CN-115933856-B - Method for saving power consumption by cross-node computing power sharing

CN115933856BCN 115933856 BCN115933856 BCN 115933856BCN-115933856-B

Abstract

The invention provides a method for saving power consumption by cross-node computing force sharing, which comprises the following steps that in a normal working state of a slave computer room node, the slave computer room node periodically collects the current network delay d_i and the used task computing force and reports the current network delay d_i and the used task computing force to a host computer room node; and the host computer room node calculates the task calculation power score of each slave computer room node in real time, and distributes tasks according to the task calculation power score of the slave computer room node. The invention provides a method for saving power consumption by cross-node computing power sharing, which is a method for saving power consumption by cross-node computing power sharing, and takes computing capacity, a calibrated PUE value and a network environment of computer room nodes as reference factors, and reasonably distributes computing tasks to each computer room node, so that the staged PUE value of each computer room node is in a reasonable interval, namely, the power consumption of a server of each computer room node and the power consumption of other equipment reach the optimal ratio, the power consumption efficiency is improved, the power consumption is saved as a whole, and the energy conservation and emission reduction are realized.

Inventors

CHEN WENBIN
QU HONGGUI
FENG CHAO
QIU FEI

Assignees

北京中电兴发科技有限公司

Dates

Publication Date: 20260512
Application Date: 20221215

Claims (4)

1. A method for saving power consumption for cross-node computing power sharing, comprising the steps of: step 1, determining one machine room node as a master_node of all machine room nodes, and other machine room nodes as slave machine room nodes, wherein the total number of the slave machine room nodes is assumed to be n, and the slave machine room nodes are respectively denoted as slave machine room nodes slave_node_1, slave_node_2, and slave_node_n; Step 2, for any slave machine room node slave_node_i, wherein i=1, 2,..n, initially, reporting basic configuration parameters to a master_node of the host machine room, including a PUE reference value and a task calculation total force value t_total_i; Step 3, in the normal working state of the slave computer room node slave_node_i, the slave computer room node slave_node_i periodically acquires the current network delay d_i and the used task calculation force Th_i and reports the current network delay d_i and the used task calculation force Th_i to the master_node of the host computer room node; Step 4, when the master_node of the host room receives a task to be distributed, obtaining a task pre-estimation force Td of the task and a task expected calculation force Tp; then, the task computing power score f_i of each slave computer room node slave_node_i is obtained through real-time computation by adopting the following formula: F_i=(1.4-PUE_i)*40+t_i/3*50+(100-d_i)*0.1 Wherein: t_i is a task force factor, and is calculated by the following method: 1) The preliminary task calculation force multiple difference tc_i from the machine room node slave_node_i is calculated using: Tc_i=(T_i-Th_i-Td)/Td Wherein: T_i represents the total calculation force of the machine room from the machine room node slave_node_i, and is a known fixed value; 2) Comparing the calculated force multiple difference Tc_i of the preparation task with the expected calculated force Tp of the task, if the calculated force multiple difference Tc_i of the preparation task is larger than the expected calculated force Tp of the task, taking the calculated force factor t_i of the task as 3, otherwise, taking the calculated force factor t_i of the task as the calculated force multiple difference Tc_i of the preparation task; And 5, the master_node of the host room sends the task to be allocated to the slave computer room node slave_node with the highest task computing power score, and the slave computer room node slave_node with the highest task computing power score executes the task.
2. The method for saving power consumption in cross-node power sharing according to claim 1, wherein in step 3, the following method is adopted from the machine room node slave_node_i to periodically acquire the current network delay d_i: The slave computer room node slave_node_i transmits network delay detection data to other n-1 slave computer room nodes slave_node respectively to obtain network delays of the other n-1 slave computer room nodes slave_node respectively, so that n-1 network delays are obtained in total, and statistical calculation is performed on the n-1 network delays to obtain final network delay d_i.
3. The method of cross-node computing power sharing power consumption savings of claim 1, wherein step 5 further comprises: If there are a plurality of slave computer room nodes slave_nodes with the highest task computing power scores, selecting the slave computer room node slave_node with the smallest network delay from the slave computer room nodes slave_node with the highest task computing power scores as the slave computer room node slave_node for executing the task.
4. The method of power consumption conservation for cross-node computing power sharing of claim 1, further comprising, after step 5: and 6, after the selected slave computer room node slave_node executes the task, returning to the step 3, collecting the current network delay and the used task calculation force, and reporting to the master_node.

Description

Method for saving power consumption by cross-node computing power sharing Technical Field The invention belongs to the technical field of machine room energy management, and particularly relates to a method for saving power consumption by cross-node computing power sharing. Background PUE (Power Usage Effectiveness) is an index for evaluating the power use efficiency of the machine room, and reflects the percentage of the power supply of the machine room, which is really used for calculating the server. Since the PUE value at each time has a fluctuation characteristic with the lapse of time (environmental change, task amount change), the PUE value calculated by the conventional method is an average value of statistics for a period of time. At present, the PUE reference value externally calibrated by a machine room constructor is an optimal value under certain specific scenes, for example, when the power consumption of the IT equipment and the power consumption of other equipment tend to be balanced under certain task quantity. When the difference between the instantaneous PUE value of the machine room and the calibrated PUE reference value is large, for example, greater than 20%, the power utilization rate of the machine room is reduced, and compared with the expected power (estimated result based on the calibrated PUE reference value), additional power consumption is generated, and waste is generated intangibly. At present, the instantaneous PUE value of the machine room cannot be kept in a more reasonable state for a long time, namely, the instantaneous PUE value floats about 20% up and down at the optimal value, so that the electricity consumption of the machine room is often higher than expected, and the operation and maintenance cost of the machine room is greatly influenced. Therefore, how to make the staged PUE value of the machine room in a reasonable interval, make the power consumption of the server and the power consumption of other devices reach the optimal ratio, improve the power consumption efficiency, and generally save the power consumption is the key point and difficulty to be solved at present. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides a method for saving power consumption by cross-node computing force sharing, which can effectively solve the problems. The technical scheme adopted by the invention is as follows: The invention provides a method for saving power consumption by cross-node computing force sharing, which comprises the following steps: step 1, determining one machine room node as a master_node of all machine room nodes, and other machine room nodes as slave machine room nodes, wherein the total number of the slave machine room nodes is assumed to be n, and the slave machine room nodes are respectively denoted as slave machine room nodes slave_node_1, slave_node_2, and slave_node_n; Step 2, for any slave machine room node slave_node_i, wherein i=1, 2,..n, initially, reporting basic configuration parameters to a master_node of the host machine room, including a PUE reference value and a task calculation total force value t_total_i; Step 3, in the normal working state of the slave computer room node slave_node_i, the slave computer room node slave_node_i periodically acquires the current network delay d_i and the used task calculation force Th_i and reports the current network delay d_i and the used task calculation force Th_i to the master_node of the host computer room node; Step 4, when the master_node of the host room receives a task to be distributed, obtaining a task pre-estimation force Td of the task and a task expected calculation force Tp; then, the task computing power score f_i of each slave computer room node slave_node_i is obtained through real-time computation by adopting the following formula: F_i=(1.4-PUE_i)*40+t_i/3*50+(100-d_i)*0.1 Wherein: t_i is a task force factor, and is calculated by the following method: 1) The preliminary task calculation force multiple difference tc_i from the machine room node slave_node_i is calculated using: Tc_i=(T_i-Th_i-Td)/Td Wherein: T_i represents the total calculation force of the machine room from the machine room node slave_node_i, and is a known fixed value; 2) Comparing the calculated force multiple difference Tc_i of the preparation task with the expected calculated force Tp of the task, if the calculated force multiple difference Tc_i of the preparation task is larger than the expected calculated force Tp of the task, taking the calculated force factor t_i of the task as 3, otherwise, taking the calculated force factor t_i of the task as the calculated force multiple difference Tc_i of the preparation task; And 5, the master_node of the host room sends the task to be allocated to the slave computer room node slave_node with the highest task computing power score, and the slave computer room node slave_node with the highest task computing power score executes the task.