A method and system for optimizing and scheduling data center task load through computer-based and thermal collaboration

Abstract

The present invention relates to the technical field of data center energy conservation, and in particular to a method and system for optimizing and scheduling data center task loads in an electric-computing-heat coordinated manner; the method of the present invention comprises: according to a data center cooling system, modeling the power, computing power and thermal power consumption involved in various areas of a data center to obtain a power, computing power and thermal power consumption model; calculating the total energy consumption of various areas of the data center according to the power, computing power and thermal power consumption model and establishing an objective function with the minimum cost of data center operation in combination with real-time electricity prices; optimizing and scheduling the task load of the data center according to the objective function; the present invention models the task load scheduling process, including physical models such as temperature changes, variable air volume systems, and load spatiotemporal distribution, taking into account real-time electricity prices and regional energy efficiency differences, achieving spatiotemporal scheduling of task loads between different areas of a single data center, and reducing the cost of data center operators.

Description

Technical Field

[0001] The present invention relates to the technical field of data center energy conservation, and in particular to a method and system for optimizing the task load scheduling of a data center with electric-computing-thermal coordination. Background Art

[0002] At present, the interactive development of digital technologies represented by cloud computing, big data, artificial intelligence, the Internet of Things and 5G blockchain, and the integration with the traditional real economy have gradually brought mankind into the digital economy era, and the amount of global data has exploded. As an important physical carrier for processing modern information technology, the scale and number of data centers are also increasing rapidly. The huge power consumption has brought a heavy economic burden to data center operators. How to tap into flexible resources, improve energy management, reduce energy consumption and reduce costs without affecting service quality has become the main problem facing its current development. In a data center, IT equipment that processes task loads and cooling systems that cool servers account for the largest proportion of energy consumption. The energy management optimization of the two is currently a focus of general concern in the industry.

[0003] Power usage effectiveness (PUE) is the most commonly used indicator to measure the global energy efficiency of a data center, and is defined as the ratio of the total annual energy consumption of a data center to the energy consumption of IT equipment. Currently, both the data center itself and the data center cluster use PUE as a constant to simply represent the energy consumption of the cooling system. However, the PUE of an actual data center is not fixed, because the output of the cooling system is related to its load rate. Even if the IT load energy consumption is the same at a certain moment, the cooling system output at different temperatures is different, the energy consumption is different, and the PUE is naturally different.

[0004] Partial Power Utilization Efficiency (Partial PUE, pPUE) is a derivative of the global PUE concept, reflecting the energy efficiency differences in different areas of a data center, such as floors and container modules. Improving the energy efficiency of areas with larger pPUE values ​​can improve the global PUE of the entire data center, which is also beneficial for reducing energy consumption and saving costs. However, few studies have considered the regional energy efficiency differences of a single data center to perform spatial scheduling of task loads. Summary of the invention

[0005] In view of the above-mentioned deficiencies of the prior art, the present invention aims to provide a data center task load scheduling method and system that coordinates electricity, computing power and heat, so as to solve the problem of energy waste caused by the existing data center task load scheduling that does not take into account the energy efficiency differences of various areas of the data center to perform task scheduling.

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

[0007] When the task load arrives, it is scheduled in an area with higher energy efficiency. If the computing resources in the area are insufficient or the temperature reaches the maximum limit, in order to meet the service quality and safe and stable operation of the server, the task load will be scheduled to the next area according to the energy efficiency. Since the distance between the areas of a single data center is very close, there is almost no delay in scheduling the task load in the spatial dimension, and the delay loss caused by the spatial scheduling of task loads in cross-regional data center clusters is not considered.

[0008] Moreover, for the data center itself, on the one hand, in order to ensure the quality of service, delay-tolerant loads can be scheduled at a time before the deadline when the electricity price is low. On the other hand, the power consumption of the cooling system can be reduced by increasing its indoor temperature set point. When the electricity price is low, the cooling system outputs more to keep the temperature at a low level. When the electricity price is high, the cooling system reduces the output by utilizing thermal inertia and can also maintain the temperature within the limit for a period of time. These two characteristics can significantly improve the flexibility of task load scheduling in the time dimension of the data center. For data center clusters, operators have data centers distributed in multiple geographical locations. Different geographical locations create different resources and electricity prices. The flexibility of the time-space scheduling of data center task loads can be achieved by allocating task loads through front-end servers.

[0009] Based on the above ideas, a data center task load optimization scheduling method with electric-computing-thermal coordination is formed, including:

[0010] According to the cooling system of the data center, the power, computing power and heat consumption involved in each area of ​​the data center are modeled to obtain the power, computing power and heat consumption model.

[0011] The total energy consumption of each area of ​​the data center is calculated based on the power, computing power and heat consumption model, and the objective function is established with the minimum cost of data center operation in combination with the real-time electricity price.

[0012] According to the objective function, the data center task load is optimized and scheduled.

[0013] Furthermore, the objective function also includes wind power energy and photovoltaic energy obtained by the data center; the objective function is:

[0014]

[0015] in, For real-time electricity prices, and are the time interval and total time respectively, and are the wind power energy and photovoltaic energy obtained by the data center at time t, is the total energy consumption of the data center.

[0016] Furthermore, the power consumption involved in the data center is modeled, and the power consumption model includes:

[0017] The power consumption of the data center is calculated based on the energy consumption of IT equipment in all computer rooms in the data center, the total energy consumption of the cooling system, and the energy consumption of lighting and uninterruptible power supply UPS, and a power consumption model is constructed:

[0018]

[0019] in, is the energy consumption of all IT equipment in the computer room at time t in area i, is the total energy consumption of the cooling system at time t, is the energy consumption of lighting and uninterruptible power supply (UPS) at time t.

[0020] Further, the thermal consumption model includes an indoor temperature model, a fan model, a chiller model and a cooling system model;

[0021] The indoor temperature model is formed by calculating the indoor temperature using the first thermodynamic formula based on the heat capacity of the hot aisle in the area, the cooling provided by the cooling unit and the fan in the area, the heat generated by the servers in the area, the heat transfer of the exterior walls in the area, the heat generated by the solar radiation on the opaque surface of the exterior walls in the area, and the heat generated by the lighting in the area, and by calculating the supply air temperature using the first thermodynamic formula based on the heat capacity of the cold aisle in the area, the cooling provided by the cooling unit and the fan.

[0022] The fan model is formed based on a quadratic function of the fan energy consumption and its air supply volume.

[0023] The chiller model is formed based on the power consumption of the chillers in a certain area and the total energy consumption of the chillers.

[0024] The cooling system model is formed based on the total energy consumption of the chiller, the total energy consumption of the fan and the constant energy consumption to obtain the total energy consumption of the entire cooling system of the data center.

[0025] Further, the indoor temperature model is:

[0026]

[0027] in, is the indoor temperature of area i at time t, is the indoor temperature of area i at time t+1, is the supply air temperature of area i at time t, is the supply air temperature of area i at time t+1, and are the heat capacity of the hot channel and the cold channel in region i, respectively. The cooling capacity provided by the cooling unit and fan for zone i at time t, The heat generated by the server in region i at time t, is the heat transfer of the exterior wall in region i at time t, is the heat generated by solar radiation on the opaque surface of the exterior wall in area i at time t, is the heat generated by lighting in area i at time t.

[0028] The cooling capacity provided by the cooling unit and fan in area i at time t :

[0029]

[0030] in, is the number of fans in area i, is the wind speed of the fan in area i at time t, is the amount of air substance.

[0031] The server in region i generates heat at time t :

[0032] .

[0033] The heat transfer of the exterior wall in area i at time t :

[0034]

[0035] in, is the heat transfer coefficient of the exterior wall in area i, is the external wall area of ​​region i, is the outdoor temperature at time t.

[0036] The heat generated by solar radiation on the opaque surface of the exterior wall in area i at time t :

[0037]

[0038] in, is the thermal radiation intensity of the sun in direction o at time t, where o includes five directions: east, south, west, north, and horizontal. is the solar radiation area of ​​the exterior wall of region i in direction o, is the heat transfer coefficient of region i, is the heat absorption coefficient of the exterior wall in area i, is the thermal resistance of region i, is the heat transfer coefficient of the exterior wall in region i.

[0039] The heat generated by lighting in area i at time t :

[0040]

[0041] in, is the unit lighting density, is the lighting area of ​​region i.

[0042] The fan model is:

[0043]

[0044] The chiller model is:

[0045]

[0046]

[0047] in, is the total energy consumption of the chiller at time t, is the full load capacity of the chiller, is the power consumption of the chiller in area i at time t.

[0048] The power consumption of the chiller in area i at time t :

[0049]

[0050] in, is the heat transfer coefficient of the chiller coil, The coefficient of performance of the chiller is:

[0051]

[0052] in, It is the performance related coefficient of the chiller.

[0053] The cooling system model is:

[0054]

[0055] in, is the energy consumption of the entire cooling system of the data center at time i, is the total energy consumption of the fan at time t, is the constant energy consumption, including the energy consumption of circulating water pump and cooling tower fan.

[0056] Furthermore, the computing power consumption involved in the data center is modeled to obtain a computing power consumption model, including: dividing the task load into delay-sensitive load and delay-tolerant load according to the task load of the data center; calculating the load that needs to be processed in a certain area at a certain moment according to the arrival amount of delay-sensitive load in a certain area at a certain moment, the category of delay-tolerant load, the number of demands for each category of delay-tolerant load, the arrival time and the deadline; calculating the energy consumption of all IT equipment in a certain computer room in a certain area at a certain moment according to the load that needs to be processed in a certain area at a certain moment, and forming a computing power consumption model.

[0057] Furthermore, the computing power consumption model is:

[0058]

[0059] in, is the number of servers in region i, is the energy consumption of the server in idle state in region i, is the energy consumption of the server in the busy state of region i, is the load that needs to be processed in region i at time t, is the total computing capacity of region i.

[0060] The load that needs to be processed in the i region at time t :

[0061]

[0062] in, is the amount of delay-sensitive load allocated to region i at time t, and , The total number of requests that need to be processed at time t for zone i of type j workload between arrival and deadline for delay-tolerant workload.

[0063] Furthermore, the cooling capacity provided by the cooling unit and fan in area i at time t is and the power consumption of the chiller in area i at time t After linearization, a mixed integer quadratic programming problem is formed, and the Gurobi solver is used to solve the mixed integer quadratic programming problem.

[0064] A data center task load optimization scheduling system with electric-computing-thermal collaboration includes a model building module, an objective function building module and an optimization scheduling module.

[0065] The model building module is used to model the power, computing power and heat consumption involved in the data center according to the data center cooling system to obtain the power, computing power and heat consumption model.

[0066] The objective function establishment module is used to calculate the total energy consumption of the data center according to the power, computing power and heat consumption model and establish the objective function with the minimum cost of data center operation in combination with the real-time electricity price.

[0067] The optimization scheduling module is used to optimize the data center task load according to the objective function.

[0068] Furthermore, the objective function also includes wind power energy and photovoltaic energy obtained by the data center; the objective function is:

[0069]

[0070] in, For real-time electricity prices, and are the time interval and total time respectively, and are the wind power energy and photovoltaic energy obtained by the data center at time t, is the total energy consumption of the data center.

[0071] The beneficial effects of the present invention are as follows: the method and system of the present invention model the task load scheduling process, including physical models such as temperature changes, variable air volume systems, and load spatiotemporal distribution, taking into account real-time electricity prices and regional energy efficiency differences, to achieve spatiotemporal scheduling of task loads among different areas of a single data center, thereby reducing the costs of data center operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of a data center task load optimization scheduling method for electrical-computing-thermal collaboration in Example 1 of the present invention.

[0073] Figure 2 Schematic diagram of a data center task load optimization scheduling system with electrical computing and thermal collaboration in Example 1 of the present invention.

[0074] Figure 3 This is a diagram of wind speed, air supply temperature and room temperature after optimized scheduling of Example 1 in Example 2 of the present invention.

[0075] Figure 4 This is the spatiotemporal distribution diagram of the task load after the optimized scheduling of Example 1 in Example 2 of the present invention.

[0076] Figure 5 This is the energy consumption graph of the data center at each moment in Example 1 in Example 2 of the present invention.

[0077] Figure 6 This is a diagram of wind speed, air supply temperature and room temperature after scheduling in Example 2 of Example 2 of Embodiment 2 of the present invention.

[0078] Figure 7This is the time-space distribution diagram of the task load after scheduling of Example 2 in Example 2 of Embodiment 2 of the present invention.

[0079] Figure 8 This is a diagram of wind speed, air supply temperature and room temperature after scheduling of Example 3 in Example 2 of the present invention.

[0080] Fig. 9 This is the time-space distribution diagram of the task load after scheduling of Example 3 in Example 2 of the present invention.

[0081] Fig.10 This is a comparison chart of energy consumption at different times of Examples 1-3 in Example 2 of the present invention. DETAILED DESCRIPTION

[0082] The present invention is further described in detail below in conjunction with specific embodiments.

[0083] It should be noted that these embodiments are only used to illustrate the present invention, rather than to limit the present invention. Simple improvements to the method based on the concept of the present invention all fall within the scope of protection claimed by the present invention.

[0084] Example 1

[0085] See also Figure 1 , which is a data center task load optimization scheduling method for electric-computing-thermal coordination, including:

[0086] S100. Model the power, computing power, and heat consumption involved in each area of ​​the data center according to the data center cooling system to obtain a power, computing power, and heat consumption model.

[0087] Generally, data centers have high cooling demand and high external temperature (>13℃). Most cooling systems use centrifugal chillers combined with precision air conditioners CRAH, and the chillers use mechanical refrigeration. The space capacity, number of servers, and energy consumption characteristics of each area are different, and the CRAHs are equipped differently. The cooling process of the entire data center is as follows: the chilled water in the coil generates cold air by exchanging heat with the indoor hot return air. The blower blows the cold air into the static pressure room (floor) and then flows into the closed cold channel through the perforated floor. Due to the positive pressure difference between the cold and hot channels, the cold air flows from the cold channel through the server into the hot channel to form hot return air that flows into the CRAH, completing a cycle in the room.

[0088] As an implementable method, the power consumption involved in the data center is modeled, and the power consumption model obtained includes:

[0089] The power consumption of the data center is calculated based on the energy consumption of IT equipment in all computer rooms in the data center, the total energy consumption of the cooling system, and the energy consumption of lighting and uninterruptible power supply UPS, and a power consumption model is constructed:

[0090]

[0091] in, is the energy consumed by the server in region i at time t, is the total energy consumption of the cooling system at time t, = is the energy consumption of lighting and uninterruptible power supply (UPS) at time t. In order to ensure the service quality for users, the data center needs to operate with uninterruptible power supply 24 / 7. The energy consumption of lighting and uninterruptible power supply (UPS) is almost unchanged and accounts for no more than 10%. Treated as a constant.

[0092] Coordinate the output of renewable energy such as wind or solar energy to reduce carbon emissions, which is in line with the concept of economic, green and sustainable development of data centers. The wind and solar energy absorbed at time t cannot exceed the maximum value at that time:

[0093] .

[0094] As an implementable, the thermal consumption model includes an indoor temperature model, a fan model, a chiller model and a cooling system model.

[0095] Among them, the indoor temperature model is based on the heat capacity of the hot channel in the area, the cooling capacity provided by the cooling unit and fan in the area, the heat generated by the servers in the area, the heat transfer of the exterior wall in the area, the heat generated by the solar radiation on the opaque surface of the exterior wall in the area, and the heat generated by the lighting in the area. The indoor temperature is calculated using the first thermodynamic formula, and the supply air temperature is calculated based on the heat capacity of the cold channel in the area, the cooling capacity provided by the cooling unit and fan.

[0096] Since the hot and cold aisles are closed, the indoor reheat cycle and leakage airflow are not considered, so the indoor temperature and supply air temperature They can be expressed by the first formula of thermodynamics:

[0097] .

[0098] Expressing it in quasi-steady-state form:

[0099] .

[0100] in, is the indoor temperature of area i at time t, is the indoor temperature of area i at time t+1, is the supply air temperature of area i at time t, is the supply air temperature of area i at time t+1, and are the heat capacity of the hot channel and the cold channel in region i, respectively. is the energy consumption of the cooling system in area i at time t, The heat generated by the server in region i at time t, is the heat transfer of the exterior wall in region i at time t, is the heat generated by solar radiation on the opaque surface of the exterior wall in area i at time t, is the heat generated by lighting in area i at time t.

[0101] The energy consumption of the cooling system in the above region i at time t is :

[0102]

[0103] in, is the number of fans in area i, is the wind speed of the fan in area i at time t, is the amount of air substance.

[0104] The server heat generation in the above region i at time t , the energy consumption of the server in region i at time t All converted into heat energy:

[0105] .

[0106] The external wall heat transfer of the above region i at time t :

[0107] .

[0108] in, is the heat transfer coefficient of the exterior wall in area i, is the external wall area of ​​region i, is the outdoor temperature at time t.

[0109] The heat generated by solar radiation on the opaque surface of the exterior wall in the above region i at time t :

[0110] .

[0111] in, is the thermal radiation intensity of the sun in direction o at time t, where o includes five directions: east, south, west, north, and horizontal. is the solar radiation area of ​​the exterior wall of region i in direction o, is the heat transfer coefficient of region i, is the heat absorption coefficient of the exterior wall in area i, is the thermal resistance of region i, is the heat transfer coefficient of the exterior wall in region i.

[0112] The heat generated by the lighting in the above area i at time t , which is usually a constant:

[0113] .

[0114] in, is the unit lighting density, is the lighting area of ​​region i.

[0115] In order to eliminate hot spots in the area and keep the server running efficiently and safely, the air supply temperature of the CRAH fan will be set at a lower value regardless of the load, and the fan air supply volume will be large, which will lead to waste of resources due to overcooling of the data center. Therefore, according to the load situation, the air supply temperature value can be reset while meeting the indoor temperature requirements. When the air supply temperature increases, the energy consumption of the chiller will be reduced accordingly. At the same time, the fluctuation of the air supply temperature cannot be too large, and the room temperature difference between adjacent time periods is not greater than 5. That is:

[0116] .

[0117] As an implementation method, the fan model is based on a quadratic function of the fan energy consumption and its air delivery volume:

[0118] In addition to changing the room temperature, the present invention also considers the variable air volume supply of the CRAH fan, that is, the fan operates at a variable wind speed, and optimizes the air supply temperature and air supply volume to control the indoor temperature. The air volume meets the output constraint, and the fan energy consumption is related to the quadratic function of its air supply volume / wind speed; the fan model is:

[0119] .

[0120] As an implementation method, the chiller model is based on the power consumption of the chillers in a certain area and the total energy consumption of the chillers is obtained:

[0121] The power consumption of the chiller in area i at time t is:

[0122] .

[0123] in, is the heat transfer coefficient of the chiller coil, The coefficient of performance of the chiller is:

[0124] .

[0125] in, are the performance coefficients of the chiller, and their values ​​are 0.0068, 0.0008, and 0.458 respectively.

[0126] The total energy consumption of a centrifugal chiller is the sum of the energy consumed in each area, and the output is not greater than its full load capacity.

[0127]

[0128] .

[0129] in, is the total energy consumption of the chiller at time t, is the full load capacity of the chiller, is the power consumption of the chiller in area i at time t.

[0130] As an implementation method, the cooling system model is formed based on the total energy consumption of the chiller, the total energy consumption of the fan and the constant energy consumption to obtain the total energy consumption of the entire cooling system of the data center:

[0131] .

[0132] in, is the energy consumption of the entire cooling system of the data center at time i, is the total energy consumption of the fan at time t, is the constant energy consumption, including the energy consumption of circulating water pump and cooling tower fan.

[0133] As an implementable method, the computing power consumption involved in the data center is modeled to obtain a computing power consumption model, including: dividing the task load into delay-sensitive load and delay-tolerant load according to the task load of the data center; calculating the load required to be processed in a certain area at a certain time according to the arrival amount of delay-sensitive load in a certain area at a certain time, the category of delay-tolerant load, the number of demands for each category of delay-tolerant load, the arrival time and the deadline; calculating the energy consumption of all IT equipment in the computer room in a certain area at a certain time according to the load required to be processed in a certain area at a certain time, and forming a computing power consumption model.

[0134] The task loads of data centers are divided into two categories: delay-sensitive and delay-tolerant. The characteristic of delay-sensitive loads is that the data center must complete the processing of data computing requests submitted by users within seconds or even milliseconds. Delay-tolerant loads are computing requests submitted by users that are allowed to be processed for a long time, up to several hours or even a day, and can be flexibly arranged to various time periods before the deadline. The volume is relatively small, but the computing amount of each load is large.

[0135] The computing power consumption model is:

[0136] .

[0137] in, is the number of servers in region i, is the rated idle power consumption of a server in zone i, is the rated full-load power consumption of a server in zone i, is the load that needs to be processed in region i at time t, is the total computing capacity of region i.

[0138] The load that needs to be handled by region i at time t :

[0139] .

[0140] in, is the amount of delay-sensitive load allocated to region i at time t, and , is the amount of delay-tolerant load of type j scheduled to region i at time t.

[0141] Furthermore, the latency-sensitive load arriving at time t needs to be processed, and the load assigned to a region i Less than the total delay-sensitive load at time t ,Right now:

[0142] .

[0143] Furthermore, latency-tolerant workloads need to complete between arrival and deadline, and the total number of requests processed by each region at each moment Does not exceed its maximum parallelism ,Right now:

[0144] .

[0145] in, is the total number of requests per class for delay-tolerant workloads, is the arrival time, is the deadline.

[0146] As an implementable method, To solve the load nonlinearly, the energy consumption of the cooling system in area i at time t is and the total energy consumption of the chiller at time t After linearization, a mixed integer quadratic programming problem is formed, and the Gurobi solver is used to solve the mixed integer quadratic programming problem.

[0147] First, the energy consumption of the cooling system in area i at time t is : Perform linearization.

[0148] There are two bilinear product variables in the formula and , and use the McCormick envelope to relax it, so , , Introducing intermediate variables , .

[0149] , ,but , expanded to:

[0150] .

[0151] , ,but , expanded to:

[0152] .

[0153] , ,but , expanded to:

[0154] .

[0155] , ,but , expanded to:

[0156] .

[0157] Similarly,

[0158] make , , Introducing intermediate variables , ,but:

[0159] .

[0160] .

[0161] .

[0162] .

[0163] .

[0164] Secondly, the power consumption of the chiller in area i at time t To perform linearization:

[0165] .

[0166] .

[0167] .

[0168] because yes The function of When it is approximately linear, it is linearized into a linear function of temperature using the least squares method. After multiplying it to the right side of the inequality, the temperature and Multiplying the two bilinear variables and relaxing them again using the McCormick envelope gives:

[0169] .

[0170] A and B are the coefficients obtained after least squares linearization, A=0.3068, B=-2.9284.

[0171] S200, calculate the total energy consumption of each area of ​​the data center based on the power, computing power and heat consumption model and establish the objective function with the minimum cost of data center operation in combination with the real-time electricity price:

[0172] .

[0173] in, For real-time electricity prices, and are the time interval and total time respectively, is the total energy consumption of the data center.

[0174] If there are wind and solar power generation devices in the data center, the objective function is:

[0175] .

[0176] in, and They are respectively the wind power energy and photovoltaic energy obtained by the data center at time t.

[0177] S300: Optimize and schedule the data center task load according to the objective function.

[0178] A data center task load scheduling system with electricity, computing power and heat coordination includes a model building module 100, an objective function establishment module 200 and an optimization scheduling module 300.

[0179] The model building module 100 is used to model the power, computing power and heat consumption involved in the data center according to the data center cooling system to obtain a power, computing power and heat consumption model.

[0180] The objective function establishment module 200 is used to calculate the total energy consumption of the data center according to the power, computing power and heat consumption model and to establish the objective function with the minimum cost of the data center operation in combination with the real-time electricity price.

[0181] The optimization scheduling module 300 is used to optimize the data center task load according to the objective function.

[0182] Example 2

[0183] Example 1: The method of the present invention.

[0184] Example 2: Ignoring the temporal and spatial scheduling of task loads, a balanced allocation method is used.

[0185] Example 3: The contribution of the variable air volume system is not considered, and only the thermal inertia and task load spatiotemporal regulation are considered.

[0186] Among them, the relevant parameters of the delay tolerance load are shown in Table 1.

[0187] surface Delay Tolerance Load Parameters

[0188]

[0189] On the same day, under the same external environment, real-time electricity price, server nature, and task load arrival, the data center has three different areas and two types of delay-sensitive loads. The total energy consumption and total cost are shown in Table 2. By comparison, it can be seen that the optimization effect of the method of the present invention is obvious.

[0190] Table 2 Comparison of three examples

[0191]

[0192] And, in Example 1, by Figure 3 and Figure 4 It can be seen that when the electricity price is high, the task load assigned to each area is small and the fan output is low. The electricity price starts to rise at time period 68, but the room temperature in each area almost starts to decrease from time period 64, which verifies the flexibility brought by the time lag effect of temperature to load scheduling. Before the electricity price increases, cooling is increased to reduce the temperature to a lower level, and cooling is reduced when the electricity price is high. Since no task load is assigned while increasing cooling, the temperature continues to decrease, but area 3 has a temperature rise point at time 67. This is because in order to complete the delay-sensitive load DT2 before the deadline, area 3 bears part of the load at time 66, and the temperature begins to rise. The chiller and fan increase their output. At time 67, the temperature rises to a high point and there is no load to be executed, and the temperature begins to continue to decrease (where RT is a delay-sensitive load).

[0193] Depend on Figure 5It can be seen that electricity price is the main factor affecting batch load allocation. During the period with the highest electricity price (from time 69 to time 77), almost no delay-tolerant task load is processed. Only real-time loads that cannot be flexibly transferred are allocated to areas 1 and 2, reducing energy consumption while ensuring service quality.

[0194] Example 2: Balanced distribution of task load is the most common task load processing method adopted by most data centers nowadays. That is, the time and space flexibility of task load is not considered. As long as the load arrives, the load is distributed according to the ratio of computing resources in the area. Case II adopts the balanced distribution method and combines the output of the variable air volume system to keep the temperature within the limit and ensure the safe and stable operation of the server. The operation results are as follows: Figure 6 and Figure 7 shown.

[0195] Depend on Figure 6 It can be seen that in the several periods before each load arrives, the regional temperature will first drop to a lower level, especially during periods with lower electricity prices, such as period 64. After the load arrives, the temperature of the server output room rises and the fan increases its output to take away the heat from the server. However, when the electricity price is high, it is impossible to reduce energy consumption and achieve cost reduction. For example, part of the delay-tolerant load DT2 is scheduled during the period with the highest electricity price. Even though the cooling has been increased since period 64 to reduce the room temperature, the subsequent load brings huge heat, the fan output is almost full load, and the room temperature is near the maximum limit, which poses a challenge to the safe and stable operation of the server.

[0196] Compared with Example 1, in Example 2, the load allocated to Area 3 increases significantly, but there are fewer air conditioners in Area 3. When the room temperature is maximum, the cooling resources provided can only be achieved by increasing the fan output, such as from time 10 to time 31, so the energy consumption increases.

[0197] Example 3 is to verify the role of the variable air volume system in task load time-space regulation. Generally, fans use a constant air volume system. As part of the data center cooling system, the fan's energy consumption and temperature have a strong correlation. Figure 8 and Fig. 9 It is the spatiotemporal calculation result of regional room temperature and task load under rated air supply mode, and its cost is the highest among the three methods.

[0198] Fig.10The energy consumption of three examples at different times is compared. All three are negatively correlated with the fluctuation of electricity prices. The energy consumption of the constant air volume system in Example 3 is the highest, and the lowest is the method proposed in the present invention. However, the energy consumption of Example 1 is not lower than that of Example 2 at every moment. For example, from time 16 to time 30, this is because the load rate of the data center in Example 1 is higher during this period, so less load needs to be processed from time 33 to time 41 when the electricity price is higher. In general, the total energy consumption and total cost of Example 1 are at the lowest level.

[0199] In summary, the method proposed in the present invention can provide a specific time-space scheduling solution for task loads and can save up to 29.69% of the cost.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described with reference to the preferred embodiments of the present invention, it should be understood by those skilled in the art that various changes may be made in form and details without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A data center task load optimization scheduling method based on computer-heat collaboration, characterized in that: include: According to the data center cooling system, the power, computing power and heat consumption involved in each area of ​​the data center are modeled to obtain the power, computing power and heat consumption model; Calculate the total energy consumption of each area of ​​the data center according to the power, computing power and heat consumption model and establish an objective function with the minimum cost of data center operation in combination with the real-time electricity price; According to the objective function, the data center task load is optimized and scheduled; The thermal consumption model includes an indoor temperature model, a fan model, a chiller model and a cooling system model; The indoor temperature model is formed by calculating the indoor temperature using the first thermodynamic formula based on the heat capacity of the hot aisle in the area, the cooling provided by the chiller and fan in the area, the heat generated by the servers in the area, the heat transfer of the exterior wall in the area, the heat generated by the solar radiation on the opaque surface of the exterior wall in the area, and the heat generated by the lighting in the area, and calculating the supply air temperature using the first thermodynamic formula based on the heat capacity of the cold aisle in the area, the cooling provided by the chiller and fan; The fan model is formed based on a quadratic function of the fan energy consumption and its air supply volume; The chiller model is formed based on the power consumption of the chiller in a certain area to obtain the total energy consumption of the chiller; The cooling system model is formed based on the total energy consumption of the chiller, the total energy consumption of the fan and the constant energy consumption to obtain the total energy consumption of the entire cooling system of the data center; The objective function also includes wind power energy and photovoltaic energy obtained by the data center; the objective function is: Among them, π t is the real-time electricity price, Δt and T are the time interval and total time respectively, P wind，t and P solar，t are the wind power energy and photovoltaic energy obtained by the data center at time t, P total，t is the total energy consumption of the data center; Calculate the power consumption of the data center based on the energy consumption of all IT equipment in the data center, the total energy consumption of the cooling system, and the energy consumption of lighting and uninterruptible power supply UPS, and build a power consumption model; According to the task load of the data center, the task load is divided into delay-sensitive load and delay-tolerant load. According to the arrival amount of delay-sensitive load in a certain area at a certain moment, the category of delay-tolerant load, the number of demands for each category of delay-tolerant load, arrival time and deadline, the load amount that needs to be processed in a certain area at a certain moment is calculated. According to the load amount that needs to be processed in a certain area at a certain moment, the energy consumption of all IT equipment in the computer room in a certain area at a certain moment is calculated to form a computing power consumption model.

2. The data center task load optimization scheduling method of computer-heat collaboration according to claim 1 is characterized in that: The power consumption model: in, is the energy consumed by the server in region i at time t, P c，t is the total energy consumption of the cooling system at time t, P other is the energy consumption of lighting and uninterruptible power supply (UPS) at time t.

3. The data center task load optimization scheduling method of computer-heat collaboration according to claim 1 is characterized in that: The indoor temperature model is: in, is the indoor temperature of area i at time t, is the indoor temperature of area i at time t+1, is the supply air temperature of area i at time t, is the air supply temperature of area i at time t+1, C i and are the heat capacity of the hot channel and the cold channel in region i, respectively. is the energy consumption of the cooling system in area i at time t, The heat generated by the server in region i at time t, is the heat transfer of the exterior wall in region i at time t, is the heat generated by solar radiation on the opaque surface of the exterior wall in area i at time t, The heat generated by lighting in area i at time t; The energy consumption of the cooling system in the i region at time t Among them, K i is the number of fans in area i, m i，t is the wind speed of the fan in area i at time t, c a is the amount of air substance; The server in region i generates heat at time t The heat transfer of the exterior wall in area i at time t in, is the heat transfer coefficient of the exterior wall in area i, is the external wall area of ​​region i, T t o is the outdoor temperature at time t; The heat generated by solar radiation on the opaque surface of the exterior wall in area i at time t Among them, S o，t is the thermal radiation intensity of the sun in direction o at time t, is the solar radiation area of ​​the exterior wall of region i in direction o, is the heat transfer coefficient of region i; The heat generated by lighting in area i at time t Among them, Q light is the unit lighting density, is the lighting area of ​​region i; The fan model is: The chiller model is: Among them, P chiller，t is the total energy consumption of the chiller at time t, is the full load capacity of the chiller, is the power consumption of the chiller in area i at time t; The power consumption of the chiller in area i at time t Where η is the heat transfer coefficient of the chiller coil and COP is the coefficient of performance of the chiller: Among them, a0, a1, and a2 are the performance-related coefficients of the chiller; The cooling system model is: Among them, P c，t is the energy consumption of the entire cooling system of the data center at time t, P fan，t is the total energy consumption of the fan at time t, The constant energy consumption includes the energy consumption of circulating water pump and cooling tower fan.

4. The data center task load optimization scheduling method of computer-heat collaboration according to claim 1 is characterized in that: The computing power consumption model is: Among them, M i is the number of servers in region i, P i idle is the rated idle power consumption of a server in region i, P i peak is the rated full-load power consumption of a server in region i, w i，t is the load that needs to be processed in region i at time t, is the total computing capacity of region i; The load w that the i region needs to handle at time t i，t : Among them, μ i，t is the amount of delay-sensitive load allocated to region i at time t, and is the amount of delay-tolerant load of type j scheduled to region i at time t.

5. The data center task load optimization scheduling method of computer-heat collaboration according to claim 4 is characterized in that: The energy consumption of the cooling system in area i at time t and the power consumption of the chiller in area i at time t After linearization, a mixed integer quadratic programming problem is formed, and the Gurobi solver is used to solve the mixed integer quadratic programming problem.

6. A data center task load optimization scheduling system with computer-heat collaboration, characterized in that: It includes model building module, objective function building module and optimization scheduling module; The model building module is used to model the power, computing power and heat consumption involved in the data center according to the data center cooling system to obtain the power, computing power and heat consumption model; The objective function establishment module is used to calculate the total energy consumption of the data center according to the power, computing power and heat consumption model and establish the objective function with the minimum cost of the data center operation in combination with the real-time electricity price; The optimization scheduling module is used to optimize the data center task load according to the objective function; The thermal consumption model includes an indoor temperature model, a fan model, a chiller model and a cooling system model; The indoor temperature model is formed by calculating the indoor temperature using the first thermodynamic formula based on the heat capacity of the hot aisle in the area, the cooling provided by the chiller and fan in the area, the heat generated by the servers in the area, the heat transfer of the exterior wall in the area, the heat generated by the solar radiation on the opaque surface of the exterior wall in the area, and the heat generated by the lighting in the area, and calculating the supply air temperature using the first thermodynamic formula based on the heat capacity of the cold aisle in the area, the cooling provided by the chiller and fan; The fan model is formed based on a quadratic function of the fan energy consumption and its air supply volume; The chiller model is formed based on the power consumption of the chiller in a certain area to obtain the total energy consumption of the chiller; The cooling system model is formed based on the total energy consumption of the chiller, the total energy consumption of the fan and the constant energy consumption to obtain the total energy consumption of the entire cooling system of the data center; The objective function also includes wind power energy and photovoltaic energy obtained by the data center; the objective function is: Among them, π t is the real-time electricity price, Δt and T are the time interval and total time respectively, P wind，t and P solar，t are the wind power energy and photovoltaic energy obtained by the data center at time t, P total，t is the total energy consumption of the data center; Calculate the power consumption of the data center based on the energy consumption of all IT equipment in the data center, the total energy consumption of the cooling system, and the energy consumption of lighting and uninterruptible power supply UPS, and build a power consumption model; According to the task load of the data center, the task load is divided into delay-sensitive load and delay-tolerant load. According to the arrival amount of delay-sensitive load in a certain area at a certain moment, the category of delay-tolerant load, the number of demands for each category of delay-tolerant load, arrival time and deadline, the load amount that needs to be processed in a certain area at a certain moment is calculated. According to the load amount that needs to be processed in a certain area at a certain moment, the energy consumption of all IT equipment in the computer room in a certain area at a certain moment is calculated to form a computing power consumption model.

Images