A method for saving power consumption by sharing computing power across nodes

Abstract

The application provides a method for saving power consumption through cross-node computing power sharing, comprising the following steps: collecting current network delay d_i and used task computing power of a slave room node in a normal working state of the slave room node, and reporting to a master room node; the master room node calculates the task computing power score of each slave room node in real time, and allocates tasks according to the task computing power score of the slave room node. The application provides a method for saving power consumption through cross-node computing power sharing, which is a method for saving power consumption through cross-node computing power sharing. The computing capacity of the room node, the calibrated PUE value and the network environment are used as reference factors, the computing task is reasonably allocated to each room node, the stage PUE value of each room node is in a reasonable range, that is, the power consumption of the server and other devices of each room node is optimized, the power utilization efficiency is improved, the total power consumption is saved, and energy saving and emission reduction are realized.

Description

Technical Field

[0001] This invention belongs to the field of data center energy management technology, specifically relating to a method for saving power consumption through cross-node computing power sharing. Background Technology

[0002] PUE (Power Usage Effectiveness) is a metric for evaluating the power usage efficiency of a data center, reflecting the percentage of power actually used for server computing. Because the PUE value fluctuates over time (due to changes in environment and workload), traditional methods calculate a PUE value as an average over a period of time. Currently, the PUE reference value publicly provided by data center developers is the optimal value for certain specific scenarios, such as the optimal value when the power consumption of IT equipment and other equipment tends to balance under a certain workload. However, when the instantaneous PUE value of the data center differs significantly from the specified PUE reference value, for example, greater than 20%, the data center's power utilization rate decreases. Compared to the expected power consumption (an estimate based on the specified PUE reference value), this results in additional power loss, leading to waste.

[0003] Currently, the instantaneous PUE value of the data center cannot be maintained at a reasonable level for a long time, that is, fluctuating around 20% above or below the optimal value. This causes the data center's power consumption to often exceed expectations, which greatly affects the data center's operation and maintenance costs.

[0004] Therefore, how to keep the periodic PUE value of the data center within a reasonable range, optimize the power consumption ratio of servers and other devices, improve power efficiency, and save power consumption overall is the key and difficult issue that needs to be addressed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for saving power consumption through cross-node computing power sharing, which can effectively solve the above-mentioned problems.

[0006] The technical solution adopted in this invention is as follows:

[0007] This invention provides a method for saving power consumption through cross-node computing power sharing, comprising the following steps:

[0008] Step 1: Among all data center nodes, determine one data center node as the master data center node (master_node), and the other data center nodes as slave data center nodes. Assume there are n slave data center nodes, represented as slave_node_1, slave_node_2, ..., slave_node_n.

[0009] Step 2: For any slave node_i, where i = 1, 2, ..., n, initially, report basic configuration parameters to the master node, including: PUE reference value and total task computing power T_total_i; where the PUE reference value of slave node_i is represented as: PUE_i;

[0010] Step 3: When the slave node_i is in normal working condition, the slave node_i periodically collects the current network latency d_i and the used task computing power Th_i, and reports it to the master node_node.

[0011] Step 4: When the master_node in the main data center receives the task to be assigned, it obtains the estimated computing power Td and the expected computing power Tp of the task.

[0012] Then, the following formula is used to calculate the task computing power score F_i for each slave node in the data center in real time:

[0013] F_i=(1.4-PUE_i)*40+t_i / 3*50+(100-d_i)*0.1

[0014] in:

[0015] t_i is the task computing power factor, calculated using the following method:

[0016] 1) The difference in computing power multiple Tc_i for the preparatory task of slave_node_i in the data center is calculated using the following formula:

[0017] Tc_i=(T_i-Th_i-Td) / Td

[0018] in:

[0019] T_i represents the total computing power of the data center from slave_node_i, which is a known fixed value;

[0020] 2) Compare the difference in computing power multiple of the preliminary task Tc_i with the expected computing power of the task Tp. If the difference in computing power multiple of the preliminary task Tc_i is greater than 3 times the expected computing power of the task Tp, then the computing power factor t_i of the task is set to 3; otherwise, the computing power factor t_i of the task is set to the difference in computing power multiple of the preliminary task Tc_i.

[0021] Step 5: The master_node sends the task to be assigned to the slave_node with the highest task computing power score, and the slave_node with the highest task computing power score executes the task.

[0022] Preferably, in step 3, the slave node i in the data center uses the following method to periodically collect the current network latency d_i:

[0023] Network latency probe data is sent from slave_node_i to each of the other n-1 slave_nodes, and the network latency with each of the other n-1 slave_nodes is obtained. Therefore, a total of n-1 network latencies are obtained. Statistical calculations are performed on the n-1 network latencies to obtain the final network latency d_i.

[0024] Preferably, step 5 further includes:

[0025] If there are multiple slave nodes with the highest task computing power score, then the slave node with the lowest network latency among the slave nodes with the highest task computing power score will be selected as the slave node to execute the task.

[0026] Preferably, after step 5, the following steps are also included:

[0027] Step 6: After the selected slave node completes the task, return to step 3, collect the current network latency and the computing power used for the task, and report it to the master node.

[0028] The method for saving power consumption through cross-node computing power sharing provided by this invention has the following advantages:

[0029] This invention provides a method for saving power consumption through cross-node computing power sharing. This method uses the computing power, calibrated PUE value, and network environment of data center nodes as reference factors to rationally allocate computing tasks to each data center node, ensuring that the periodic PUE value of each data center node is within a reasonable range. In other words, it optimizes the power consumption ratio between servers and other devices on each data center node, improving power efficiency and saving overall power consumption, thus achieving energy conservation and emission reduction. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating a method for saving power consumption through cross-node computing power sharing provided by the present invention. Detailed Implementation

[0031] To make the technical problems solved, the technical solutions, and the beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the invention.

[0032] This invention provides a method for saving power consumption through cross-node computing power sharing. This method uses the computing power, calibrated PUE value, and network environment of data center nodes as reference factors to rationally allocate computing tasks to each data center node, ensuring that the periodic PUE value of each data center node is within a reasonable range. In other words, it optimizes the power consumption ratio between servers and other devices on each data center node, improving power efficiency and saving overall power consumption, thus achieving energy conservation and emission reduction.

[0033] refer to Figure 1 The present invention provides a method for saving power consumption through cross-node computing power sharing, comprising the following steps:

[0034] Step 1: Among all data center nodes, determine one data center node as the master data center node (master_node), and the other data center nodes as slave data center nodes. Assume there are n slave data center nodes, represented as slave_node_1, slave_node_2, ..., slave_node_n.

[0035] Step 2: For any slave node in the data center, where i = 1, 2, ..., n, initially, it reports basic configuration parameters to the master node in the data center, including: the PUE reference value and the total computing power of the task, T_total_i; where the PUE reference value of the slave node is represented as: PUE_i; the typical PUE reference value is around 1.2-2.4.

[0036] Step 3: When the slave node_i is in normal working condition, the slave node_i periodically collects the current network latency d_i and the used task computing power Th_i, and reports it to the master node_node.

[0037] In step 3, the slave node in the data center uses the following method to periodically collect the current network latency d_i: network latency d_i represents the network environment data of the slave node in the data center, in milliseconds.

[0038] Network latency probe data is sent from slave_node_i to each of the other n-1 slave_nodes, and the network latency with each of these slave_nodes is obtained. Therefore, a total of n-1 network latency data are obtained. Statistical calculations are then performed on these n-1 network latency data to obtain the final network latency d_i. The statistical calculation can be performed by averaging the n-1 network latency data.

[0039] Step 4: When the master_node in the main data center receives the task to be assigned, it obtains the estimated computing power Td and the expected computing power Tp of the task.

[0040] Then, the following formula is used to calculate the task computing power score F_i for each slave node in the data center in real time:

[0041] F_i=(1.4-PUE_i)*40+t_i / 3*50+(100-d_i)*0.1

[0042] in:

[0043] t_i is the task computing power factor, calculated using the following method:

[0044] 1) The difference in computing power multiple Tc_i for the preparatory task of slave_node_i in the data center is calculated using the following formula:

[0045] Tc_i=(T_i-Th_i-Td) / Td

[0046] in:

[0047] T_i represents the total computing power of slave_node_i in the data center, which is a known fixed value; a typical total computing power of a data center is 100 TPS, that is, 100 target analyses are processed per second. The total computing power of each slave node varies depending on the data center configuration.

[0048] 2) Compare the difference in computing power multiple of the preliminary task Tc_i with the expected computing power of the task Tp. If the difference in computing power multiple of the preliminary task Tc_i is greater than 3 times the expected computing power of the task Tp, then the computing power factor t_i of the task is set to 3; otherwise, the computing power factor t_i of the task is set to the difference in computing power multiple of the preliminary task Tc_i.

[0049] Step 5: The master_node sends the task to be assigned to the slave_node with the highest task computing power score, and the slave_node with the highest task computing power score executes the task.

[0050] Step 5 also includes:

[0051] If there are multiple slave nodes with the highest task computing power score, then the slave node with the lowest network latency among the slave nodes with the highest task computing power score will be selected as the slave node to execute the task.

[0052] After step 5, the following also includes:

[0053] Step 6: After the selected slave node completes the task, return to step 3, collect the current network latency and the computing power used for the task, and report it to the master node.

[0054] As a specific implementation, this invention provides a method for saving power consumption through cross-node computing power sharing, applied in the field of data center energy management, specifically in scenarios where multiple data center nodes are managed simultaneously. A computing power sharing management program is installed on the master node, and client programs are installed on the other slave data center nodes, forming a computing power sharing system. The computing power sharing management program allocates computing tasks based on the real-time computing power, PUE value, and network environment (i.e., network latency) of each slave data center node, ensuring that each slave node maintains a reasonable PUE value for an extended period, thereby saving power consumption overall. Specifically, the computing power sharing management program is primarily responsible for collecting data, coordinating tasks, and allocating them to the slave data center nodes; the client programs mainly collect basic data periodically and report it to the computing power sharing management program.

[0055] Therefore, in the management of data center clusters, the method of this invention can be used to allocate computing tasks to appropriate slave nodes. While making reasonable use of the network and computing power of idle slave nodes, it can also keep the instantaneous PUE value of each slave node within the optimal range for a long time, thereby improving power efficiency, reducing power waste as a whole, and achieving the goal of energy conservation and emission reduction.

[0056] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for saving power consumption through cross-node computing power sharing, characterized in that, Includes the following steps: Step 1: Among all data center nodes, determine one data center node as the master data center node (master_node), and the other data center nodes as slave data center nodes. Assume there are n slave data center nodes, represented as slave_node_1, slave_node_2, ..., slave_node_n. Step 2: For any slave node_i, where i = 1, 2, ..., n, initially, report basic configuration parameters to the master node, including: PUE reference value and total task computing power T_total_i; where the PUE reference value of slave node_i is represented as: PUE_i; Step 3: When the slave node_i is in normal working condition, the slave node_i periodically collects the current network latency d_i and the used task computing power Th_i, and reports it to the master node_node. Step 4: When the master_node in the main data center receives the task to be assigned, it obtains the estimated computing power Td and the expected computing power Tp of the task. Then, the following formula is used to calculate the task computing power score F_i for each slave node in the data center in real time: F_i=(1.4-PUE_i)*40+t_i / 3*50+(100-d_i)*0.1 in: t_i is the task computing power factor, calculated using the following method: 1) The difference in computing power multiple Tc_i for the preparatory task of slave_node_i in the data center is calculated using the following formula: Tc_i=(T_i-Th_i-Td) / Td in: T_i represents the total computing power of the data center from slave_node_i, which is a known fixed value; 2) Compare the difference in computing power multiple of the preliminary task Tc_i with the expected computing power of the task Tp. If the difference in computing power multiple of the preliminary task Tc_i is greater than 3 times the expected computing power of the task Tp, then the computing power factor t_i of the task is set to 3; otherwise, the computing power factor t_i of the task is set to the difference in computing power multiple of the preliminary task Tc_i. Step 5: The master_node sends the task to be assigned to the slave_node with the highest task computing power score, and the slave_node with the highest task computing power score executes the task.

2. The method for saving power consumption through cross-node computing power sharing according to claim 1, characterized in that, In step 3, the slave node in the data center uses the following method to periodically collect the current network latency d_i: Network latency probe data is sent from slave_node_i to each of the other n-1 slave_nodes, and the network latency with each of the other n-1 slave_nodes is obtained. Therefore, a total of n-1 network latencies are obtained. Statistical calculations are performed on the n-1 network latencies to obtain the final network latency d_i.

3. The method for saving power consumption through cross-node computing power sharing according to claim 1, characterized in that, Step 5 also includes: If there are multiple slave nodes with the highest task computing power score, then the slave node with the lowest network latency among the slave nodes with the highest task computing power score will be selected as the slave node to execute the task.

4. The method for saving power consumption through cross-node computing power sharing according to claim 1, characterized in that, After step 5, the following also includes: Step 6: After the selected slave node completes the task, return to step 3, collect the current network latency and the computing power used for the task, and report it to the master node.

Images