cost cost prediction process, the contribution of the power consumption of the infrastructure on which the schedulable unit depends on the cost of the schedulable unit can be considered to make the predicted cost The cost price is more reasonable and accurate, and the rationality of resource scheduling based on the cost price can be improved.

Description of the drawings

The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application, and do not constitute an improper limitation of the application. In the attached picture:

FIG. 1 is a schematic structural diagram of a computer room system provided by an exemplary embodiment of this application;

FIG. 2a is a block diagram of a working principle of a resource scheduling device provided by an exemplary embodiment of this application;

2b is a block diagram of another working principle of the resource scheduling device provided by an exemplary embodiment of this application;

FIG. 3 is a schematic structural diagram of a data center system provided by an exemplary embodiment of this application;

4 is a schematic structural diagram of another computer room system provided by an exemplary embodiment of this application;

Fig. 5a is a block diagram of a working principle of an information prediction device provided by an exemplary embodiment of this application;

FIG. 5b is a block diagram of another working principle of the information prediction device provided by an exemplary embodiment of this application;

Fig. 6a is a schematic structural diagram of another data center system provided by an exemplary embodiment of this application;

FIG. 6b is a schematic structural diagram of an edge computing system provided by an exemplary embodiment of this application;

FIG. 7a is a schematic flowchart of a resource scheduling method provided by an exemplary embodiment of this application;

FIG. 7b is a schematic flowchart of a model training method provided by an exemplary embodiment of this application;

FIG. 7c is a schematic flowchart of an information prediction method provided by an exemplary embodiment of this application;

FIG. 8 is a schematic structural diagram of a resource scheduling device provided by an exemplary embodiment of this application;

FIG. 9 is a schematic structural diagram of an information prediction device provided by an exemplary embodiment of this application.

detailed description

In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described clearly and completely in conjunction with specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

In view of the technical problem of how to reduce the energy consumption of the computer room system or the data center system, in some embodiments of the present application, the prediction of the cost price is combined with the resource scheduling, and the cost price of the schedulable unit is predicted to realize the resource scheduling based on the cost price. , It can prioritize resources with lower cost, and can reduce the probability of local hot spots, which is helpful to reduce the overall power consumption. In the cost prediction process, the power consumption of the infrastructure on which the schedulable unit depends can be considered at the same time. The contribution of the cost cost of the scheduling unit can make the predicted cost cost more reasonable and accurate, and further can improve the rationality of resource scheduling based on the cost cost.

The technical solutions provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

FIG. 1 is a schematic diagram of the structure and working principle of a computer room system provided by an exemplary embodiment of this application. As shown in FIG. 1, the computer room system 100 of this embodiment includes: a computer room. The computer room refers to a physical place where machinery and equipment are stored, for example, a room or a factory building. The computer room system 100 can be built separately or placed in other buildings. Further, as shown in FIG. 1, the computer room system 100 further includes: at least one physical device 101 and a resource scheduling device 102 located in the computer room. This embodiment does not limit the number of physical devices 101, and it may be one or multiple.

In this embodiment, the device form of the physical device 101 is not limited. As shown in FIG. 1, the physical device 101 may be an IT-type device in a computer room and a refrigeration device for cooling the IT-type device, such as an air-conditioning device. For example, the at least one physical device 101 may include, but is not limited to: cabinet equipment, server equipment, terminal equipment, printers, hubs, power supply equipment, storage equipment, network switching equipment, air conditioning equipment, and so on. The server device may include, but is not limited to, a conventional server, a server array, or a cloud server. The power supply equipment can be storage battery equipment, dry battery equipment, or uninterruptible power supply (UPS). Storage devices may include, but are not limited to: disks, disk arrays, hard disks, network storage devices (NAS), and so on.

Similarly, this embodiment does not limit the device form of the resource scheduling device 102 either. The resource scheduling device can be any computer device with certain resource scheduling capabilities, such as terminal devices such as smart phones, laptops, desktop computers, tablets, or smart watches, or server devices such as conventional servers, cloud servers, or server arrays .

In this embodiment, from the perspective of resource scheduling, the physical device 101 in the computer room is divided into several schedulable units and several infrastructures. In the embodiments of the present application, "several" means an indeterminate number, which may mean one or more. Among them, a schedulable unit refers to a resource device, module, or unit that can provide a certain service, function, or resource (for example, computing, storage, or network) to the outside in the computer room. For example, the schedulable unit may be a physical device such as a server or printer in the computer room, or a hardware module such as a CPU, GPU, memory, network card, or hard disk on the server, or a virtual machine running on a physical device, Software modules or units such as cloud services. Generally speaking, physical servers are the most common schedulable unit. Infrastructure refers to the facilities in the computer room that provide basic services for dispatchable units. According to the different schedulable units, the infrastructure will be different, which is not limited. Take the server as an example. The schedulable unit is a server. The infrastructure can include cabinets that provide support services for the servers, air-conditioning equipment, fans or water-cooling equipment for cooling the servers, power equipment for powering the servers, switches that provide network connections for the servers, etc. .

Among them, the schedulable unit is used as an available resource in the computer room system 100, and the schedulable unit can be reasonably scheduled for resources through the resource scheduling device 102. Reasonable resource scheduling is of great significance for improving the resource utilization efficiency of the computer room system 100, saving energy, improving resource sharing, and reducing operating costs. In this embodiment, resource scheduling is a process of allocating schedulable units to users who need resources. For scenarios where resource overload (demand is greater than system capacity) or scenarios where demand and capacity change dynamically over time, the resource scheduling device 102 may dynamically reallocate resources so as to use available resources more efficiently. For example, the resource scheduling device 102 may allocate certain resource-demanding user applications to certain schedulable units or units to run. For another example, the resource scheduling device 102 may allocate resources such as CPU, storage, or network bandwidth provided by a certain schedulable unit or units to certain resource demand users, and the resource demand users use these resources according to actual conditions. The "application program" in the embodiments of the present application refers to various programs, codes, software, business systems, or application systems in a broad sense.

In this embodiment, the resource scheduling device 102 simultaneously considers the cost of the schedulable unit during resource scheduling of the schedulable unit. Resource scheduling based on cost can not only prioritize resources with lower cost, but also reduce the probability of local hot spots, which is beneficial to reducing the overall power consumption of the computer room system 100. Among them, in order to use the cost price of the schedulable unit, the resource scheduling device 102 also has a cost price prediction capability for predicting the cost price of the schedulable unit. Among them, the physical meaning represented by the "cost of the schedulable unit" can be flexibly defined according to application requirements, which is not limited in this embodiment. For example, the cost of the schedulable unit may be the cost of power consumption, or the cost of maintenance, or the cost of configuration, and so on. In this embodiment, the description is focused on the cost of power consumption as an example.

Among them, the direct factor that has an impact on the cost is the power consumption of the schedulable unit itself, in addition to many other factors. In this embodiment, the factor of power consumption of the infrastructure on which the schedulable unit depends is mainly considered, and the power consumption of the infrastructure on which the schedulable unit depends is converted into the cost of the schedulable unit, which can improve resource scheduling. rationality. In this embodiment, the factors that have an impact on the cost of the schedulable unit are collectively referred to as the cost impact factor. In this embodiment, the cost impact factor includes at least two. For example, the power consumption of the schedulable unit itself is a cost impact factor, and the power consumption of the infrastructure on which the schedulable unit depends is another cost impact factor. . Among these cost impact factors, the value of some impact factors changes dynamically, and the value of some impact factors may not change. Regardless of whether the value of the impact factor changes, in the embodiments of the present application, the value of the cost impact factor at the current moment or the current time period is referred to as the current data corresponding to the cost impact factor; the cost impact factor is in the history The value in the time or historical period is called the historical data corresponding to the cost impact factor. According to the different cost impact factors, the data content contained in current data and historical data will also be different.

For the resource scheduling device 102, at least one schedulable unit can be determined from the schedulable units included in the computer room system 100 during each resource scheduling process. The at least one schedulable unit may be part of the schedulable unit included in the computer room system 100, or may be all schedulable units included in the computer room system 100. Then, the resource scheduling device 102 obtains the current data corresponding to the cost impact factor of at least one schedulable unit. In this embodiment, the cost impact factor used includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit depends. Based on this, the current data includes: Current power consumption data of the dependent infrastructure. Then, the resource scheduling device 102 can predict at least one schedulable unit and the current power consumption data of the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price. The cost of the dispatch unit. Finally, the resource scheduling device 102 may perform resource scheduling on at least one schedulable unit according to the cost of the at least one schedulable unit.

In this embodiment, the manner in which the resource scheduling device 102 determines at least one schedulable unit from the schedulable units included in the computer room system 100 is not limited. For example, taking this resource scheduling as an example, the resource scheduling device 102 may determine at least one schedulable unit that is schedulable this time from the schedulable units included in the computer room system 100 according to the resource demand corresponding to this resource scheduling. The resource requirements corresponding to this resource scheduling include some requirements of this resource scheduling for schedulable units, such as type, location, or capacity, and schedulable units that meet these requirements can participate in this resource scheduling.

Further, in addition to considering the resource requirements corresponding to this resource scheduling, at least one schedulable unit that can be scheduled this time can be determined from the schedulable units included in the computer room system 100 in combination with the device topology relationship of the computer room system 100. The device topology relationship of the computer room system 100 describes the topology relationship between the physical devices 101 in the computer room system 100, and the distance between each schedulable unit and the resource scheduling device 102 can be clearly known through the topology relationship. Based on this, according to the device topology relationship of the computer room system 100, at least one schedulable unit whose distance to the resource scheduling device performing this resource scheduling meets the set requirement can be determined from among the schedulable units included in the computer room system 100. The distance here can be a physical distance or a network transmission distance. This embodiment does not limit the "setting requirements", and can be flexibly set according to application requirements. The following example illustrates:

Example 1: The setting requirement refers to being located in the same computer room. According to the device topology of the computer room system 100, at least one of the schedulable units included in the computer room system 100 can be determined to be located in the same computer room as the resource scheduling device 102. Scheduling unit. To put it simply, it is to obtain a schedulable unit that meets the resource demand corresponding to this resource scheduling from each schedulable unit included in the computer room system 100.

Example 2: The setting requirement refers to being located in the same channel. According to the device topology relationship of the computer room system 100, at least one of the schedulable units included in the computer room system 100 can be determined to be located in the same channel as the resource scheduling device 102. Scheduling unit. The computer room system 100 includes a number of channels, each channel includes a number of cabinets, and each cabinet includes a number of servers. The resource scheduling device 102 is deployed in a certain channel. To put it simply, it is to obtain a schedulable unit that meets the resource requirements corresponding to this resource scheduling in the schedulable unit located in the channel where the resource scheduling device 102 is located.

Example 3: The setting requirement refers to being located in the same cabinet. According to the device topology of the computer room system 100, it can be determined from the schedulable units included in the computer room system 100 that at least one schedulable unit is located in the same cabinet as the resource scheduling device 102. Scheduling unit. The computer room system 100 includes a number of channels, each channel includes a number of cabinets, and each cabinet includes a number of servers. The resource scheduling device 102 is deployed in a certain cabinet. To put it simply, it is deployed in the schedulable unit in the cabinet where the resource scheduling device 102 is located, and obtains the schedulable unit that meets the resource requirements corresponding to this resource scheduling.

Example 4: The set requirement means that the physical distance is less than the set distance threshold, and the physical distance to the resource scheduling device 102 can be determined from the schedulable units included in the computer room system 100 according to the device topology relationship of the computer room system 100 At least one schedulable unit less than the set distance threshold.

In any manner, after determining at least one schedulable unit, the resource scheduling device 102 may obtain the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends. Optionally, the resource scheduling device 102 may obtain the current power consumption data of each physical device 101 in the computer room system 100, and analyze the current power consumption data of each schedulable unit in the computer room system 100 and the current power consumption data of each infrastructure; Then, the current power consumption data of at least one schedulable unit that can be schedulable this time is selected from the current power consumption data of each schedulable unit, and combined with the device topology relationship of the computer room system 100, the current power consumption data of each infrastructure Select the current power consumption data of the infrastructure on which at least one schedulable unit each depends.

In some embodiments of the present application, the cost prediction model corresponding to each schedulable unit can be pre-trained offline, and the cost cost prediction model corresponding to each schedulable unit can reflect the cost impact factor and cost corresponding to the schedulable unit itself. Based on this, the cost prediction model corresponding to at least one schedulable unit is used to predict the cost on the line to obtain the cost of at least one schedulable unit.

Wherein, FIG. 2a shows a block diagram of a working principle of the resource scheduling device 102. Referring to FIG. 2a, the resource scheduling device 102 may obtain historical data corresponding to the cost impact factor of at least one schedulable unit, and the historical data includes historical power consumption data of the at least one schedulable unit and the infrastructure on which it depends. The historical power consumption data here includes power consumption data at multiple historical moments or historical time periods; model training is performed according to the historical power consumption data of at least one schedulable unit and the infrastructure on which it depends to obtain at least one schedulable unit Each corresponding cost price prediction model. Further, referring to Figure 2a, in the resource scheduling process, the resource scheduling device 102 obtains the current power consumption data of at least one schedulable unit and the infrastructure on which it depends, and compares the current power consumption data of at least one schedulable unit and the infrastructure on which it depends. The current power consumption data of each is input into the corresponding cost cost prediction model for cost cost prediction to obtain the cost cost of at least one schedulable unit; and then based on the cost cost of the at least one schedulable unit, resource scheduling is performed on at least one schedulable unit .

Among them, the cost cost prediction model shown in FIG. 2a mainly reflects the influence relationship of the cost cost influence factor of the power consumption of the infrastructure on which the schedulable unit depends on the cost cost of the schedulable unit. Optionally, the influence relationship can be expressed as the following formula (1), but it is not limited to this.

In the formula (1), P _i represents the power consumption data may depend on the scheduling unit i-th infrastructure; n-represents the total number of units may depend infrastructure scheduling, n being a positive integer; beta] _i represents the i The weight system corresponding to the power consumption data of each infrastructure can reflect the contribution of the power consumption of the i-th infrastructure to the cost of the schedulable unit; P ₀ represents the power consumption data of the schedulable unit; P _v represents the schedulable unit The cost of the unit.

Based on the influence relationship of the cost influence factor of the infrastructure power consumption expressed by the above formula (1) on the cost cost of the schedulable unit, taking the target schedulable unit as an example, the process of predicting the cost of the resource scheduling device 102 will be described. Wherein, the target schedulable unit is any schedulable unit among at least one schedulable unit.

For the target schedulable unit, the resource scheduling device 102 inputs the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model corresponding to the target schedulable unit; the cost prediction model corresponding to the target schedulable unit Inside the model, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the first type of weight coefficients that have been trained, and the result of the weighted summation is combined with the current power consumption data of the target schedulable unit Add them to get the cost of the target schedulable unit. In order to facilitate the distinction and description, the weighting system β _i in formula (1) is called the first type of weighting coefficient; the first type of weighting system reflects the impact of the cost impact factor of infrastructure power consumption on the cost of schedulable units. influences.

It should be noted that the factors that affect the cost of the schedulable unit, that is, the cost impact factor is in addition to the power consumption of the schedulable unit itself (referred to as its own power consumption) and the power consumption of the infrastructure on which the schedulable unit depends (referred to as Infrastructure power consumption) In addition to these two, there are many others, such as the price of electricity, the type of application running on the schedulable unit (referred to as the application type), and the physical location of the schedulable unit (referred to as the physical location) Wait. Among them, different types of applications have different power consumption usage patterns, which are generally required in advance in the SLA.

Based on the above, in some other embodiments of the present application, in addition to considering the two cost impact factors of the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit depends, the cost of the schedulable unit can also be considered. At least one of the physical location, the price of electricity in the area where the schedulable unit is located, and the type of program running on the schedulable unit, and other cost impact factors. Among them, the type of programs running on the schedulable unit can reflect the power consumption of the schedulable unit to a certain extent. For example, some types of applications consume more power, such as applications that are used more frequently or frequently, or some audio and video playback applications; some types of applications consume relatively less power, such as those that are more frequently used. Few applications, or some text processing applications. The electricity price of the area where the dispatchable unit is located is different, and the cost of the dispatchable unit will also be different. The price of electricity in some areas is relatively high. Even if the power consumption of schedulable units (such as servers) deployed in these areas is not high, the cost of electricity may be higher when combined with the price of electricity; the price of electricity in some areas is relatively high. Low, even if the power consumption of schedulable units (such as servers) deployed in these areas is relatively high, the cost price may be relatively low when combined with the price of electricity. In addition to regions, the price of electricity may also change over time, and fluctuations in the price of electricity will affect the cost of dispatchable units.

As shown in FIG. 2b, it is a block diagram of another working principle of the resource scheduling device 102. To simplify the illustration, the working principle of the resource scheduling device 102 is described by taking the target schedulable unit as an example in FIG. 2b.

Referring to FIG. 2b, for the target schedulable unit, the resource scheduling device 102 can obtain historical data corresponding to the cost impact factor of the target schedulable unit. In this embodiment, the cost impact factor includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit is based, as well as at least one of physical location, electricity price, and application type, which are affected by these costs. The historical data corresponding to the factor includes: historical power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and the target schedulable unit At least one type of data from the type of historical program running on.

Further, the resource scheduling device 102 is based on the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and the target schedulable unit Model training is performed on at least one type of data in the historical program type running on it, and a cost prediction model corresponding to the target schedulable unit is obtained.

Further, referring to FIG. 2b, in the resource scheduling process, the resource scheduling device 102 may obtain the current data corresponding to the cost impact factor of the target schedulable unit. In this embodiment, the cost impact factor includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit is based, as well as at least one of physical location, electricity price, and application type, which are affected by these costs. The current data corresponding to the factor includes: the current power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the target schedulable unit At least one type of data in the type of program that needs to be run. After that, the resource scheduling device 102 combines the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the target schedulable unit At least one data of the program types that need to be run is input into the cost prediction model corresponding to the target schedulable unit to perform cost cost prediction to obtain the cost of the target schedulable unit.

Among them, the cost prediction model shown in Figure 2b mainly reflects the power consumption of the infrastructure on which the schedulable unit depends, the price of electricity, the types of applications running on the schedulable unit, and the physical location of the schedulable unit. The influence of factors on the cost of schedulable units. Optionally, the influence relationship can be expressed as the following formula (2), but it is not limited to this.

In equation (2), P _i represents the power consumption data may depend on the scheduling unit infrastructure i-th, n is the total number of cells showing depends infrastructure schedulable, n being a positive integer; beta] _i represents the i The weighting system corresponding to the power consumption data of the infrastructure can reflect the contribution of the power consumption of the i-th infrastructure to the cost of the schedulable unit, which belongs to the first type of weighting system; P ₀ represents the power consumption of the schedulable unit Data; P _v represents the cost of the schedulable unit; α represents the second type of weight system. Among them, the first type of weight coefficient β _i reflects the impact of the cost of infrastructure power consumption on the cost of schedulable units, and the second type of weight coefficient α reflects at least one of the physical location, electricity price, and program type. The impact of cost impact factor on the cost of schedulable units.

Based on the influence relationship of multiple cost cost influencing factors on the cost cost of the schedulable unit expressed by the above formula (2), still taking the target schedulable unit as an example, the process of predicting the cost of the resource scheduling device 102 will be described.

For the target schedulable unit, the resource scheduling device 102 calculates the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and Data of at least one of the types of programs to be run on the target schedulable unit is input into the cost prediction model corresponding to the target schedulable unit. In the cost prediction model corresponding to the target schedulable unit, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the first type of weight coefficients that have been trained, and the weighted sum is the result Add the current power consumption data of the target schedulable unit to obtain the initial cost of the target schedulable unit; and use the second type of weight coefficient that has been trained to correct the initial cost to obtain the cost of the target schedulable unit .

In the influence relationship shown in formula (2), the method of using the second type of weight coefficient to modify the initial cost is to calculate the product of the second type of weight coefficient and the initial cost as the cost of the target schedulable unit. However, it should be noted that the method of using the second type of weight coefficient to modify the initial cost is not limited to the formula (2), which is not limited in the embodiment of the present application.

In summary, in the computer room system 100 provided by the embodiment of the present application, the resource scheduling device 102 combines the prediction of cost and resource scheduling. By predicting the cost of schedulable units, resource scheduling based on cost can be realized. Priority scheduling of resources with lower cost costs can also reduce the probability of local hot spots and help reduce overall power consumption; in addition, in the cost cost prediction process, the power consumption of the infrastructure on which the schedulable unit depends is considered in the schedulable The contribution in the cost of the unit can make the predicted cost more reasonable and accurate, and thus can improve the rationality of resource scheduling based on the cost.

It should be noted that the resource scheduling scheme that combines cost prediction and resource scheduling provided by the embodiments of the present application is not only used for computer room systems, but also other systems with resource scheduling requirements, such as data center systems.

FIG. 3 is a schematic structural diagram of a data center system provided by an exemplary embodiment of this application. As shown in FIG. 3, the data center system 300 includes: at least one computer room 301 and a resource scheduling device 302. The data center system 300 can be built separately, or placed in other buildings, or distributed in different geographical locations. Among them, each computer room 301 includes at least one physical device 303. The number of physical devices contained in each computer room 301 may be one or multiple.

The computer room 301 in this embodiment is similar to the computer room in the embodiment shown in FIG. 1. For the related description of the physical device 303 and the computer room 301, please refer to the description in the embodiment shown in FIG. 1, which will not be repeated here.

In this embodiment, from the perspective of resource scheduling, the physical equipment 303 in each computer room 301 is divided into several schedulable units and several infrastructures. In the embodiments of the present application, "several" means an indeterminate number, which may mean one or more. Among them, the schedulable unit refers to the resource equipment, module or unit that can provide a certain service, function or resource (for example, computing, storage or network) in the computer room. For example, the schedulable unit can be a physical device such as a server, a printer, or a switch in the computer room, or a hardware module such as a CPU, GPU, memory, network card, or hard disk on the server, or a virtual device running on a physical device. Software modules or units such as computers and cloud services. Infrastructure refers to the facilities in the computer room that provide basic services for dispatchable units. According to the different schedulable units, the infrastructure will be different, which is not limited. Taking the server as the schedulable unit as an example, the infrastructure may include cabinets that provide support services for the servers, air-conditioning equipment or water-cooling equipment for cooling the servers, power equipment for supplying power to the servers, and so on.

The difference between this embodiment and the embodiment shown in FIG. 1 is that the resource scheduling device 302 does not belong to a certain computer room, but belongs to the entire data center system 300, and resource scheduling of schedulable units in each computer room 301 is required. Although the resource scheduling device 302 does not belong to any computer room, in terms of physical deployment, it can be deployed in a certain computer room or independently deployed outside each computer room.

In the same way, the resource scheduling device 302 in this embodiment simultaneously considers the cost of the schedulable unit during resource scheduling of the schedulable unit. For the resource scheduling device 302, at least one schedulable unit can be determined from the schedulable units included in the data center system 300 during each resource scheduling process. The at least one schedulable unit may be part of the schedulable unit included in the data center system 300, or may be all schedulable units included in the data center system 300. In addition, it should be noted that at least one schedulable unit can be from the same computer room 301 or from a different computer room 301. Then, the resource scheduling device 302 obtains the current data corresponding to the cost impact factor of at least one schedulable unit. In this embodiment, the cost impact factor used includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit depends. Based on this, the current data includes: at least one schedulable unit and its dependent Current power consumption data for your infrastructure. Then, the resource scheduling device 302 may predict at least one schedulable unit and the current power consumption data of the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price. The cost of the dispatch unit. Finally, the resource scheduling device 302 may perform resource scheduling on at least one schedulable unit according to the cost of the at least one schedulable unit.

Regarding the detailed process of resource scheduling of the schedulable unit by the resource scheduling device 302 in combination with the cost of the schedulable unit, refer to the foregoing embodiment, which will not be repeated here.

In the above-mentioned embodiment of the present application, the cost prediction process and the resource scheduling process of the schedulable unit are deployed on the resource scheduling device, but it is not limited to this. For example, a distributed deployment method may also be adopted, and the cost prediction process and the resource scheduling process of the schedulable unit may be separately deployed on different devices for implementation. In the following embodiments, a computer room system and a data center system will be used as examples to illustrate the implementation of distributed deployment.

Fig. 4 is a schematic structural diagram of another computer room system provided by an exemplary embodiment of this application. As shown in FIG. 4, the computer room system 400 includes: a computer room. The computer room refers to a physical place where machinery and equipment are stored, for example, a room or a factory building. The computer room system 400 can be constructed separately or placed in other buildings. Further, as shown in FIG. 4, the computer room system 400 further includes: at least one physical device 401, a resource scheduling device 402, and an information prediction device 403 located in the computer room. This embodiment does not limit the number of physical devices 401, and it may be one or multiple.

In this embodiment, the device form of the physical device 401 is not limited. As shown in FIG. 4, the physical device 401 may be an IT-type device in a computer room and a refrigeration device for cooling the IT-type device, such as an air-conditioning device. For example, the at least one physical device 401 may include, but is not limited to: cabinet equipment, server equipment, computer equipment, printers, hubs, power supply equipment, storage equipment, network switching equipment, air conditioning equipment, and the like. The server device may include, but is not limited to, a conventional server, a server array, or a cloud server. The power supply equipment can be storage battery equipment, dry battery equipment, or uninterruptible power supply (UPS). Storage devices may include, but are not limited to: disks, disk arrays, hard disks, network storage devices (NAS), and so on.

Similarly, this embodiment does not limit the device form of the resource scheduling device 402 and the information prediction device 403 either. The resource scheduling device 402 can be any computer device with certain resource scheduling capabilities, for example, it can be a terminal device such as a smart phone, a notebook computer, a desktop computer, a tablet computer, or a smart watch, or a server such as a conventional server, a cloud server, or a server array. equipment. The information prediction device 403 can be any computer device with cost prediction capability, for example, it can be a terminal device such as a smart phone, a notebook computer, a desktop computer, a tablet computer, or a smart watch, or a server such as a conventional server, a cloud server, or a server array. equipment.

In this embodiment, from the perspective of resource scheduling, the physical device 401 in the computer room is divided into several schedulable units and several infrastructures. In the embodiments of the present application, "several" means an indeterminate number, which may mean one or more. Among them, a schedulable unit refers to a resource device, module, or unit that can provide a certain service, function, or resource (for example, computing, storage, or network) to the outside in the computer room. For example, the schedulable unit can be a physical device such as a server, a printer, or a switch in the computer room, or a hardware module such as a CPU, GPU, memory, network card, or hard disk on the server, or a virtual device running on a physical device. Software modules or units such as computers and cloud services. Infrastructure refers to the facilities in the computer room that provide basic services for dispatchable units. According to the different schedulable units, the infrastructure will be different, which is not limited. Taking the server as the schedulable unit as an example, the infrastructure may include cabinets that provide support services for the servers, air-conditioning equipment or water-cooling equipment for cooling the servers, power equipment for supplying power to the servers, and so on.

Among them, the schedulable unit is used as an available resource in the computer room system 400, and the schedulable unit can be reasonably scheduled for resources through the resource scheduling device 402. Reasonable resource scheduling is of great significance to improving the resource utilization efficiency of the computer room system 400, saving energy, improving resource sharing, and reducing operating costs. In this embodiment, resource scheduling is a process of allocating schedulable units to users who need resources. For scenarios where resources are overloaded (demand is greater than system capacity), or scenarios where demand and capacity will dynamically change over time, the resource scheduling device 402 can dynamically reallocate resources so as to use available resources more efficiently. For example, the resource scheduling device 402 may allocate the application programs of users with certain resource requirements to certain schedulable units or units to run. For another example, the resource scheduling device 402 may allocate resources such as CPU, storage, or network bandwidth provided by one or some schedulable units to certain resource demand users, and the resource demand users use these resources according to actual conditions. The "application program" in the embodiments of the present application refers to various programs, codes, or software in a broad sense.

In this embodiment, the resource scheduling device 402 simultaneously considers the cost of the schedulable unit during resource scheduling of the schedulable unit. Resource scheduling based on cost can not only prioritize resources with lower cost, but also reduce the probability of local hot spots, which is beneficial to reducing the overall power consumption of the computer room system 400. Among them, in order to use the cost of the schedulable unit, the computer room system 400 of this embodiment further includes: an information prediction device 403 with cost-cost prediction capabilities, which is mainly used to predict the cost of the schedulable unit and provide it to the resource scheduling device 402, For the resource scheduling device 402 to perform resource scheduling on the schedulable resource according to the cost of the schedulable unit. Among them, the physical meaning represented by the "cost of the schedulable unit" can be flexibly defined according to application requirements, which is not limited in this embodiment. For example, the cost of the schedulable unit may be the cost of power consumption, or the cost of maintenance, or the cost of configuration, and so on. In the embodiments of the present application, the cost in terms of power consumption is taken as an example for description.

Among them, the direct factor that has an impact on the cost is the power consumption of the schedulable unit itself, in addition to many other factors. In this embodiment, the main consideration is the power consumption of the infrastructure on which the schedulable unit depends (referred to as infrastructure power consumption), and the power consumption of the infrastructure on which the schedulable unit depends is converted to the cost of the schedulable unit. In the cost, the rationality of resource scheduling can be improved. In this embodiment, the factors that have an impact on the cost of the schedulable unit are collectively referred to as the cost impact factor. In this embodiment, the cost impact factor includes at least two. For example, the power consumption of the schedulable unit itself is a cost impact factor, and the power consumption of the infrastructure on which the schedulable unit depends is another cost impact factor. . Among these cost impact factors, the value of some impact factors changes dynamically, and the value of some impact factors may not change. Regardless of whether the value of the impact factor changes, in the embodiments of the present application, the value of the cost impact factor at the current moment or the current time period is referred to as the current data corresponding to the cost impact factor; the cost impact factor is in the history The value in the time or historical period is called the historical data corresponding to the cost impact factor. According to the different cost impact factors, the data content contained in current data and historical data will also be different.

For the resource scheduling device 402, at least one schedulable unit can be determined from the schedulable units included in the computer room system 400 during each resource scheduling process. Among them, the at least one schedulable unit may be part of the schedulable unit included in the computer room system 400, or may be all schedulable units included in the computer room system 400. In other words, at least one schedulable unit comes from several schedulable units included in the computer room system 400. Optionally, the resource scheduling device 402 may notify the information prediction device 403 for the information prediction device 403 to predict the cost of at least one schedulable unit.

For the information prediction device 403, the current data corresponding to the cost impact factor of at least one schedulable unit can be obtained. In this embodiment, the cost impact factor used includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit depends. Based on this, the current data corresponding to these cost impact factors include: at least one The current power consumption data of the schedulable unit and the infrastructure on which it depends. Then, the information prediction device 403 can predict at least one schedulable unit and the current power consumption data of the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price. The cost of the scheduling unit; the cost of at least one schedulable unit is provided to the resource scheduling device 402. The resource scheduling device 402 receives the cost price of the at least one schedulable unit provided by the information prediction device 403, and performs resource scheduling on the at least one schedulable unit according to the cost price of the at least one schedulable unit.

In this embodiment, the manner in which the resource scheduling device 402 determines at least one schedulable unit from the schedulable units included in the computer room system 400 is not limited. For example, taking this resource scheduling as an example, the resource scheduling device 402 may determine at least one schedulable unit that can be scheduled this time from the schedulable units included in the computer room system 400 according to the resource demand corresponding to this resource scheduling. The resource requirements corresponding to this resource scheduling may include some requirements for the schedulable unit of this resource scheduling, such as type, location, or capacity, etc. The schedulable unit that meets these requirements can participate in this resource scheduling.

Further, in addition to considering the resource requirements corresponding to this resource scheduling, at least one schedulable unit that can be scheduled this time can be determined from the schedulable units included in the computer room system 400 in combination with the device topology relationship of the computer room system 400. The device topology relationship of the computer room system 400 describes the topology relationship between the physical devices 401 in the computer room system 400, and the distance between each schedulable unit and the resource scheduling device 402 can be clearly known through the topology relationship. Based on this, according to the device topology relationship of the computer room system 400, at least one schedulable unit whose distance from the resource scheduling device performing this resource scheduling meets the set requirements can be determined from among the schedulable units included in the computer room system 400. The distance here can be a physical distance or a network transmission distance. This embodiment does not limit the "setting requirements", and can be flexibly set according to application requirements. The following example illustrates:

Example 1: The setting requirement refers to being located in the same computer room. According to the device topology relationship of the computer room system 400, from the schedulable units included in the computer room system 400, at least one of the schedulable units located in the same computer room as the resource scheduling device 402 can be determined. Scheduling unit. To put it simply, it is to obtain a schedulable unit that meets the resource demand corresponding to this resource scheduling from each schedulable unit included in the computer room system 400.

Example 2: The setting requirement refers to being located in the same channel, then according to the device topology relationship of the computer room system 400, from the schedulable units included in the computer room system 400, at least one of the schedulable units located in the same channel as the resource scheduling device 402 can be determined. Scheduling unit. The computer room system 400 includes a number of channels, each channel includes a number of cabinets, and each cabinet includes a number of servers. The resource scheduling device 402 is deployed in a certain channel. To put it simply, it is to obtain a schedulable unit that meets the resource requirements corresponding to this resource scheduling in the schedulable unit in the channel where the resource scheduling device 402 is located.

Example 3: The setting requirement refers to being located in the same cabinet, then according to the equipment topology of the computer room system 400, from the schedulable units included in the computer room system 400, at least one of the schedulable units that is located in the same cabinet as the resource scheduling device 402 can be determined. Scheduling unit. The computer room system 400 includes a number of channels, each channel includes a number of cabinets, and each cabinet includes a number of servers. The resource scheduling device 402 is deployed in a certain cabinet. To put it simply, it is deployed in the schedulable unit in the cabinet where the resource scheduling device 402 is located, and obtains the schedulable unit that meets the resource requirements corresponding to this resource scheduling.

Example 4: The setting requirement means that the physical distance is less than the set distance threshold, and the physical distance to the resource scheduling device 402 can be determined from the schedulable units included in the computer room system 400 according to the device topology of the computer room system 400 At least one schedulable unit less than the set distance threshold.

In any manner, after determining at least one schedulable unit, the information prediction device 403 may obtain the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends. Optionally, the information prediction device 403 may obtain the current power consumption data of each physical device 401 in the computer room system 400, and analyze the current power consumption data of each schedulable unit in the computer room system 400 and the current power consumption data of each infrastructure; Then, the current power consumption data of at least one schedulable unit that can be schedulable this time is selected from the current power consumption data of each schedulable unit, and combined with the device topology relationship of the computer room system 400, the current power consumption data of each infrastructure Select the current power consumption data of the infrastructure on which at least one schedulable unit each depends.

In some embodiments of the present application, the cost prediction model corresponding to each schedulable unit can be pre-trained offline, and the cost prediction model corresponding to each schedulable unit can reflect the cost and cost impact factor corresponding to the schedulable unit itself. The relationship between the cost and the cost; based on this, the cost prediction model corresponding to at least one schedulable unit can be used online to predict the cost, and the cost of at least one schedulable unit can be obtained.

Wherein, FIG. 5a shows a block diagram of a working principle of the information prediction device 403. Referring to FIG. 5a, the information prediction device 403 may obtain historical data corresponding to the cost impact factor of at least one schedulable unit, and the historical data includes historical power consumption data of the at least one schedulable unit and the infrastructure on which it depends. The historical power consumption data here includes power consumption data at multiple historical moments or historical time periods; model training is performed according to the historical power consumption data of at least one schedulable unit and the infrastructure on which it depends to obtain at least one schedulable unit Each corresponding cost price prediction model. Further, referring to FIG. 5a, in the resource scheduling process, the information prediction device 403 obtains the current power consumption data of at least one schedulable unit and the infrastructure on which it depends, and compares the current power consumption data of at least one schedulable unit and the infrastructure on which it depends. The current power consumption data of each is input into respective cost cost prediction models to perform cost cost prediction to obtain the cost cost of at least one schedulable unit; and then the cost cost of at least one schedulable unit is provided to the resource scheduling device 402 for resources The scheduling device 402 performs resource scheduling on at least one schedulable unit in combination with the cost of the at least one schedulable unit.

Among them, the cost cost prediction model shown in FIG. 5a mainly reflects the influence of the cost cost influencing factor of infrastructure power consumption on the cost cost of schedulable units. Optionally, the influence relationship can be expressed as the aforementioned formula (1), but it is not limited to this. For the related description of formula (1), please refer to the foregoing embodiment.

Based on the influence relationship of the cost influence factor of the infrastructure power consumption expressed by the above formula (1) on the cost cost of the schedulable unit, taking the target schedulable unit as an example, the process of predicting the cost of the information prediction device 403 will be described. Wherein, the target schedulable unit is any schedulable unit among at least one schedulable unit.

For the target schedulable unit, the information prediction device 403 inputs the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model corresponding to the target schedulable unit; the cost prediction model corresponding to the target schedulable unit Inside the model, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the trained weight coefficients of the first type, and the result of the weighted summation is combined with the current power consumption data of the target schedulable unit Add them to get the cost of the target schedulable unit. Among them, the first type of weight coefficient reflects the impact of the cost impact factor of infrastructure power consumption on the cost of schedulable units.

It should be noted that the factors that affect the cost of the schedulable unit, that is, the cost impact factor, in addition to the power consumption data of the schedulable unit itself and the power consumption data of the infrastructure on which the schedulable unit depends, there are also Many others, such as the price of electricity, the types of applications running on the dispatchable unit, and the physical location of the dispatchable unit. Among them, different types of applications have different power consumption usage patterns, which are generally required in advance in the SLA.

Based on the foregoing, in some other embodiments of the present application, in addition to considering the two cost impact factors of the power consumption data of the schedulable unit itself and the power consumption data of the infrastructure on which the schedulable unit depends, the schedulable unit can also be considered. At least one of the physical location of the unit, the price of electricity in the area where the schedulable unit is located, and the type of programs running on the schedulable unit, and other cost impact factors. As shown in FIG. 5b, it is a block diagram of another working principle of the information prediction device 403. To simplify the illustration, the working principle of the information prediction device 403 is described by taking the target schedulable unit as an example in FIG. 5b.

Referring to FIG. 5b, for the target schedulable unit, the information prediction device 403 may obtain historical data corresponding to the cost impact factor of the target schedulable unit. In this embodiment, the cost impact factor includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit is based, as well as at least one of physical location, electricity price, and application type, which are affected by these costs. The historical data corresponding to the factor includes: historical power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and the target schedulable unit At least one type of data from the type of historical program running on. The information prediction device 403 runs on the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and the target schedulable unit. Model training is performed on at least one data of the historical program types to obtain a cost prediction model corresponding to the target schedulable unit.

Further, referring to Fig. 5b, in the resource scheduling process, the information prediction device 403 can obtain the current data corresponding to the cost impact factor of the target schedulable unit. In this embodiment, the cost impact factor includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit is based, as well as at least one of physical location, electricity price, and application type, which are affected by these costs. The current data corresponding to the factor includes: the current power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the target schedulable unit At least one type of data in the type of program that needs to be run. After that, the information prediction device 403 combines the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the target schedulable unit At least one of the types of programs that need to be run is input into the cost cost prediction model corresponding to the target schedulable unit to perform cost cost prediction to obtain the cost cost of the target schedulable unit; and then provide the cost cost of at least one schedulable unit to The resource scheduling device 402 is used for the resource scheduling device 402 to perform resource scheduling on at least one schedulable unit in combination with the cost of the at least one schedulable unit.

Among them, the cost prediction model shown in Figure 5b mainly reflects the influence of multiple cost influencing factors such as infrastructure power consumption, electricity price, application type, and physical location of schedulable units on the cost of schedulable units. Optionally, the influence relationship can be expressed as the following formula (2), but it is not limited to this. For the related description of formula (2), please refer to the foregoing embodiment.

Based on the influence relationship of multiple cost cost influencing factors on the cost cost of the schedulable unit expressed by the above formula (2), still taking the target schedulable unit as an example, the process of predicting the cost cost of the information prediction device 403 will be described.

For the target schedulable unit, the information prediction device 403 will calculate the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and Data of at least one of the types of programs that need to be run on the target schedulable unit is input into the cost prediction model corresponding to the target schedulable unit. In the cost prediction model corresponding to the target schedulable unit, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the first type of weight coefficients that have been trained, and the weighted sum is the result Add the current power consumption data of the target schedulable unit to obtain the initial cost of the target schedulable unit; and use the second type of weight coefficient that has been trained to correct the initial cost to obtain the cost of the target schedulable unit .

Fig. 6a is a schematic structural diagram of another data center system provided by an exemplary embodiment of this application. As shown in FIG. 6a, the data center system 600 includes: at least one computer room 601, a resource scheduling device 602, and an information prediction device 604. The data center system 600 can be built separately, or placed in other buildings, or distributed in different geographical locations. Among them, each computer room 601 includes at least one physical device 603. The number of physical devices contained in each computer room 601 may be one or multiple.

The computer room 601 in this embodiment is similar to the computer room in the embodiment shown in FIG. 4. For related descriptions of the physical equipment 603 and the computer room 601, please refer to the description in the embodiment shown in FIG. 4, which will not be repeated here.

In this embodiment, from the perspective of resource scheduling, the physical equipment 603 in each computer room 601 is divided into several schedulable units and several infrastructures. In the embodiments of the present application, "several" means an indeterminate number, which may mean one or more. Among them, a schedulable unit refers to a resource device, module, or unit that can provide a certain service, function, or resource (for example, computing, storage, or network) to the outside in the computer room. For example, the schedulable unit can be a physical device such as a server, a printer, or a switch in the computer room, or a hardware module such as a CPU, GPU, memory, network card, or hard disk on the server, or a virtual device running on a physical device. Software modules or units such as computers and cloud services. Infrastructure refers to the facilities in the computer room that provide basic services for dispatchable units. According to the different schedulable units, the infrastructure will be different, which is not limited. Taking the server as the schedulable unit as an example, the infrastructure may include cabinets that provide support services for the servers, air-conditioning equipment or water-cooling equipment for cooling the servers, power equipment for supplying power to the servers, and so on.

The difference between this embodiment and the embodiment shown in FIG. 4 is that the information prediction device 604 and the resource scheduling device 602 do not belong to a certain computer room, but belong to the entire data center system 600, and the schedulable units in each computer room 601 Perform cost prediction and resource scheduling. Although the information prediction device 604 and the resource scheduling device 602 do not belong to any computer room, in terms of physical deployment, they can be deployed in a certain computer room or independently deployed outside each computer room.

In the same way, the information prediction device 604 and the resource scheduling device 602 of this embodiment can cooperate with each other, and the cost of the schedulable unit is considered in the process of resource scheduling for the schedulable unit. For the resource scheduling device 602, at least one schedulable unit can be determined from the schedulable units included in the data center system 600 during each resource scheduling process. Among them, the at least one schedulable unit may be part of the schedulable unit included in the data center system 600, or may be all schedulable units included in the data center system 600. In addition, it should be noted that at least one schedulable unit may come from the same computer room 601 or different computer rooms 601. Then, the information prediction device 604 obtains the current data corresponding to the cost impact factor of at least one schedulable unit. In this embodiment, the cost impact factor used includes the power consumption of the schedulable unit and the power consumption of the infrastructure on which the schedulable unit depends. Based on this, the current data corresponding to these cost impact factors include: at least one The current power consumption data of the schedulable unit and the infrastructure on which it depends. Then, the information prediction device 604 can predict at least one schedulable unit and the current power consumption data of the infrastructure on which it depends, and the pre-trained cost impact factor and the impact relationship between the cost price. The cost of the scheduling unit; and then the cost of at least one schedulable unit is provided to the resource scheduling device 602. Finally, the resource scheduling device 602 may perform resource scheduling on at least one schedulable unit according to the cost of the at least one schedulable unit.

For the detailed process of the information prediction device 604 predicting the cost of the schedulable unit, and the detailed process of the resource scheduling device 602 combining the cost of the schedulable unit to perform resource scheduling on the schedulable unit, please refer to the foregoing embodiment, which will not be repeated here. .

In summary, in the computer room system or data center system provided by the embodiments of the present application, the resource scheduling equipment and the information prediction equipment cooperate with each other, so that the prediction of the cost and the resource scheduling are combined, and the cost of the schedulable unit is predicted to achieve Resource scheduling based on cost cost can prioritize resources with lower cost cost and reduce the probability of local hot spots, which is conducive to reducing overall power consumption. In addition, in the cost cost prediction process, the schedulable unit depends on at the same time. The contribution of the power consumption of the infrastructure to the cost of the schedulable unit can make the predicted cost more reasonable and accurate, and thus can improve the rationality of resource scheduling based on the cost.

In each of the foregoing computer room system embodiments or data center system embodiments, after the cost of at least one schedulable unit is obtained, resource scheduling can be performed on at least one schedulable unit based on the cost of at least one schedulable unit.

In an optional embodiment, resource scheduling may be performed on at least one schedulable unit based solely on the cost of the at least one schedulable unit. Wherein, based on the cost of at least one schedulable unit alone, resource scheduling methods for at least one schedulable unit include but are not limited to the following:

Manner 1: According to the cost of all at least one schedulable unit, resource scheduling is performed on the at least one schedulable unit in the descending order of the cost. In method 1, resource scheduling can be performed on at least one schedulable unit in the order of ascending cost.

Manner 2: According to the cost of at least one schedulable unit, the schedulable unit with the smallest cost is prioritized for resource scheduling. In method 2, in each resource scheduling, resource scheduling is allowed for the schedulable unit with the least cost.

Manner 3: According to the cost price of all at least one schedulable unit, priority is given to resource scheduling for the schedulable unit whose cost cost is less than the set threshold. In mode 3, a threshold is set, and each resource scheduling allows resource scheduling of schedulable units whose cost is lower than the threshold.

In another optional embodiment, resource scheduling may be performed on at least one schedulable unit according to the service requirements corresponding to this resource scheduling and the cost of at least one schedulable unit. In the embodiments of the present application, the business requirements are not limited. For example, the service requirements corresponding to this resource scheduling include service performance requirements. Based on this, a schedulable unit that meets the service performance requirements can be selected from at least one schedulable unit according to the service performance requirements corresponding to this resource scheduling; According to the cost of the schedulable unit that meets the service performance requirement, resource scheduling is performed on the schedulable unit that meets the service performance requirement. It should be noted that the method of combining service performance requirements and cost for resource scheduling is not limited to the one listed here. Among them, the schedulable unit that meets the service performance requirement may include: a schedulable unit (such as a server) with a performance level greater than a set level, or may include a newer batch of schedulable units (such as a server), and so on. Among them, the newer batch can be understood as the relatively short time to put into use.

In addition to the above-mentioned computer room system and data center system, the resource scheduling solution that combines power consumption cost cost prediction and resource scheduling provided by the embodiments of the present application can also be applied to cluster systems or edge computing systems that have other resource scheduling requirements. In the system.

Fig. 6b is a schematic structural diagram of an edge computing system provided by an exemplary embodiment of this application. As shown in Figure 6b, the edge computing system includes several edge computing nodes, several infrastructures and servers. In this embodiment, "several" means an indeterminate number, and it may mean one or more.

Among them, an edge computing node refers to a node device at the edge of the network logic, and usually includes hardware modules, drivers for hardware modules, operating systems, and related application programs. Hardware modules include but are not limited to: CPU, network card, memory, etc. Several infrastructures provide basic services for several edge computing nodes. The equipment forms of edge computing nodes are different, and the infrastructure that provides basic services for edge computing nodes will also be different. For example, the edge computing node may be a personal computer in the home, and the infrastructure that can provide basic services for the personal computer includes, but is not limited to: household power supplies, routers, and so on. For another example, if the edge computing node is a street lamp, the infrastructure that can provide basic services for the street lamp includes, but is not limited to: solar panels, solar batteries, or urban power supply systems. For another example, if the edge computing node is a traffic signal light, etc., the infrastructure that can provide basic services for the traffic signal light includes, but is not limited to: power supply systems (such as solar cells, storage batteries, or urban power supply systems), controllers, and so on. For another example, if the edge computing node is an electronic eye or camera, the infrastructure that can provide basic services for the electronic eye or camera includes, but is not limited to: power supply system (such as solar cells, batteries or urban power supply systems), signal transmission lines, installation poles, control System and so on.

Among them, there are one or more servers, which can be deployed in the cloud or in the client room. The server communicates with the edge computing node through the network, and the server can respond to the request of the edge computing node and provide related cloud services for the edge computing node. In addition, the server can also control, operate and maintain edge computing nodes. In addition, the edge computing node is a schedulable unit, and the server can also perform resource scheduling on it. The server performs resource scheduling on the edge computing node, and can perform operations such as some applications, data and services from the network center node (such as the server) to the edge computing node on the network logic for processing, thereby realizing a distributed operation Architecture.

In this embodiment, the server performs resource scheduling on the edge computing node in combination with the cost of the edge computing node. Cost-based resource scheduling can prioritize low-cost edge computing nodes and reduce the probability of local hot spots, which is beneficial to improve the overall resource utilization of the edge computing system and reduce the overall power consumption of the edge computing system. Among them, the physical meaning represented by the "cost of edge computing node" can be flexibly defined according to application requirements, which is not limited in this embodiment. For example, the cost of edge computing nodes may be the cost of power consumption, or the cost of maintenance, or the cost of configuration, and so on. In this embodiment, the description is focused on the cost of power consumption as an example.

Among them, the direct factor that has an impact on the cost is the power consumption of the edge computing node itself. In addition, it also includes many other factors. In this embodiment, the power consumption of the infrastructure on which the edge computing node depends is mainly considered, and the power consumption of the infrastructure on which the edge computing node depends is converted to the cost of the edge computing node, which can improve resource scheduling. rationality. In this embodiment, the factors that have an impact on the cost of the edge computing node are collectively referred to as the cost impact factor. In this embodiment, the cost impact factor includes at least two. For example, the power consumption of the edge computing node itself is a cost impact factor, and the power consumption of the infrastructure on which the edge computing node depends is another cost impact factor. . Among these cost impact factors, the value of some impact factors changes dynamically, and the value of some impact factors may not change. Regardless of whether the value of the impact factor changes, in the embodiments of the present application, the value of the cost impact factor at the current moment or the current time period is referred to as the current data corresponding to the cost impact factor; the cost impact factor is in the history The value in the time or historical period is called the historical data corresponding to the cost impact factor. According to the different cost impact factors, the data content contained in current data and historical data will also be different.

For the server, in each resource scheduling process, at least one schedulable edge computing node can be determined from a number of edge computing nodes included in the edge computing system. The at least one edge computing node may be part of the edge computing node included in the edge computing system, or may be all edge computing nodes included in the edge computing system. Then, the server obtains the current data corresponding to the cost impact factor of the at least one edge computing node, and the current data includes the current power consumption data of the at least one edge computing node and the infrastructure on which it depends. Then, the server predicts the cost of at least one edge computing node based on the current power consumption data of at least one edge computing node and the infrastructure on which it depends, and the pre-trained cost impact factor and the impact relationship between the cost price. . Finally, according to the cost of at least one edge computing node, resource scheduling is performed on at least one edge computing node.

In this embodiment, the manner in which the server determines at least one schedulable edge computing node from the edge computing nodes included in the edge computing system is not limited. For example, taking this resource scheduling as an example, the server may determine at least one schedulable edge computing node from each edge computing node included in the edge computing system according to resource requirements corresponding to this resource scheduling. The resource requirements corresponding to this resource scheduling include some requirements for edge computing nodes of this resource scheduling, such as type, location, or capacity, and edge computing nodes that meet these requirements can participate in this resource scheduling. For example, the server can select an edge computing node of the type required for this resource scheduling according to the resource requirements corresponding to this resource scheduling, or select an edge computing node located in the geographical area required by this resource scheduling, or choose to meet this time. The edge computing node with computing capability required for resource scheduling is used as the schedulable edge computing node this time. Of course, the resource requirements corresponding to this resource scheduling can also include two or more conditions, and the server can also combine two or more conditions to determine the edge computing nodes that can be scheduled this time.

Further, in addition to considering the resource requirements corresponding to this resource scheduling, the device topology of the edge computing system can also be combined to determine at least one edge computing node that can be scheduled this time from the edge computing nodes included in the edge computing system. For example, according to the device topology of the edge computing system, from among the edge computing nodes included in the edge computing system, at least one edge computing node whose physical distance to the server or network transmission distance is less than a set distance threshold can be determined.

The main difference between this embodiment and the previous embodiment is that the schedulable unit is specifically an edge computing node. For the detailed process of the server predicting the cost of the edge computing node, and the detailed process of the server performing resource scheduling on the edge computing node in combination with the cost of the edge computing node, refer to the foregoing embodiment, and will not be repeated here.

In the edge computing system of this embodiment, the server combines cost cost prediction with resource scheduling, and by predicting the cost cost of edge computing nodes, it realizes cost cost-based resource scheduling, which can prioritize resources with lower costs. It can also reduce the probability of local hot spots and help reduce overall power consumption. In addition, in the cost prediction process, the power consumption of the infrastructure on which the edge computing node depends is considered at the same time as the contribution to the cost of the edge computing node. The predicted cost is more reasonable and accurate, and the rationality of resource scheduling based on cost can be improved.

In addition to the foregoing computer room system, data center system, and edge computing system embodiments, this application also provides some method embodiments, which are described below.

Fig. 7a is a schematic flowchart of a resource scheduling method provided by an exemplary embodiment of this application. This method is described from the perspective of resource scheduling equipment. As shown in Figure 7a, the method includes:

71a. Acquire current data corresponding to the cost impact factor of at least one schedulable unit, where the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends.

72a. Predict the cost of at least one schedulable unit based on the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, and the pre-trained impact factor of the cost cost and the impact relationship between the cost cost. .

73a. Perform resource scheduling on at least one schedulable unit according to the cost of the at least one schedulable unit.

In an optional embodiment, the cost prediction model corresponding to each schedulable unit can be pre-trained offline, and the cost cost prediction model corresponding to each schedulable unit can reflect the cost impact factor and cost corresponding to the schedulable unit itself. Based on this, the cost prediction model corresponding to at least one schedulable unit is used to predict the cost on the line to obtain the cost of at least one schedulable unit. Based on this, an implementation manner of step 72a includes: inputting the current power consumption data of at least one schedulable unit and the infrastructure on which it depends, respectively, into their respective cost prediction models for cost cost prediction, so as to obtain at least one schedulable unit. The cost of the scheduling unit; wherein, the cost prediction model corresponding to each schedulable unit reflects the influence relationship existing between the corresponding cost cost influencing factor and the cost cost.

Further optionally, before using the cost prediction model, a model training process is also included. Among them, as shown in Figure 7b, the process of a model training method includes the following steps:

71b. Acquire historical data corresponding to the cost impact factor of at least one schedulable unit, where the historical data includes historical power consumption data of the at least one schedulable unit and the infrastructure on which it depends.

72b. Perform model training according to the historical power consumption data of the at least one schedulable unit and the infrastructure on which it depends, respectively, to obtain a cost prediction model corresponding to each of the at least one schedulable unit.

In the embodiments of the present application, the model training process is not limited. For example, a supervised model training method can be used, or an unsupervised model training method can also be used. Similarly, the embodiments of the present application do not limit the algorithm used for model training, for example, linear regression, logistic regression, or various deep learning algorithms can be used.

In an optional embodiment, the cost cost prediction model mainly reflects the influence relationship of the cost cost influencing factor of the power consumption of the infrastructure on the cost cost of the schedulable unit. Optionally, the influence relationship can be expressed as the aforementioned formula (1), but it is not limited to this. For the relevant description of the formula (1), please refer to the foregoing embodiment, which will not be repeated here.

Based on the above formula (1), the influence relationship of the cost impact factor of the power consumption of the infrastructure on which the schedulable unit depends on the cost of the schedulable unit, take any schedulable unit in at least one schedulable unit as an example , The implementation process of step 72a is described. For ease of description, any schedulable unit is referred to as a target schedulable unit. Among them, an implementation process of step 72a includes: for the target schedulable unit, input the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model corresponding to the target schedulable unit; Within the cost prediction model corresponding to the scheduling unit, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is performed using the first-type weight coefficients that have been trained, and the result of the weighted sum is combined with the target achievable value. The current power consumption data of the scheduling unit is added to obtain the cost of the target schedulable unit.

It should be noted that the factors that affect the cost of the schedulable unit, that is, the cost impact factor, in addition to the power consumption of the schedulable unit itself and the power consumption of the infrastructure on which the schedulable unit depends, there are many other factors. , Such as the price of electricity, the types of applications running on the schedulable unit, and the physical location of the schedulable unit. Among them, different types of applications have different power consumption usage patterns, which are generally required in advance in the SLA.

Based on the foregoing, in other embodiments of the present application, in addition to considering the two cost impact factors of the power consumption of the schedulable unit itself and the power consumption of the infrastructure, the physical location of the schedulable unit and the location of the schedulable unit can also be considered. At least one of the regional electricity price and the type of program running on the schedulable unit, and other cost impact factors. Based on this, taking the target schedulable unit as an example, in step 71b, in addition to obtaining the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, the historical physical location, At least one of the historical electricity price of the area where the target schedulable unit is located and the historical program type running on the target schedulable unit; accordingly, in step 72b, the target schedulable unit and the infrastructure on which it depends Historical power consumption data, and at least one of the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and the historical program type running on the target schedulable unit for model training to obtain the target schedulable unit Corresponding cost prediction model.

Further, in step 71a, in addition to obtaining the current power consumption data of the target schedulable unit and the infrastructure on which it depends, the current physical location of the target schedulable unit and the current electricity bill of the area where the target schedulable unit is located can also be obtained. At least one of the price and the type of program that needs to be run on the target schedulable unit; correspondingly, in step 72a, the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the target schedulable unit At least one of the current physical location of the unit, the current price of electricity in the area where the target schedulable unit is located, and the type of program that needs to be run on the target schedulable unit are input into the cost and price prediction model corresponding to the target schedulable unit to predict the cost and price. Get the cost of the target schedulable unit.

The above cost prediction model mainly reflects the power consumption of the infrastructure on which the schedulable unit depends, the price of electricity, the types of applications running on the schedulable unit, and the physical location of the schedulable unit. The impact of cost. Optionally, the influence relationship can be expressed as the aforementioned formula (2), but it is not limited to this. For the related description of formula (2), please refer to the foregoing embodiment, which will not be repeated here.

Based on the influence relationship of multiple cost cost influencing factors on the cost cost of the schedulable unit expressed by the above formula (2), the target schedulable unit is still taken as an example to illustrate the implementation process of step 72a. Among them, an implementation process of step 72a includes: for the target schedulable unit, the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, and the target schedulable unit At least one of the current price of electricity in the area and the type of program that needs to be run on the target schedulable unit is input into the cost prediction model corresponding to the target schedulable unit. In the cost prediction model corresponding to the target schedulable unit, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the first type of weight coefficients that have been trained, and the weighted sum is the result Add the current power consumption data of the target schedulable unit to obtain the initial cost of the target schedulable unit; and use the second type of weight coefficient that has been trained to correct the initial cost to obtain the cost of the target schedulable unit .

In an optional embodiment, before step 71a, it further includes: determining the schedulable unit from the schedulable units included in the computer room system, data center system, or edge computing system according to the resource demand corresponding to this resource scheduling. At least one schedulable unit.

Further optionally, determining at least one schedulable unit that can be schedulable this time from the schedulable units included in the computer room system, data center system, or edge computing system according to the resource requirements corresponding to this resource scheduling includes: according to the computer room system , The device topology relationship of the data center system or edge computing system, from the schedulable units included in the computer room system, data center system or edge computing system, determine that the distance to the resource scheduling device performing this resource scheduling satisfies the setting At least one schedulable unit required.

Wherein, an implementation manner for determining at least one schedulable unit whose distance from the resource scheduling device performing this resource scheduling meets the set requirement includes any one of the following:

According to the device topology relationship of the computer room system or the data center system, determine at least one schedulable unit located in the same computer room as the resource scheduling device from among the schedulable units included in the computer room system or the data center system;

According to the device topology relationship of the computer room system or the data center system, determine at least one schedulable unit that is located in the same channel as the resource scheduling device from each schedulable unit included in the computer room system or the data center system;

Determine at least one schedulable unit located in the same cabinet as the resource scheduling device from each schedulable unit included in the computer room system or the data center system according to the device topology relationship of the computer room system or the data center system;

According to the device topology relationship of the computer room system or the data center system, from each schedulable unit included in the computer room system or the data center system, it is determined that the physical distance to the resource scheduling device is less than at least a set distance threshold. A schedulable unit.

In an optional embodiment, for the computer room system or the data center system, the schedulable unit included in the computer room system or the data center system is a server in the computer room system or the data center system, but it is not limited to this. Correspondingly, for the edge computing system, the schedulable unit included in the edge computing system is the edge computing node in the edge computing system, but it is not limited to this.

In an optional embodiment, in step 73a, resource scheduling may be performed on at least one schedulable unit based on the power consumption cost of the at least one schedulable unit alone. Among them, the implementation manners for resource scheduling on at least one schedulable unit include but are not limited to the following:

According to the cost of at least one schedulable unit, perform resource scheduling on at least one schedulable unit in the descending order of cost; or

According to the cost of at least one schedulable unit, the schedulable unit with the lowest cost is prioritized for resource scheduling; or

Based on the cost of at least one schedulable unit, priority is given to resource scheduling for schedulable units whose cost cost is less than the set threshold.

In another optional embodiment, in step 73a, resource scheduling may be performed on at least one schedulable unit according to the service requirements corresponding to this resource scheduling and the cost of at least one schedulable unit. For example, the service requirements corresponding to this resource scheduling include service performance requirements. Based on this, a schedulable unit that meets the service performance requirements can be selected from at least one schedulable unit according to the service performance requirements corresponding to this resource scheduling; To meet the cost of the schedulable unit that meets the service performance requirement, resource scheduling is performed on the schedulable unit that meets the service performance requirement. It should be noted that the method of combining service performance requirements and cost for resource scheduling is not limited to the one listed here. Among them, the schedulable unit that meets the service performance requirement may include: a schedulable unit (such as a server) with a performance level greater than a set level, or may include a newer batch of schedulable units (such as a server), and so on. Among them, the newer batch can be understood as the relatively short time to put into use.

For a detailed description of each step in this embodiment, please refer to the foregoing system embodiment, which will not be repeated here.

In summary, in this embodiment, the cost cost prediction is combined with resource scheduling. By predicting the cost cost of schedulable units, cost cost-based resource scheduling can be realized, which can prioritize resources with lower cost costs and also It can reduce the probability of local hot spots and help reduce overall power consumption; in addition, in the cost prediction process, the power consumption of the infrastructure on which the schedulable unit depends on the schedulable unit is considered to contribute to the cost of the schedulable unit. The predicted cost is more reasonable and accurate, and the rationality of resource scheduling based on cost can be improved.

FIG. 7c is a schematic flowchart of an information prediction method provided by an exemplary embodiment of this application. This method is described from the perspective of the information prediction device or the server in Figure 6b. As shown in Figure 7c, the method includes:

71c. Acquire current data corresponding to the cost impact factor of the target schedulable unit, where the current data includes current power consumption data of the target schedulable unit and the infrastructure on which it depends.

72c. Predict the cost of the target schedulable unit based on the current power consumption data of the target schedulable unit and the infrastructure on which it depends, and the pre-trained impact factor of the cost cost and the impact relationship between the cost cost.

In an optional embodiment, the cost prediction model corresponding to the target schedulable unit can be pre-trained offline, and the cost cost prediction model corresponding to the target schedulable unit can reflect the cost impact factor and cost cost corresponding to the target schedulable unit. The influence relationship exists between. Based on this, an implementation manner of step 72c includes: inputting the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model to perform cost cost prediction, so as to obtain the cost cost of the target schedulable unit.

Further optionally, before using the cost prediction model, a model training process is also included. The process of a model training method includes: obtaining historical data of the target schedulable unit corresponding to the cost impact factor, the historical data including historical power consumption data of the target schedulable unit and the infrastructure on which it depends; schedulable according to the target Model training is performed on the historical power consumption data of the unit and the infrastructure on which it depends, and the cost prediction model is obtained.

In an optional embodiment, the cost cost prediction model mainly reflects the influence relationship of the cost cost influence factor of the power consumption of the infrastructure on which the target schedulable unit depends on the cost cost of the target schedulable unit. Optionally, the influence relationship can be expressed as the aforementioned formula (1), but it is not limited to this. For the relevant description of the formula (1), please refer to the foregoing embodiment, which will not be repeated here.

Based on the influence relationship expressed by the above formula (1), an implementation of step 72c includes: inputting the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model; inside the cost prediction model , Use the first-type weight coefficients that have been trained to perform a weighted summation of the current power consumption data of the infrastructure on which the target schedulable unit depends, and add the result of the weighted summation to the current power consumption data of the target schedulable unit , In order to get the cost of the target schedulable unit. In this embodiment, the first type of weight coefficient reflects the impact of the cost impact factor of the power consumption of the infrastructure on which the target schedulable unit depends on the cost of the target schedulable unit.

It should be noted that the factors that affect the cost of the target schedulable unit, that is, the cost impact factor is in addition to the power consumption data of the target schedulable unit itself and the power consumption data of the infrastructure on which the target schedulable unit depends. , There are many others, such as the price of electricity, the type of application running on the target schedulable unit, and the physical location of the target schedulable unit. Among them, different types of applications have different power consumption usage patterns, which are generally required in advance in the SLA.

Based on the foregoing, in other embodiments of the present application, in addition to considering the two cost impact factors of the target schedulable unit's own power consumption and the power consumption of the infrastructure on which the target schedulable unit depends, the target schedulable unit can also be considered. At least one of the physical location of the dispatching unit, the price of electricity in the area where the target dispatchable unit is located, and the type of program running on the target dispatchable unit, and other cost influencing factors. Based on this, in the model training process, in addition to obtaining the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, the historical physical location of the target schedulable unit and the area where the target schedulable unit is located can also be obtained. At least one of the historical electricity price and the historical program type running on the target schedulable unit; accordingly, it can be based on the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, and the history of the target schedulable unit Model training is performed on at least one of the physical location, the historical electricity price price of the area where the target schedulable unit is located, and the historical program type running on the target schedulable unit to obtain a cost prediction model corresponding to the target schedulable unit.

Further, in step 71c, in addition to obtaining the current power consumption data of the target schedulable unit and the infrastructure on which it depends, the current physical location of the target schedulable unit and the current electricity bill of the area where the target schedulable unit is located can also be obtained. At least one of the price and the type of program that needs to be run on the target schedulable unit; accordingly, in step 72c, the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the target schedulable unit At least one of the current physical location of the unit, the current price of electricity in the area where the target schedulable unit is located, and the type of program that needs to be run on the target schedulable unit are input into the cost and price prediction model corresponding to the target schedulable unit to predict the cost and price. Get the cost of the target schedulable unit.

The above cost prediction model mainly reflects the power consumption of the infrastructure on which the target schedulable unit depends, the price of electricity, the type of application running on the target schedulable unit, and the physical location of the target schedulable unit. The impact of the cost of schedulable units. Optionally, the influence relationship can be expressed as the aforementioned formula (2), but it is not limited to this. For the related description of formula (2), please refer to the foregoing embodiment, which will not be repeated here.

Based on the influence of multiple cost-cost influencing factors on the cost-cost of the target schedulable unit expressed by the above formula (2), an implementation process of step 72c includes: the current status of the target schedulable unit and the infrastructure on which it depends Power consumption data, together with at least one of the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the type of program that needs to be run on the target schedulable unit, enter the cost of the target schedulable unit Forecast model. In the cost prediction model corresponding to the target schedulable unit, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the first type of weight coefficients that have been trained, and the weighted sum is the result Add the current power consumption data of the target schedulable unit to obtain the initial cost of the target schedulable unit; and use the second type of weight coefficient that has been trained to correct the initial cost to obtain the cost of the target schedulable unit . Among them, the weight coefficient of the first type reflects the impact of the cost impact factor of the power consumption of the infrastructure on which the target schedulable unit depends on the cost of the target schedulable unit; the weight coefficient of the second type reflects the physical location of the schedulable unit, The impact of at least one cost factor in the price of electricity and the program type on the cost of the target schedulable unit.

In an optional embodiment, an implementation manner of step 71c includes: determining the current power consumption data of the target schedulable unit from the current power consumption data of each schedulable unit in the computer room system or the data center system; According to the device topology relationship of the computer room system or data center system, from the current power consumption data of each infrastructure of the computer room system or data center system, the current power of the infrastructure on which the target schedulable unit depends is filtered out. Consumption data.

In the embodiments of the present application, the implementation form of the target schedulable unit is not limited, and it may be any device, module, or unit that can provide a certain service, function, or capability to the outside. For example, the target schedulable unit is a server in a computer room system or a data center system, and a CPU, GPU, memory, or network card on the server.

It should be noted that the execution subject of each step of the method provided in the foregoing embodiment may be the same device, or different devices may also be the execution subject of the method. For example, the execution subject of steps 71a to 73a may be device A; for another example, the execution subject of

steps

71a and 72a may be an information prediction device, and the execution subject of step 73a may be a resource scheduling device; and so on.

In addition, in some of the processes described in the above embodiments and drawings, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may be performed out of the order in which they appear in this document or performed in parallel. The sequence numbers of operations, such as 71a, 72a, etc., are only used to distinguish different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions of "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit the "first" and "second" Are different types.

FIG. 8 is a schematic structural diagram of a resource scheduling device provided by an exemplary embodiment of this application. As shown in FIG. 8, the resource scheduling device includes: a memory 81 and a processor 82.

The memory 81 is used to store computer programs, and can be configured to store various other data to support operations on the resource scheduling device. Examples of these data include instructions, messages, pictures, videos, etc. for any application or method operating on the resource scheduling device.

The processor 82 is coupled with the memory 81, and is configured to execute a computer program in the memory 81 to obtain current data corresponding to the cost impact factor of at least one schedulable unit, and the current data includes the data of at least one schedulable unit and The current power consumption data of the infrastructure on which it depends; the current power consumption data of at least one schedulable unit and the infrastructure on which it depends, and the pre-trained cost impact factor and the impact relationship between the cost price , Predict the cost of at least one schedulable unit; according to the cost of the at least one schedulable unit, perform resource scheduling on the at least one schedulable unit. Among them, the physical meaning represented by the "cost of the schedulable unit" can be flexibly defined according to application requirements, which is not limited in this embodiment. For example, the cost of the schedulable unit may be the cost of power consumption, or the cost of maintenance, or the cost of configuration, and so on. In this embodiment, the description is focused on the cost of power consumption as an example.

In an optional embodiment, when predicting the cost of at least one schedulable unit, the processor 82 is specifically configured to: input the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, respectively, into the respective corresponding The cost cost prediction model of the schedulable unit performs cost cost prediction to obtain the cost cost of at least one schedulable unit; among them, the cost cost prediction model corresponding to each schedulable unit reflects the impact between the corresponding cost cost influencing factor and the cost cost relationship.

In an optional embodiment, the processor 82 is further configured to obtain historical data corresponding to the cost impact factor of at least one schedulable unit, the historical data including the history of the at least one schedulable unit and the infrastructure on which it depends Power consumption data; model training is performed according to the historical power consumption data of at least one schedulable unit and the infrastructure on which it depends, respectively, to obtain a cost prediction model corresponding to each of the at least one schedulable unit.

Optionally, the processor 82 may specifically use linear regression, logistic regression or various deep learning algorithms to perform model training during the model training process.

Further, based on the aforementioned cost cost prediction model, when the processor 82 performs cost cost prediction, it is specifically configured to: for the target schedulable unit, input the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the target schedulable unit. The cost and cost prediction model corresponding to the scheduling unit; inside the cost and cost prediction model corresponding to the target schedulable unit, the current power consumption data of the infrastructure on which the target schedulable unit depends is weighted by using the first type of weight coefficients that have been trained Sum, and add the result of the weighted summation to the current power consumption data of the target schedulable unit to obtain the cost of the target schedulable unit;

Among them, the weight coefficient of the first type reflects the impact of the cost impact factor of infrastructure power consumption on the cost of the schedulable unit; the target schedulable unit is any schedulable unit among at least one schedulable unit.

In an optional embodiment, the historical data acquired by the processor 82 further includes at least one of the following: the historical physical location of the at least one schedulable unit, the historical electricity price of the area where the at least one schedulable unit is located, and the historical data on the at least one schedulable unit. Types of historical programs run. Based on this, when the processor 82 performs model training, it is specifically used to: for the target schedulable unit, according to the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, and the historical physical location of the target schedulable unit , Model training is performed on at least one of the historical electricity price price of the area where the target schedulable unit is located and the historical program type running on the target schedulable unit to obtain the cost prediction model corresponding to the target schedulable unit; where the target schedulable unit is Any schedulable unit in at least one schedulable unit.

Correspondingly, the current data acquired by the processor 82 also includes at least one of the following: the current physical location of the at least one schedulable unit, the current electricity price of the area where the at least one schedulable unit is located, and the type of program that needs to be run on the at least one schedulable unit . Based on this, when the processor 82 predicts the cost, it is specifically used to: for the target schedulable unit, combine the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical value of the target schedulable unit. At least one of the location, the current electricity price of the area where the target schedulable unit is located, and the type of program that needs to be run on the target schedulable unit is input into the cost prediction model corresponding to the target schedulable unit to predict the cost to obtain the target schedulable unit The cost of the unit.

Further optionally, the processor 82 is specifically configured to: compare the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, and the target schedulable unit. At least one of the current price of electricity in the area where the dispatch unit is located and the type of program that needs to be run on the target schedulable unit is input into the cost prediction model corresponding to the target schedulable unit;

In the cost prediction model corresponding to the target schedulable unit, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the first type of weight coefficients that have been trained, and the weighted sum is the result Add the current power consumption data of the target schedulable unit to obtain the initial cost of the target schedulable unit; and use the second type of weight coefficient that has been trained to correct the initial cost to obtain the cost of the target schedulable unit ；

Among them, the first type of weight coefficient reflects the impact of the cost of infrastructure power consumption on the cost of schedulable units, and the second type of weight coefficient reflects at least one of the cost of physical location, electricity price, and program type. The impact of the impact factor on the cost of schedulable units.

In an optional embodiment, the processor 82 is further configured to: according to the resource requirements corresponding to this resource scheduling, determine from the schedulable units included in the computer room system, data center system or edge computing system at least A schedulable unit.

Further optionally, when the processor 82 determines at least one schedulable unit that can be schedulable this time, it is specifically configured to: according to the device topology relationship of the computer room system, data center system, or edge computing system, from the computer room system, data center system, or Among the schedulable units included in the edge computing system, determine at least one schedulable unit whose distance from the resource scheduling device performing this resource scheduling meets the set requirement.

Furthermore, the processor 82 is specifically configured to perform any one of the following operations when determining at least one schedulable unit that is schedulable this time:

According to the resource demand corresponding to this resource scheduling and the device topology relationship of the computer room system or data center system, from the schedulable units included in the computer room system or data center system, determine at least one schedulable unit located in the same computer room as the resource scheduling device unit;

According to the resource demand corresponding to this resource scheduling and the device topology relationship of the computer room system or data center system, from each schedulable unit included in the computer room system or data center system, determine at least one schedulable unit located in the same channel as the resource scheduling device unit;

According to the resource demand corresponding to this resource scheduling and the device topology relationship of the computer room system or data center system, from each schedulable unit included in the computer room system or data center system, determine at least one schedulable unit that is located in the same cabinet as the resource scheduling device unit;

According to the resource demand corresponding to this resource scheduling and the device topology relationship of the computer room system or data center system, from the schedulable units included in the computer room system or data center system, determine that the physical distance to the resource scheduling device is less than the set distance At least one schedulable unit of the threshold.

In an optional embodiment, when the processor 82 performs resource scheduling, it is specifically configured to: perform resource scheduling on the at least one schedulable unit according to the cost of the at least one schedulable unit in a descending order of the cost; Or, based on the cost of at least one schedulable unit, priority is given to the schedulable unit with the lowest cost to perform resource scheduling; or, based on the cost of at least one schedulable unit, priority is given to the schedulable unit whose cost is less than the set threshold. Resource Scheduling.

In an optional embodiment, when the processor 82 performs resource scheduling, it is specifically configured to: according to the service requirements corresponding to the resource scheduling this time, and in combination with the cost of the at least one schedulable unit, compare the cost of the at least one schedulable unit. The unit performs resource scheduling.

Furthermore, when the processor 82 is performing resource scheduling, it is specifically configured to: according to the service performance requirements corresponding to this resource scheduling, select a schedulable unit that meets the service performance requirements from the at least one schedulable unit; According to the cost of the schedulable unit required by the service performance, resource scheduling is performed on the schedulable unit meeting the service performance requirement.

Further, as shown in FIG. 8, the resource scheduling device further includes: a communication component 83, a display 84, a power supply component 85, an audio component 85 and other components. Only some of the components are schematically shown in FIG. 8, which does not mean that the resource scheduling device only includes the components shown in FIG. 8. In addition, according to the different implementation forms of the resource scheduling device, the components in the dashed box in FIG. 8 are optional components, not mandatory components. For example, when the resource scheduling device is implemented as a terminal device such as a smart phone, a tablet computer, or a desktop computer, it can include the components in the dashed box in Figure 8; when the resource scheduling device is implemented as a server device such as a conventional server, a cloud server, or a server array , The components in the dashed box in Figure 8 may not be included.

Correspondingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, which can implement each step in the method embodiment shown in FIG. 7a and FIG. 7b when the computer program is executed.

FIG. 9 is a schematic structural diagram of an information prediction device provided by an exemplary embodiment of this application. As shown in FIG. 9, the information prediction device includes: a memory 91 and a processor 92.

The memory 91 is used to store computer programs, and can be configured to store various other data to support operations on the information prediction device. Examples of these data include instructions, messages, pictures, videos, etc. for any application or method operating on the information prediction device.

The processor 92 is coupled with the memory 91, and is configured to execute a computer program in the memory 91 to obtain current data of the target schedulable unit corresponding to the cost impact factor, and the current data includes the target schedulable unit and all the data of the target schedulable unit. The current power consumption data of the dependent infrastructure; according to the current power consumption data of the target schedulable unit and the dependent infrastructure, as well as the influence relationship between the pre-trained cost cost influencing factor and the cost cost, the target is predicted The cost of schedulable units.

In an optional embodiment, when the processor 92 predicts the cost of the target schedulable unit, it is specifically configured to: input the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost and cost prediction model. Cost cost prediction to obtain the cost cost of the target schedulable unit; among them, the cost cost prediction model reflects the impact relationship between the cost cost influencing factor and the cost cost.

Further, the processor 92 is further configured to: obtain historical data corresponding to the cost impact factor of the target schedulable unit, the historical data including historical power consumption data of the target schedulable unit and the infrastructure on which it depends; schedulable according to the target Model training is performed on the historical power consumption data of the unit and the infrastructure on which it depends, and the cost prediction model is obtained.

Further optionally, based on the aforementioned cost cost prediction model, when the processor 92 performs cost cost prediction, it is specifically configured to: input current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost cost prediction model; In the cost prediction model, the weight coefficient of the first type that has been trained is used to perform a weighted summation of the current power consumption data of the infrastructure on which the target schedulable unit depends, and the result of the weighted summation is compared with the value of the target schedulable unit. The current power consumption data is added to obtain the cost price of the target schedulable unit; among them, the first type of weight coefficient reflects the impact of the cost impact factor of infrastructure power consumption on the cost of the schedulable unit.

In an optional embodiment, the historical data acquired by the processor 92 further includes at least one of the following data: the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and the data running on the target schedulable unit. History program type. Based on this, the processor 92 is specifically used for model training according to the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the historical physical location of the target schedulable unit, and the location of the target schedulable unit. Model training is performed on at least one of the historical electricity price price of the area and the historical program type running on the target schedulable unit to obtain a cost prediction model.

Correspondingly, the current data acquired by the processor 92 also includes at least one of the following data: the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the type of program that needs to be run on the target schedulable unit. Based on this, when the processor 92 predicts the cost, it is specifically configured to: combine the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, and the target schedulable unit. At least one of the current electricity price in the area and the type of program that needs to be run on the target schedulable unit is input into a cost prediction model to perform cost cost prediction to obtain the cost of the target schedulable unit.

Further optionally, when the processor 92 performs cost cost prediction, it is specifically configured to: combine the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, and the target schedulable unit. At least one of the current electricity price in the area where the dispatching unit is located and the type of program that needs to be run on the target schedulable unit are input into the cost prediction model; inside the cost prediction model, the first type of weight coefficient pair that has been trained is used Perform a weighted summation of the current power consumption data of the infrastructure on which the target schedulable unit depends, and add the result of the weighted summation to the current power consumption data of the target schedulable unit to obtain the initial cost of the target schedulable unit; And use the second type of weight coefficients that have been trained to modify the initial cost cost to obtain the cost cost of the target schedulable unit;

Among them, the first type of weight coefficient reflects the impact of the cost of infrastructure power consumption on the cost of schedulable units, and the second type of weight coefficient reflects at least one of the cost of physical location, electricity price, and program type. Impact on the cost of schedulable units.

Further optionally, when the processor 92 modifies the initial cost cost, it is specifically configured to calculate the product of the second type of weight coefficient and the initial cost cost as the cost cost of the target schedulable unit.

In an optional embodiment, when the processor 92 obtains the current data corresponding to the cost impact factor of the target schedulable unit, it is specifically used to: from the current power consumption of each schedulable unit in the computer room system or the data center system In the data, determine the current power consumption data of the target schedulable unit; according to the equipment topological relationship of the computer room system or data center system, filter out the target schedulable unit from the current power consumption data of each infrastructure of the computer room system or data center system Current power consumption data of the infrastructure on which it depends.

Further, as shown in FIG. 9, the resource scheduling device further includes: a communication component 93, a display 94, a power supply component 95, an audio component 95 and other components. Only some components are schematically shown in FIG. 9, which does not mean that the resource scheduling device only includes the components shown in FIG. 9. In addition, according to the different implementation forms of the resource scheduling device, the components in the dashed box in FIG. 9 are optional components, not mandatory components. For example, when the resource scheduling device is implemented as a terminal device such as a smart phone, a tablet computer, or a desktop computer, it can include the components in the dashed box in Figure 9; when the resource scheduling device is implemented as a server device such as a conventional server, a cloud server, or a server array , The components in the dashed box in Figure 9 may not be included.

Correspondingly, the embodiment of the present application also provides a computer-readable storage medium storing a computer program, which can implement each step in the method embodiment shown in FIG. 7c when the computer program is executed.

The communication components in Figs. 8 and 9 described above are configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G/LTE, 5G and other mobile communication networks, or a combination of them. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component may further include a near field communication (NFC) module, radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology Wait.

The above-mentioned display in Figs. 8 and 9 includes a screen, and the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation.

The power supply components in Figures 8 and 9 above provide power for various components of the equipment where the power supply components are located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device where the power supply component is located.

The audio components in Figs. 8 and 9 described above may be configured to output and/or input audio signals. For example, the audio component includes a microphone (MIC). When the device where the audio component is located is in an operating mode, such as call mode, recording mode, and voice recognition mode, the microphone is configured to receive external audio signals. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component further includes a speaker for outputting audio signals.

Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

The present invention is described with reference to flowcharts and/or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing equipment are generated It is a device that realizes the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device. The device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to produce computer-implemented processing, so as to execute on the computer or other programmable equipment. The instructions provide steps for implementing the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

In a typical configuration, the computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include non-permanent memory in a computer readable medium, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer readable media.

Computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, Magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. According to the definition in this article, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.

It should also be noted that the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or equipment including a series of elements not only includes those elements, but also includes Other elements that are not explicitly listed, or they also include elements inherent to such processes, methods, commodities, or equipment. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, commodity, or equipment that includes the element.

The foregoing descriptions are only examples of the present application, and are not used to limit the present application. For those skilled in the art, this application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in the scope of the claims of this application.

Claims

A resource scheduling method, characterized in that it comprises:

Acquiring current data corresponding to the cost impact factor of at least one schedulable unit, where the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends;

According to the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, and the pre-trained impact factor of the cost cost and the impact relationship between the cost cost, predict the performance of the at least one schedulable unit Cost price

According to the cost of the at least one schedulable unit, resource scheduling is performed on the at least one schedulable unit.
The method according to claim 1, characterized in that there is a relationship between the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, and the pre-trained cost impact factor and the cost price. The impact relationship of predicting the cost of the at least one schedulable unit includes:

Input the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, respectively, into the respective cost cost prediction models to perform cost cost prediction, so as to obtain the cost cost of the at least one schedulable unit;

Among them, the cost prediction model corresponding to each schedulable unit reflects the influence relationship existing between the corresponding cost cost influencing factor and the cost cost.
The method according to claim 2, characterized in that, before the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends are respectively input into respective cost prediction models for cost cost prediction, Also includes:

Acquiring historical data corresponding to the cost impact factor of the at least one schedulable unit, the historical data including historical power consumption data of the at least one schedulable unit and the infrastructure on which it depends;

Model training is performed according to the historical power consumption data of the at least one schedulable unit and the infrastructure on which it depends, respectively, to obtain a cost prediction model corresponding to each of the at least one schedulable unit.
The method according to claim 3, characterized in that the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends are respectively input into respective cost prediction models for cost cost prediction, so as to obtain The cost of the at least one schedulable unit includes:

For the target schedulable unit, input the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model corresponding to the target schedulable unit;

In the cost prediction model corresponding to the target schedulable unit, the weight coefficient of the first type that has been trained is used to perform a weighted summation of the current power consumption data of the infrastructure on which the target schedulable unit depends, and the weighted The sum result is added to the current power consumption data of the target schedulable unit to obtain the cost of the target schedulable unit;

Wherein, the weight coefficient of the first type reflects the impact of the cost-cost influencing factor of infrastructure power consumption on the cost of the schedulable unit; the target schedulable unit is any schedulable unit in the at least one schedulable unit .
The method according to claim 3, wherein the historical data further includes at least one of the following: the historical physical location of the at least one schedulable unit, the historical electricity price of the area where the at least one schedulable unit is located, and The type of historical program running on the at least one schedulable unit;

Model training is performed according to the historical power consumption data of the at least one schedulable unit and the infrastructure on which it depends, respectively, to obtain the cost prediction model corresponding to each of the at least one schedulable unit, including:

For the target schedulable unit, according to the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the historical physical location of the target schedulable unit, and the historical electricity bill of the area where the target schedulable unit is located Model training is performed on at least one of the price and historical program types running on the target schedulable unit to obtain a cost prediction model corresponding to the target schedulable unit;

Wherein, the target schedulable unit is any schedulable unit in the at least one schedulable unit.
The method according to claim 5, wherein the current data further comprises at least one of the following: the current physical location of the at least one schedulable unit, the current price of electricity in the area where the at least one schedulable unit is located, and The type of program that needs to be run on the at least one schedulable unit;

Inputting the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends respectively into their respective cost prediction models to perform cost cost prediction to obtain the cost cost of the at least one schedulable unit includes:

For the target schedulable unit, the current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, and the current electricity bill of the area where the target schedulable unit is located At least one data of the price and the type of program that needs to be run on the target schedulable unit is input into the cost price prediction model corresponding to the target schedulable unit to perform cost price prediction, so as to obtain the cost price of the target schedulable unit.
The method according to claim 6, wherein the current power consumption data of the target schedulable unit and the infrastructure on which it depends are combined with the current physical location of the target schedulable unit and the target schedulable unit. At least one of the current electricity price price of the area where the dispatch unit is located and the type of program that needs to be run on the target schedulable unit is input into the cost prediction model corresponding to the target schedulable unit to predict the cost to obtain the target The cost of schedulable units includes:

The current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the target schedulable unit Data of at least one of the types of programs that need to be run on the scheduling unit is input into the cost prediction model corresponding to the target schedulable unit;

In the cost prediction model corresponding to the target schedulable unit, the weight coefficient of the first type that has been trained is used to perform a weighted summation of the current power consumption data of the infrastructure on which the target schedulable unit depends, and the weighted The result of the summation is added to the current power consumption data of the target schedulable unit to obtain the initial cost cost of the target schedulable unit; and the second type of weight coefficient that has been trained is used to perform the initial cost cost Correction to obtain the cost of the target schedulable unit;

Wherein, the weight factor of the first type reflects the impact of the cost factor of infrastructure power consumption on the cost of schedulable units, and the weight factor of the second type reflects at least one of the physical location, electricity price, and program type. The impact of a cost cost influence factor on the cost cost of a schedulable unit.
The method according to any one of claims 1-7, wherein before obtaining the current data corresponding to the cost impact factor of at least one schedulable unit, the method further comprises:

According to the resource demand corresponding to this resource scheduling, at least one schedulable unit that can be schedulable this time is determined from the schedulable units included in the computer room system, data center system, or edge computing system.
The method according to claim 8, wherein at least one schedulable unit is determined from the schedulable units included in the computer room system, data center system, or edge computing system according to the resource demand corresponding to this resource scheduling. The dispatch unit includes:

According to the device topology relationship of the computer room system, data center system, or edge computing system, determine and execute the resource scheduling device for this resource scheduling from each schedulable unit included in the computer room system, data center system, or edge computing system At least one schedulable unit whose distance meets the set requirements.
The method according to claim 9, characterized in that, according to the device topological relationship of the computer room system or the data center system, the current resource is determined and executed from the schedulable units included in the computer room system or the data center system At least one schedulable unit whose distance of the scheduled resource scheduling device meets the set requirement includes any one of the following:

According to the device topology relationship of the computer room system or the data center system, determine at least one schedulable unit located in the same computer room as the resource scheduling device from each schedulable unit included in the computer room system or the data center system;

According to the device topology relationship of the computer room system or the data center system, determine at least one schedulable unit that is located in the same channel as the resource scheduling device from each schedulable unit included in the computer room system or the data center system;

Determine at least one schedulable unit located in the same cabinet as the resource scheduling device from each schedulable unit included in the computer room system or the data center system according to the device topology relationship of the computer room system or the data center system;

According to the device topology relationship of the computer room system or the data center system, from each schedulable unit included in the computer room system or the data center system, it is determined that the physical distance to the resource scheduling device is less than at least a set distance threshold. A schedulable unit.
The method according to claim 8, wherein the schedulable unit included in the computer room system or the data center system is a server in the computer room system or the data center system;

The schedulable unit included in the edge computing system is an edge computing node in the edge computing system.
The method according to any one of claims 1-7, wherein the resource scheduling of the at least one schedulable unit according to the cost of the at least one schedulable unit comprises:

According to the cost of the at least one schedulable unit, perform resource scheduling on the at least one schedulable unit in the descending order of the cost; or

According to the cost of the at least one schedulable unit, the schedulable unit with the least cost is prioritized for resource scheduling; or

According to the cost of the at least one schedulable unit, priority is given to resource scheduling for schedulable units whose cost cost is less than the set threshold.
The method according to any one of claims 1-7, wherein the resource scheduling of the at least one schedulable unit according to the cost of the at least one schedulable unit comprises:

Perform resource scheduling on the at least one schedulable unit according to the service requirements corresponding to the current resource scheduling and in combination with the cost of the at least one schedulable unit.
The method according to claim 13, wherein the resource scheduling of the at least one schedulable unit according to the service requirements corresponding to this resource scheduling and the cost of the at least one schedulable unit comprises:

According to the service performance requirement corresponding to this resource scheduling, select a schedulable unit that meets the service performance requirement from the at least one schedulable unit;

According to the cost of the schedulable unit that meets the service performance requirement, resource scheduling is performed on the schedulable unit that meets the service performance requirement.
An information prediction method, characterized in that it includes:

Acquiring current data corresponding to the cost impact factor of the target schedulable unit, where the current data includes current power consumption data of the target schedulable unit and the infrastructure on which it depends;

Predict the cost cost of the target schedulable unit based on the current power consumption data of the target schedulable unit and the infrastructure on which it depends, and the pre-trained impact factor of the cost cost and the impact relationship between the cost cost .
The method according to claim 15, characterized in that, according to the current power consumption data of the target schedulable unit and the infrastructure on which it depends, and the influence relationship between the pre-trained cost cost influencing factor and the cost cost , Predict the cost of the target schedulable unit, including:

Input the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model to perform cost cost prediction, so as to obtain the cost cost of the target schedulable unit;

Wherein, the cost price prediction model reflects the influence relationship existing between the cost price influencing factor and the cost price.
The method according to claim 16, characterized in that, before inputting the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model for cost cost prediction, the method further comprises:

Acquiring historical data of the target schedulable unit corresponding to the cost impact factor, where the historical data includes historical power consumption data of the target schedulable unit and the infrastructure on which it depends;

Model training is performed according to the historical power consumption data of the target schedulable unit and the infrastructure on which it depends to obtain the cost prediction model.
The method according to claim 17, characterized in that the current power consumption data of the target schedulable unit and the infrastructure on which it depends are input into a cost prediction model for cost cost prediction, so as to obtain the target schedulable unit The cost of the unit, including:

Input the current power consumption data of the target schedulable unit and the infrastructure on which it depends into the cost prediction model;

In the cost prediction model, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the weight coefficients of the first type that have been trained, and the weighted sum result is combined with all Adding the current power consumption data of the target schedulable unit to obtain the cost of the target schedulable unit;

Wherein, the weight coefficient of the first type reflects the impact of the cost impact factor of infrastructure power consumption on the cost of the schedulable unit.
The method according to claim 17, wherein the historical data further comprises at least one of the following data: the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and all Describe the historical program types running on the target schedulable unit;

The performing model training according to the historical power consumption data of the target schedulable unit and the infrastructure on which it depends to obtain the cost prediction model includes:

According to the historical power consumption data of the target schedulable unit and the infrastructure on which it depends, as well as the historical physical location of the target schedulable unit, the historical electricity price of the area where the target schedulable unit is located, and the target schedulable unit Model training is performed on at least one type of data from historical program types running on the scheduling unit to obtain the cost prediction model.
The method according to claim 19, wherein the current data further includes at least one of the following data: the current physical location of the target schedulable unit, the current price of electricity in the area where the target schedulable unit is located, and all Describe the types of programs that need to be run on the target schedulable unit;

The current power consumption data of the target schedulable unit and the infrastructure on which it depends are input into the cost cost prediction model to perform cost cost prediction to obtain the cost cost of the target schedulable unit, including:

The current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the target schedulable unit At least one data of the type of program to be run on the scheduling unit is input into the cost cost prediction model to perform cost cost prediction, so as to obtain the cost cost of the target schedulable unit.
The method according to claim 20, wherein the current power consumption data of the target schedulable unit and the infrastructure on which it depends are combined with the current physical location of the target schedulable unit and the target schedulable unit. At least one data of the current electricity price price of the area where the dispatch unit is located and the type of program that needs to be run on the target schedulable unit are input into the cost prediction model for cost cost prediction to obtain the cost of the target schedulable unit The cost includes:

The current power consumption data of the target schedulable unit and the infrastructure on which it depends, together with the current physical location of the target schedulable unit, the current electricity price of the area where the target schedulable unit is located, and the target schedulable unit Data of at least one of the types of programs that need to be run on the scheduling unit are input into the cost prediction model together;

In the cost prediction model, the weighted sum of the current power consumption data of the infrastructure on which the target schedulable unit relies is weighted and summed with the weight coefficients of the first type that have been trained, and the weighted sum result is combined with all the weighted sums. Adding the current power consumption data of the target schedulable unit to obtain the initial cost of the target schedulable unit; and using the second type of weight coefficient that has been trained to correct the initial cost to obtain the target The cost of schedulable units;

Wherein, the weight coefficient of the first type reflects the impact of the cost of infrastructure power consumption on the cost of schedulable units, and the weight coefficient of the second type reflects at least one of physical location, electricity price, and program type The impact of the cost impact factor on the cost of schedulable units.
The method according to claim 21, characterized in that, using the second type of weight coefficients that have been trained to correct the initial cost cost to obtain the cost cost of the target schedulable unit comprises:

The product of the weight coefficient of the second type and the initial cost is calculated as the cost of the target schedulable unit.
The method according to any one of claims 15-22, wherein obtaining current data corresponding to the cost impact factor of the target schedulable unit comprises:

Determine the current power consumption data of the target schedulable unit from the current power consumption data of each schedulable unit in the computer room system, data center system or edge computing system;

According to the device topology relationship of the computer room system, data center system, or edge computing system, filter out the target schedulable unit from the current power consumption data of each infrastructure of the computer room system, data center system, or edge computing system Current power consumption data of the infrastructure on which it depends.
The method according to any one of claims 15-22, wherein the target schedulable unit is a server in a computer room system or a data center system; or, the target schedulable unit is an edge computing system in an edge computing system. calculate node.
A resource scheduling device, characterized by comprising: a memory and a processor;

The memory is used to store a computer program;

The processor, coupled with the memory, is configured to execute the computer program for:

Acquiring current data corresponding to the cost impact factor of at least one schedulable unit, where the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends;

According to the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, and the pre-trained impact factor of the cost cost and the impact relationship between the cost cost, predict the performance of the at least one schedulable unit Cost price

According to the cost of the at least one schedulable unit, resource scheduling is performed on the at least one schedulable unit.
An information prediction device, characterized by comprising: a memory and a processor;

The memory is used to store a computer program;

The processor, coupled with the memory, is configured to execute the computer program for:

Acquiring current data corresponding to the cost impact factor of the target schedulable unit, where the current data includes current power consumption data of the target schedulable unit and the infrastructure on which it depends;

Predict the cost cost of the target schedulable unit based on the current power consumption data of the target schedulable unit and the infrastructure on which it depends, and the pre-trained impact factor of the cost cost and the impact relationship between the cost cost .
A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to implement the steps in the method of any one of claims 1-24.
A computer room system, characterized by comprising: several schedulable units, several infrastructures and resource scheduling equipment; the several infrastructures provide basic services for the several schedulable units;

The resource scheduling device is configured to obtain current data corresponding to the cost impact factor of at least one schedulable unit, and the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends Predict the at least one schedulable unit based on the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price According to the cost price of the at least one schedulable unit, perform resource scheduling on the at least one schedulable unit; wherein the at least one schedulable unit comes from the several schedulable units.
A computer room system, characterized by comprising: a number of schedulable units, a number of infrastructures, information prediction equipment and resource scheduling equipment; the several infrastructures provide basic services for the several schedulable units;

The information prediction device is configured to obtain current data corresponding to the cost impact factor of at least one schedulable unit, and the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends Predict the at least one schedulable unit based on the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price The cost of the cost; the cost of the at least one schedulable unit is provided to the resource scheduling device;

The resource scheduling device is configured to perform resource scheduling on the at least one schedulable unit according to the cost of the at least one schedulable unit provided by the information prediction device; wherein the at least one schedulable unit is from The number of schedulable units.
A data center system is characterized by comprising: at least one computer room and resource scheduling equipment; each computer room includes several schedulable units and several infrastructures, and the several infrastructures provide for the several schedulable units Basic services;

The resource scheduling device is configured to obtain current data corresponding to the cost impact factor of at least one schedulable unit, and the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends Predict the at least one schedulable unit based on the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price According to the cost of the at least one schedulable unit, resource scheduling is performed on the at least one schedulable unit; wherein the at least one schedulable unit comes from the schedulable unit in the at least one computer room.
A data center system is characterized by comprising: at least one computer room, information prediction equipment, and resource scheduling equipment; each computer room includes a number of schedulable units and a number of infrastructures, the number of infrastructures being the number of The schedulable unit provides basic services;

The information prediction device is configured to obtain current data corresponding to the cost impact factor of at least one schedulable unit, and the current data includes current power consumption data of the at least one schedulable unit and the infrastructure on which it depends Predict the at least one schedulable unit based on the current power consumption data of the at least one schedulable unit and the infrastructure on which it depends, as well as the pre-trained cost impact factor and the impact relationship between the cost price The cost of the cost; the cost of the at least one schedulable unit is provided to the resource scheduling device;

The resource scheduling device is configured to perform resource scheduling on the at least one schedulable unit according to the cost of the at least one schedulable unit provided by the information prediction device; wherein the at least one schedulable unit is from The schedulable unit in the at least one computer room.
An edge computing system, characterized in that it comprises: several edge computing nodes, several infrastructures and servers; the several infrastructures provide basic services for the several edge computing nodes;

The server is configured to obtain current data corresponding to the cost impact factor of at least one edge computing node, where the current data includes current power consumption data of the at least one edge computing node and the infrastructure on which it depends; Describe the current power consumption data of the at least one edge computing node and the infrastructure on which it depends, and the pre-trained cost impact factor and the impact relationship between the cost price, and predict the cost price of the at least one edge computing node; According to the cost of the at least one edge computing node, resource scheduling is performed on the at least one edge computing node; wherein the at least one edge computing node comes from the plurality of edge computing nodes.