CN115658282A - Server computing power management distribution method, system, network device and storage medium - Google Patents
Server computing power management distribution method, system, network device and storage medium Download PDFInfo
- Publication number
- CN115658282A CN115658282A CN202210991170.5A CN202210991170A CN115658282A CN 115658282 A CN115658282 A CN 115658282A CN 202210991170 A CN202210991170 A CN 202210991170A CN 115658282 A CN115658282 A CN 115658282A
- Authority
- CN
- China
- Prior art keywords
- task
- computing
- target
- calculation
- computing power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The application provides a server computing power management and distribution method, a server computing power management and distribution system, network equipment and a storage medium, and belongs to the technical field of computers. The server computing power management and distribution method identifies a target computing task by acquiring the target computing task; extracting relevant parameters of the target calculation task according to the identification result of the target calculation task; acquiring current computing power resource information; distributing computing power resources for the target computing task according to the current computing power resource information and the relevant parameters of the target computing task; the method comprises the steps of identifying a target computing task to obtain relevant parameters of the target computing task, and matching corresponding computing resources for the target computing task at least according to the type, size and timeliness requirements of the target computing task, so that the reasonability of computing management and distribution is greatly improved, computing cost and hardware cost waste are reduced, and more appropriate computing resources are matched; the method can improve the completion efficiency of the target calculation task and improve the satisfaction degree of the user.
Description
Technical Field
The present application relates to the field of computer technologies, and in particular, to a server computing power management and distribution method, system, network device, and storage medium.
Background
Nowadays, with the rapid development of computer data, computer, artificial intelligence and big data service in various fields such as industry, agriculture, medicine, national defense, economy, education and the like have penetrated various aspects of industry and life, and the computing requirements of various industries are continuously improved.
Many enterprises deploy local servers to meet computing needs, or rent or purchase cloud servers, or both. Regardless of the deployment mode of any server, the management and allocation of computing power are very important, and most of users adopt the mode of simply stacking the computing power to configure more servers with stronger computing power to meet the improvement of computing power requirements at present; a more reasonable and flexible dynamic allocation scheme is lacked in the aspect of computing power management allocation, various computing tasks are difficult to adapt, and the unreasonable allocation of the computing tasks causes the waste of computing power cost and server hardware cost.
Disclosure of Invention
The application provides a server power management distribution method, a server power management distribution system, network equipment and a storage medium, and aims to solve the problems that power management distribution is unreasonable, power cost is wasted, hardware cost is wasted and the like in the prior art.
In order to solve the technical problem, the present application is implemented as follows:
in a first aspect, an embodiment of the present application provides a server computing power management allocation method, including:
acquiring a target calculation task, and identifying the target calculation task;
extracting relevant parameters of the target calculation task according to the identification result of the target calculation task; wherein the relevant parameters of the target computing task at least comprise: the type of the target calculation task, the size of the target calculation task and the timeliness requirement of the target calculation task;
acquiring current computing power resource information; wherein the computing resource information at least comprises: calculating the type, storage capacity and calculation speed of the resource;
and distributing computing power resources for the target computing task according to the current computing power resource information and the relevant parameters of the target computing task.
In a second aspect, an embodiment of the present application provides a server computing power management distribution system, including:
the task identification module is used for acquiring a target calculation task and identifying the target calculation task;
the task parameter extraction module is used for extracting relevant parameters of the target calculation task according to the identification result of the target calculation task; wherein the relevant parameters of the target computing task at least comprise: the type of the target calculation task, the size of the target calculation task and the timeliness requirement of the target calculation task;
the computing resource monitoring module is used for acquiring current computing resource information; wherein the computing power resource information at least comprises: calculating the type, storage capacity and calculation speed of the resource;
and the computing power resource allocation module is used for allocating computing power resources for the target computing task according to the current computing power resource information and the relevant parameters of the target computing task.
In a third aspect, an embodiment of the present application provides a network device, including: a processor, a memory and a program stored on the memory and executable on the processor, the program, when executed by the processor, implementing the steps of the server power management allocation method of the first aspect described above.
In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the server power management allocation method of the first aspect.
The beneficial effects brought by the technical scheme provided by the embodiment of the application at least comprise: identifying a target computing task by acquiring the target computing task; extracting relevant parameters of the target calculation task according to the identification result of the target calculation task; acquiring current computing power resource information; distributing computing power resources for the target computing task according to the current computing power resource information and the relevant parameters of the target computing task; the target computing task is identified to obtain the relevant parameters of the target computing task, and the corresponding computing resources are matched at least according to the type, size and timeliness requirements of the target computing task, so that the rationality of computing management and distribution is greatly improved, the computing cost and hardware cost waste are reduced, the completion efficiency of the target computing task can be improved by matching the more appropriate computing resources, and the satisfaction degree of a user is improved.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the application. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:
fig. 1 is a flowchart of a server computing power management allocation method according to an embodiment of the present disclosure;
fig. 2 is a schematic diagram of a calculation power fitness score list provided in an embodiment of the present application;
FIG. 3 is a schematic diagram illustrating a computing power distribution flow of a target computing task according to an embodiment of the present disclosure;
fig. 4 is a schematic structural diagram of a server computing power management distribution system according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a network device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The server computing power management and distribution method provided by the embodiment of the application can be deployed on a server or other related equipment for managing the server, can be deployed on local equipment or cloud equipment, and can be used for computing power management and distribution of various single or multiple servers or server groups capable of providing computing power, such as a single server, a multi-server unit, combination of the local server and the cloud server group, and the like.
Referring to fig. 1, fig. 1 is a flowchart illustrating a server computing power management allocation method according to an embodiment of the present application, where the method includes:
and 11, acquiring a target calculation task and identifying the target calculation task.
Specifically, the target computing task can be identified by reading task information carried by a target computing task data packet, the target computing task can also be identified by scanning the data packet, and the target computing task can also be identified by setting keywords, in short, the identification of the target computing task is to acquire information of the target computing task, so that parameter information of the target computing task required by computing power distribution can be acquired subsequently; the task information may include task publisher information, task data type information (image data, audio data, etc.), storage space required by the task, feedback time requirements (real-time, non-real-time, deadline, etc.) for the task, security level of the task, privacy requirements for the task, and so on.
Specifically, the type of the target computing task, that is, the type of data to be processed by the target computing task, is as follows: image data, audio data, text data, mathematical operations, and the like; specifically, the type of the target computing task can be classified according to the application scene of the target computing task, such as voice recognition, character recognition, graphic recognition, image processing, and the like; computing resources more suitable for the computing task can be matched for the target computing task through classification, so that the computing efficiency is improved; for example, if the target computing task belongs to image data, when computing resources are allocated, GPU computing resources suitable for image data processing are preferentially allocated to the target computing task; if the target calculation task belongs to mathematical operation, when the calculation power resource is distributed, the calculation power resource of the CPU suitable for calculation is preferentially distributed to the target calculation task.
For example, the related parameters of the target computing task may also be adjusted according to an actual computing power allocation policy to update the number and types of the related parameters that need to be acquired, for example, in order to perform computing power allocation, the related parameter category that needs to be acquired is C1 (for example, 10 items of the related parameters of each target computing task need to be acquired in the C1 category table), and when the computing power resources that can be currently allocated are changed, the related parameter category may be updated to C2 (for example, 12 items of the related parameters of each target computing task need to be acquired in the C2 category table, and the types and numbers of the related parameters of C2 and C1 may be different) in order to facilitate computing power allocation.
Specifically, the type of the computing resources corresponds to the type of the target computing task, in order to better match the computing resources, the data types of the computing resources and/or the task requirement type need to be obtained, and the computing resources with higher computing power matched to the corresponding type can also enable the target computing task to be executed more quickly.
And step 14, distributing computing power resources for the target computing task according to the current computing power resource information and the relevant parameters of the target computing task.
Optionally, in some embodiments of the present application, the method further includes:
classifying the target computing task according to the type of the target computing task;
determining the data type of the target computing task according to the classification result;
and matching the calculation resources of the data type suitable for the target calculation task according to the data type of the target calculation task.
Specifically, the target computing task may be classified for the first time according to the type of the target computing task when or after the target computing task is identified, the data type of the target computing task is determined through the classification for the first time, and computing resources suitable for the data type are matched for the target computing task according to the data type of the target computing task; after the first classification, the type of the target computing task can be classified for the second time to obtain a secondary classification result of the target computing task, and a computing mode applicable to the target computing task is determined according to the classification result of the target computing task.
Illustratively, when computing resources are matched for the target computing task, the type of the target computing task is obtained and can be matched with more suitable computing resources, so that computing allocation is more reasonable, and computing cost and server hardware cost are reduced; for example, for the target computing task A subjected to primary classification, the type of the target computing task A is image data, and GPU computing resources suitable for image data processing are matched for the image data; if the secondary classification is further performed, the target calculation task a is a face image, and face recognition is required, and calculation resources suitable for face recognition, such as a calculation resource module carrying a face recognition algorithm, are matched for the face image.
Illustratively, the data types include at least image data, audio data, textual data, mathematical operations, and the like; if the data type of the target computing task A is image data, and the image data is subjected to secondary classification, determining that the type of the image data is as follows: image data, namely a face image, which needs face recognition; then, computational resources suitable for face recognition, such as computational resource modules carrying face recognition algorithms or computational resources with higher computational power in face image processing, can be matched for the target computation task a according to the result of the secondary classification.
Illustratively, the type of the secondary classification further includes that, if the data type of the target computing task is image data, the secondary classification result of the target computing task at least includes one of the following: still images, video images, face images, animal images, other object images, images requiring motion recognition, images requiring character recognition.
Specifically, the calculation methods of the video and the still image are different, and the video and the still image are used for recognizing human faces, animals and other objects, or for motion recognition, and the calculation methods specifically adapted to the image data are different, and even if the target calculation task obtained by one-time classification belongs to the image data, it is obviously not as efficient to match a calculation resource used for image data processing at will as to further match a calculation resource carrying a corresponding processing algorithm or having higher efficiency in calculating corresponding type image data.
Illustratively, the secondary classification type further includes, if the target computing task data type is audio data, the secondary classification result of the target computing task at least includes one of the following: voice data, music data, audio detection data (sonar, etc.), and other sound data.
Illustratively, the secondary classification type further includes that, if the target computing task data type is text data, the secondary classification result of the target computing task at least includes one of the following: character recognition is required, semantic recognition is required, and classification is required.
Optionally, in some embodiments of the present application, the allocating computing resources to the target computing task according to the current computing resource information and the relevant parameters of the target computing task includes:
evaluating all current computing resources according to the relevant parameters of the target computing task and the current computing resource information to obtain a computing power adaptation degree score of each computing resource to the target computing task;
generating a calculation power adaptation degree grading list according to calculation power adaptation degree grading results of all current calculation power resources on the target calculation task;
and selecting proper computing power resources from the computing power suitability grade list for executing the target computing task according to a preset computing power distribution rule.
Illustratively, if the current computing resource has m computing resources numbered 1-m, the m computing resources are evaluated first, and the computing power suitability score result S = { S } of each computing resource is obtained 1 ，S 2 ，……，S m }; generating a computation power suitability score list according to the result, specifically, a computation power suitability score list is shown with reference to fig. 2The schematic diagram comprises information such as the calculation force resource number, the calculation force adaptation degree score, the total capacity of the calculation force resource, the current available capacity, the data transmission speed, the current load and the like; obviously, the required categories in the calculation power suitability scoring list can be adjusted according to the actual use requirement; and then, distributing proper computing power resources for the target computing task by combining with a specific computing power distribution rule.
It should be noted that the suitable computation resource is not necessarily the computation resource with the highest computation fitness score, and if the target computation task B is a real-time task, the computation resource with the highest computation fitness score may be unstable at the transmission speed or have a common transmission speed for the task B, then for the computation tasks such as the task B, the computation resource may be preferentially allocated according to the transmission speed under the condition that the threshold of the computation fitness score is satisfied; if the target computing task C is a non-real-time task and the time tolerance is high, the computing power resource with the highest computing power suitability score is possibly insufficient in available capacity for the target computing task C, and needs to be split into a plurality of subtasks for matching, the target computing task C is convenient to distribute and concise, and the target computing task C can be distributed with a computing power resource with a large capacity under the conditions of meeting the computing power suitability threshold and meeting the aging requirement without splitting, so that the task can be prevented from being split and combined, and the simplicity of task data management is facilitated. In the practical application process, the operation and maintenance personnel can dynamically adjust the calculation power distribution rule according to the current requirement, and can also automatically distribute the calculation power according to the preset calculation power distribution rule. It should be noted that, in the case that the computation power fitness threshold is met, the computation power resource finally applicable may be selected and allocated according to specific situations, and the final allocation result is not necessarily the computation power resource with the highest computation power fitness score.
Optionally, in some embodiments of the present application, the evaluating all current computing resources according to the relevant parameters of the target computing task and the current computing resource information includes obtaining a computing power suitability score of each computing resource for the target computing task by a computing power suitability scoring formula:
S m scoring the computing power suitability of the computing power resource m to the target computing task, wherein m is the number of the computing power resource;
L i calculating the type of the task for the target;
b is the size of the target calculation task;
t is the time efficiency requirement of the target calculation task;
k T a time tolerance coefficient corresponding to the aging requirement T;
V m-Li for computing power m is directed to type L i Computing power capability during the computing task of (1);
Q m is the computing power floating coefficient of the computing power resource m.
In particular, the computing task type L i ∈{L 1 ，L 2 ，……，L n H, each task type number L i Corresponding to a specific type of computing task;
V m-Li for computing power capability, it means that the computing power resource m is L for the type i The calculation efficiency and speed when calculating the task type are one performance parameter of the calculation power resource m;
Q m the computing power capability floating coefficient of the computing power resource m represents the floating of the computing performance of the computing power resource m and is influenced by factors such as task occupancy rate, hardware state, network delay and the like of the computing power resource m;
further, the computational resource has a storage capacity of R; the type of computing resources is also L i (one-to-one correspondence with the type of target computing task).
When computing power matching is carried out, each current computing power resource is evaluated according to the relevant parameters of the target computing task and the current computing power resource information, a computing power adaptation degree score S of the target computing task is obtained, and the adaptation degrees of all current computing power resources and the target computing task can be obtained accurately.
In particular, with respect to Q m The computing power capability floating coefficient of the computing power resource m can greatly slide down when the task occupancy rate of a certain computing power resource, such as the computing power resource m, is too high; the health state of the hardware behind the computing resources also influences the computing capacity of the computing resources, and if the hardware has overhigh temperature, faults and the like, the computing capacity of the computing resources also decreases along with the abnormal health of the hardware; the network state of the non-local computing resources is also a factor to be considered when computing power capacity is considered; therefore, the current computing power of the computing power resource needs to be adjusted by the computing power floating coefficient Q of the computing power resource, and the computing power floating coefficient of the computing power resource m can be obtained by the following formula:
Q m ＝E m ·A m ·W m ·k m
E m for calculating the task occupancy rate coefficient of the force resource m, due to the determination of the task occupancy rate, a mapping table of the task occupancy rate coefficient and the task occupancy rate can be preset according to the requirement;
A m the hardware health coefficient of the computational resource m is determined by the hardware health state of the computational resource m, and the mapping table of the hardware health coefficient and the hardware health condition can be preset according to requirements;
W m the network state coefficient of the computational resource m is determined by the network state of the computational resource m, and a mapping table between the network state coefficient and the network delay and the data transmission speed in the network can be preset according to requirements;
k m setting and adjusting the adjustment coefficient of the computing power resource m in computing the computing power floating coefficient according to requirements so as to adjust errors or convert numerical values, so that Q m Can be applied to calculate the fitness score S m In the evaluation of (1).
Optionally, in some embodiments of the present application, the method further includes:
matching the same or similar historical tasks in a historical task database according to the related parameters of the target calculation task;
and if the same or similar historical tasks are matched in the historical task database, distributing computing power resources for the target computing task according to the historical computing power distribution records of the same or similar historical tasks, the current computing power resource information and the related parameters of the target computing task.
Optionally, the method further includes:
if the similarity between the target calculation task and the target historical task exceeds a similarity threshold value;
the target historical task is a similar task of the target calculation task;
and distributing computing power to the target computing task according to the similarity of the target historical task from high to low until the distribution is successful or all the distribution fails.
For example, the same or similar historical tasks may be obtained by determining whether the similarity value between the target computing task and the historical task exceeds a threshold, such as the target computing task a, which has a similarity of 80% to the historical task X001, a similarity of 50% to the historical task X002, a similarity of 90% to the historical task X003, a similarity of 30% to the historical task X004, and a similarity threshold of 70%; then, the X001 and the X003 with the similarity exceeding 70% are used as similar tasks of the calculation task A, 90% is more than 80%, so that calculation power distribution is performed by preferentially referring to the X003 historical task for the target calculation task A, if distribution fails due to problems such as current calculation power resource state problems (such as occupation, data transmission speed and faults) and the like, calculation power adaptation degree evaluation is performed on distribution calculation power according to calculation power distribution information pairs of the next similarity-oriented historical task X002, and so on until distribution succeeds, or distribution failure of all historical tasks meeting the similarity threshold requirement is performed.
Through the historical task data matching function, calculation power distribution of calculation tasks can be rapidly achieved, calculation power distribution efficiency is improved, and historical calculation task data can be used.
Optionally, in some embodiments of the present application, the method further includes:
matching the same or similar historical tasks in a historical task database according to the related parameters of the target calculation task;
if the same or similar historical tasks are not matched in the historical task database and the type of the target computing task cannot be determined, randomly extracting a plurality of subtasks from the target computing task;
calculating the plurality of subtasks, and evaluating the computing power requirement of the target computing task according to the calculation result;
and distributing computing power resources for the target computing task according to the computing power demand evaluation result and the related parameters of the target computing task.
For example, when history matching cannot be matched with a related or similar task, and a specific type of the task cannot be identified, extracting part of data in the target calculation as subtasks, and performing calculation by the subtasks to determine the required calculation power and the appropriate calculation type, thereby performing calculation power distribution for the target task. By the method, the reasonable calculation power distribution can be performed on the unknown calculation tasks which are not processed by the calculation power distribution system.
Optionally, in some embodiments of the present application, the method further includes:
detecting the state of the target calculation task to obtain relevant parameters of the target calculation task at the moment t;
inputting relevant parameters of the target calculation task at the time t into a calculation power distribution model to obtain a calculation power distribution scheme of the target calculation task at the time t;
dynamically adjusting the computing power resource of the target computing task according to the computing power distribution scheme of the target computing task at the time t;
an iteration operation is performed at time t + 1.
Wherein the calculation force distribution model is obtained based on historical similar task training of the target calculation task.
Optionally, in some embodiments of the present application, the inputting the relevant parameter of the target computing task at time t into the computation power allocation model to obtain the computation power allocation scheme of the target computing task at time t includes:
acquiring a calculation force distribution strategy set of a target calculation task at the time t according to relevant parameters of the target calculation task at the time t;
acquiring a calculation power distribution scheme of the target calculation task at the time t according to the calculation power distribution strategy set of the target calculation task at the time t;
and storing the relevant parameters of the target computing task at the time t, the computing power distribution strategy set of the target computing task at the time t and the computing power distribution scheme of the target computing task at the time t into a historical task database.
Specifically, the calculation power distribution scheme selects one calculation power distribution scheme from a plurality of strategies in the calculation power distribution strategy set to update the target calculation task, so that the dynamic adjustment of the calculation power distribution is realized.
The method for acquiring the computing power distribution strategy set of the target computing task at the moment t according to the relevant parameters of the target computing task at the moment t comprises the following steps:
acquiring a calculation power estimation allocation strategy of the target calculation task at the time t according to relevant parameters of the target calculation task at the time t;
obtaining the value G of an estimation strategy through a first evaluation network according to the relevant parameters of a target calculation task at the time t and the calculation power estimation distribution strategy of the target calculation task at the time t e ；
Obtaining the value G of the target strategy according to the relevant parameters of the target calculation task at the moment t through a second evaluation network t ；
Value G according to target strategy t Estimating value of strategy G e And acquiring a calculation power distribution strategy set of the target calculation task, and updating parameters of the first evaluation network and the second evaluation network.
Specifically, a feedback value of the execution target strategy at the moment t is obtained;
executing the feedback value of the target strategy at the time t and the value G of the target strategy t And estimating the value G of the strategy e The difference is input into an evaluation network, and the value of an estimation strategy at the moment t is fed back by the first evaluation networkThe gradient is used for adjusting the second evaluation network to obtain a strategy closer to the target;
meanwhile, parameters of the first evaluation network and the second evaluation network are updated;
all the parameter updating and learning experiences are stored in a historical database;
all networks can call all learned parameters and experiences of the computational power allocation task from a historical database;
the iteration operation is performed at the next instant.
Specifically, the allocation strategy is evaluated through an evaluation network, the first evaluation network evaluates the value of the estimation strategy, the estimation strategy can be a plurality of strategies, and the value G of the obtained estimation strategy e Can be the value G of multiple estimation strategies e A set of compositions; the second evaluation network obtains the value G of the target strategy through the relevant parameters of the target calculation task at the next moment t +1 t Value G of target strategy at time t +1 t And the value G of the estimation strategy e The difference between the two network resources and the feedback value of the execution target strategy at the simulation time t are used for adjusting the second evaluation network, so that a better calculation resource control strategy is obtained.
Specifically, the method further comprises:
obtaining the value G of the calculation power distribution strategy in use at the moment t of the target calculation task through a third evaluation network y ；
Value G according to target strategy t Estimating value of strategy G e And value G of calculation power distribution strategy in use y And acquiring a calculation power distribution strategy set of the target calculation task, and updating parameters of the first evaluation network, the second evaluation network and the third evaluation network.
Specifically, a third evaluation network is added to obtain the current ongoing calculation power distribution strategy, and the value G of the target strategy at the moment t +1 is passed t And the value G of the estimation strategy e And value G of the calculation power distribution strategy in use at the current moment y The difference between them to obtain the whole calculation force distribution modelAnd adjusting the parameters to obtain a more efficient force distribution strategy. The gap between the value of the estimation strategy and the value of the target strategy at the next moment and the value of the calculation power distribution strategy currently in use is fully considered.
Specifically, the value G of the target strategy at the moment t +1 can be passed t And the value G of the estimation strategy e And value G of the calculation power distribution strategy in use at the current moment y The difference between the two strategies and the feedback value of the simulation execution target strategy at the time t are obtained, and the strategy closer to the target is obtained by adjusting the second evaluation network;
the feedback value of the execution target strategy can be used for embodying the following steps: computing power capability floating coefficient, strategy execution value and/or target computing task state;
that is, the feedback value for executing the target strategy can be adjusted according to the actual requirement, so that the better effect can be achieved, and the feedback value can be a value gradient, or a comprehensive value containing the computing power capability of the computing power resource, the computing power resource state and the target computing task state, or a gradient of the comprehensive value.
Referring to fig. 3, a schematic diagram of a computing power distribution flow of a target computing task provided in an embodiment of the present application is shown; when a target calculation task needs to be subjected to calculation force distribution, firstly identifying the target calculation task; secondly, extracting task related parameters respectively to obtain current calculation force information; then, historical task matching is carried out on the target computing task; if the similar tasks are matched, calculating force distribution is carried out according to the similar tasks; if the calculation power distribution according to the similar tasks fails, calculating power resources are distributed according to the relevant parameters of the target calculation task and the current calculation power information; if the similar tasks are not matched, identifying and classifying two conditions according to the type of the target calculation task, and when the type of the target calculation task can be identified, distributing calculation force resources according to the relevant parameters of the target calculation task and the current calculation force information; and when the type of the target calculation task cannot be identified, extracting the target calculation task, evaluating the calculation power requirement, and then distributing calculation power resources according to the relevant parameters of the target calculation task and the current calculation power information.
To sum up, the server computing power management and distribution method provided by the embodiment of the application can achieve the following beneficial technical effects: calculating power distribution is carried out on the target calculation task according to relevant parameters capable of combining the target calculation task and current calculating power information; particularly, the target computing task can be matched with the appropriate computing power resource through the type of the target computing task; meanwhile, a calculation power adaptation degree evaluation mechanism is introduced, the influence of various parameters in the target calculation task and the current calculation power resource on the calculation power capability of the calculation power resource is comprehensively considered, the rationality of calculation power distribution is greatly improved by matching with the use of calculation power distribution rules, a more suitable and efficient calculation power distribution result is obtained, the completion efficiency of the calculation task is improved, and the calculation power cost and the hardware cost waste are reduced; moreover, a historical task information matching mechanism can be supported, and the calculation power distribution efficiency is improved; and unknown calculation tasks can be classified after being evaluated, so that the calculation power distribution efficiency is improved.
Referring to fig. 4, an embodiment of the present application provides a server power management distribution system 40, including:
the task identification module 41 is configured to acquire a target computing task and identify the target computing task;
a task parameter extraction module 42, configured to extract, according to the recognition result of the target computing task, a relevant parameter of the target computing task; wherein the relevant parameters of the target computing task at least comprise: the type of the target calculation task, the size of the target calculation task and the timeliness requirement of the target calculation task;
a calculation resource monitoring module 43, configured to obtain current calculation resource information; wherein the computing resource information at least comprises: calculating the type, storage capacity and calculation speed of the resource;
and the calculation resource allocation module 44 is configured to allocate calculation resources to the target calculation task according to the current calculation resource information and the relevant parameters of the target calculation task.
Optionally, in some embodiments of the present application, the system 40 further includes:
a task classification module 45, configured to classify the target computing task according to the type of the target computing task; determining the data type of the target computing task according to the classification result;
the calculation power resource allocation module 44 is further configured to match, for the target calculation task, the calculation power resource of the data type suitable for the target calculation task according to the data type of the target calculation task.
Optionally, in some embodiments of the present application, the computational resource allocation module 44 is further configured to:
evaluating all current computing resources according to the relevant parameters of the target computing task and the current computing resource information to obtain a computing power adaptation degree score of each computing resource to the target computing task;
generating a calculation power adaptation degree grading list according to calculation power adaptation degree grading results of all current calculation power resources on the target calculation task;
and selecting proper computing power resources from the computing power suitability grade list for executing the target computing task according to a preset computing power distribution rule.
Optionally, in some embodiments of the present application, the computational resource allocation module 44 is further configured to:
obtaining the calculation power adaptation degree score of each calculation power resource to the target calculation task through a calculation power adaptation degree scoring formula:
S m scoring the computing power suitability of the computing power resource m to the target computing task, wherein m is the number of the computing power resource;
L i calculating the type of the task for the target;
b is the size of the target calculation task;
t is the time efficiency requirement of the target calculation task;
k T to correspond to the time tolerance requirement TA coefficient;
V m-Li for computing power m is directed to type L i Computing power capability during the computing task of (1);
Q m is the computing power floating coefficient of the computing power resource m.
Optionally, in some embodiments of the present application, the computing power resource allocation module 44 is further configured to:
matching the same or similar historical tasks in a historical task database according to the related parameters of the target calculation task;
and if the same or similar historical tasks are matched in the historical task database, distributing computing power resources for the target computing task according to the historical computing power distribution records of the same or similar historical tasks, the current computing power resource information and the related parameters of the target computing task.
Optionally, in some embodiments of the present application, the computational resource allocation module 44 is further configured to:
matching the same or similar historical tasks in a historical task database according to the related parameters of the target calculation task;
if the same or similar historical tasks are not matched in the historical task database and the type of the target computing task cannot be determined, randomly extracting a plurality of subtasks from the target computing task;
calculating the plurality of subtasks, and evaluating the computing power requirement of the target computing task according to the calculation result;
and distributing computing power resources for the target computing task according to the computing power demand evaluation result and the related parameters of the target computing task.
Optionally, in some embodiments of the present application, the system 40 further includes:
the task monitoring module 46 is configured to detect a state of the target computing task, and obtain a relevant parameter of the target computing task at time t;
the calculation power resource allocation module 44 is further configured to input relevant parameters of the target calculation task at time t into the calculation power allocation model, and obtain a calculation power allocation scheme of the target calculation task at time t;
dynamically adjusting the computing power resource of the target computing task according to the computing power distribution scheme of the target computing task at the time t;
an iteration operation is performed at time t + 1.
Wherein the calculation power distribution model is obtained based on historical similar task training of the target calculation task.
Optionally, in some embodiments of the present application, the computing power resource allocation module 44 is further configured to:
acquiring a calculation force distribution strategy set of a target calculation task at the time t according to relevant parameters of the target calculation task at the time t;
acquiring a calculation power distribution scheme of the target calculation task at the time t according to the calculation power distribution strategy set of the target calculation task at the time t;
and storing the relevant parameters of the target computing task at the time t, the computing power distribution strategy set of the target computing task at the time t and the computing power distribution scheme of the target computing task at the time t into a historical task database.
Specifically, the calculation power distribution scheme selects one calculation power distribution scheme from a plurality of strategies in the calculation power distribution strategy set to update the target calculation task, so that the dynamic adjustment of the calculation power distribution is realized.
Wherein the computational power allocation model includes a first evaluation network, a second evaluation network, and the computational power resource allocation module 44 is further configured to:
acquiring a calculation power estimation allocation strategy of the target calculation task at the time t according to relevant parameters of the target calculation task at the time t;
obtaining the value G of an estimation strategy through a first evaluation network according to the relevant parameters of a target calculation task at the time t and the calculation power estimation distribution strategy of the target calculation task at the time t e ；
Obtaining the value G of the target strategy according to the relevant parameters of the target calculation task at the moment t through a second evaluation network t ；
Value G according to target strategy t Estimating value of strategy G e And acquiring a calculation power distribution strategy set of the target calculation task, and updating parameters of the first evaluation network and the second evaluation network.
Specifically, a feedback value of the execution target strategy at the moment t is obtained;
executing the feedback value of the target strategy at the time t and the value G of the target strategy t And estimating the value G of the strategy e Inputting the difference into an evaluation network, and feeding back the value gradient of the estimation strategy at the moment t by the first evaluation network so as to adjust the second evaluation network to obtain a strategy closer to the target;
meanwhile, parameters of the first evaluation network and the second evaluation network are updated;
all the parameter updating and learning experiences are stored in a historical database;
all networks can call all learned parameters and experiences of the computational power allocation task from a historical database;
the iteration operation is performed at the next instant.
Specifically, the allocation strategy is evaluated through an evaluation network, the first evaluation network evaluates the value of the estimation strategy, the estimation strategy can be a plurality of strategies, and the value G of the obtained estimation strategy e May be the value G of multiple estimation strategies e A set of compositions; the second evaluation network obtains the value G of the target strategy through the relevant parameters of the target calculation task at the next moment t +1 t Value G of target strategy through time t +1 t And the value G of the estimation strategy e The difference between the two network resources and the feedback value of the execution target strategy at the simulation time t are used for adjusting the second evaluation network, so that a better calculation resource control strategy is obtained.
Specifically, the computation power allocation model further includes a third evaluation network, and the computation power resource allocation module 44 is further configured to:
obtaining the value G of the calculation power distribution strategy in use at the moment t of the target calculation task through a third evaluation network y ；
Value G according to target strategy t Estimating value of strategy G e And value G of calculation power distribution strategy in use y And acquiring a calculation power distribution strategy set of the target calculation task, and updating parameters of the first evaluation network, the second evaluation network and the third evaluation network.
The server power management and distribution system provided in the embodiment of the application can implement each process of the server power management and distribution method embodiment, and can achieve the same technical effect, and is not described here again to avoid repetition.
Referring to fig. 5, an embodiment of the present application further provides a network device 50, which includes a processor 51, a memory 52, and a computer program stored in the memory 52 and capable of running on the processor 51, where the computer program is executed by the processor 51 to implement each process of the server power management and distribution method embodiment, and can achieve the same technical effect, and is not described herein again to avoid repetition.
The embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when executed by a processor, the computer program implements each process of the server computing power management and distribution method embodiment, and can achieve the same technical effect, and is not described herein again to avoid repetition. The computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
Through the description of the foregoing embodiments, it is clear to those skilled in the art that the method of the foregoing embodiments may be implemented by software plus a necessary general hardware platform, and certainly may also be implemented by hardware, but in many cases, the former is a better implementation. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present application.
While the present embodiments have been described with reference to the accompanying drawings, it is to be understood that the invention is not limited to the precise embodiments described above, which are meant to be illustrative and not restrictive, and that various changes may be made therein by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (11)
1. A server computing power management allocation method, the method comprising:
acquiring a target calculation task, and identifying the target calculation task;
extracting relevant parameters of the target calculation task according to the identification result of the target calculation task; wherein the relevant parameters of the target computing task at least comprise: the type of the target calculation task, the size of the target calculation task and the timeliness requirement of the target calculation task;
acquiring current computing power resource information; wherein the computing resource information at least comprises: calculating the type, storage capacity and calculation speed of the resource;
and distributing the computing power resources for the target computing task according to the current computing power resource information and the related parameters of the target computing task.
2. The server computing force management allocation method according to claim 1, wherein the method further comprises:
classifying the target computing task according to the type of the target computing task;
determining the data type of the target computing task according to the classification result;
and matching the calculation resources of the data type suitable for the target calculation task according to the data type of the target calculation task.
3. The server computing power management allocation method according to claim 2, wherein the allocating computing power resources to the target computing task according to the current computing power resource information and the relevant parameters of the target computing task comprises:
evaluating all current computing resources according to the relevant parameters of the target computing task and the current computing resource information to obtain a computing power adaptation degree score of each computing resource to the target computing task;
generating a calculation power adaptation degree grading list according to calculation power adaptation degree grading results of all current calculation power resources on the target calculation task;
and selecting proper computing power resources from the computing power suitability grade list for executing the target computing task according to a preset computing power distribution rule.
4. The server computing power management distribution method according to claim 3, wherein the evaluating all current computing power resources according to the relevant parameters of the target computing task and the current computing power resource information comprises obtaining a computing power fitness score of each computing power resource for the target computing task through a computing power fitness scoring formula:
S m scoring the computing power suitability of the computing power resource m to the target computing task, wherein m is the number of the computing power resource;
L i calculating the type of the task for the target;
b is the size of the target computing task;
t is the time efficiency requirement of the target calculation task;
k T a time tolerance coefficient corresponding to the aging requirement T;
V m-Li for computing power m is directed to type L i Computing power capability during the computing task of (1);
Q m is the computing power floating coefficient of the computing power resource m.
5. The server power management allocation method of claim 4, wherein the method further comprises:
matching the same or similar historical tasks in a historical task database according to the related parameters of the target calculation task;
and if the same or similar historical tasks are matched in the historical task database, distributing computing power resources for the target computing task according to the historical computing power distribution records of the same or similar historical tasks, the current computing power resource information and the related parameters of the target computing task.
6. The server power management allocation method of claim 5, further comprising:
matching the same or similar historical tasks in a historical task database according to the related parameters of the target calculation task;
if the same or similar historical tasks are not matched in the historical task database and the type of the target computing task cannot be determined, randomly extracting a plurality of subtasks from the target computing task;
calculating the plurality of subtasks, and evaluating the computing power requirement of the target computing task according to the calculation result;
and distributing computing power resources for the target computing task according to the computing power demand evaluation result and the related parameters of the target computing task.
7. The server computing force management allocation method according to claim 4, wherein the method further comprises:
detecting the state of the target calculation task to obtain relevant parameters of the target calculation task at the moment t;
inputting relevant parameters of the target calculation task at the time t into a calculation power distribution model to obtain a calculation power distribution scheme of the target calculation task at the time t;
dynamically adjusting the computing power resource of the target computing task according to the computing power distribution scheme of the target computing task at the time t;
performing iterative operation at the moment t + 1;
wherein the calculation force distribution model is obtained based on historical similar task training of the target calculation task.
8. The server computing power management distribution method according to claim 7, wherein the step of inputting relevant parameters of the target computing task at time t into the computing power distribution model to obtain a computing power distribution scheme of the target computing task at time t comprises:
acquiring a calculation force distribution strategy set of a target calculation task at the time t according to relevant parameters of the target calculation task at the time t;
acquiring a calculation power distribution scheme of the target calculation task at the time t according to the calculation power distribution strategy set of the target calculation task at the time t;
and storing the relevant parameters of the target computing task at the time t, the computing power distribution strategy set of the target computing task at the time t and the computing power distribution scheme of the target computing task at the time t into a historical task database.
9. A server computing power management distribution system, the system comprising:
the task identification module is used for acquiring a target calculation task and identifying the target calculation task;
the task parameter extraction module is used for extracting relevant parameters of the target calculation task according to the identification result of the target calculation task; wherein the relevant parameters of the target computing task at least comprise: the type of the target calculation task, the size of the target calculation task and the timeliness requirement of the target calculation task;
the computing resource monitoring module is used for acquiring current computing resource information; wherein the computing resource information at least comprises: calculating the type, storage capacity and calculation speed of the resource;
and the computing power resource allocation module is used for allocating computing power resources for the target computing task according to the current computing power resource information and the relevant parameters of the target computing task.
10. A network device, comprising: a processor, a memory and a program stored on the memory and executable on the processor, the program when executed by the processor implementing the steps of the server computing power management distribution method of any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that a computer program is stored thereon, which computer program, when being executed by a processor, realizes the steps of the server power management distribution method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210991170.5A CN115658282A (en) | 2022-08-18 | 2022-08-18 | Server computing power management distribution method, system, network device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210991170.5A CN115658282A (en) | 2022-08-18 | 2022-08-18 | Server computing power management distribution method, system, network device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115658282A true CN115658282A (en) | 2023-01-31 |
Family
ID=84983420
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210991170.5A Pending CN115658282A (en) | 2022-08-18 | 2022-08-18 | Server computing power management distribution method, system, network device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115658282A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116361006A (en) * | 2023-03-31 | 2023-06-30 | 山东省计算中心（国家超级计算济南中心） | Method and system for collaborative scheduling of computing network resources oriented to memory computing separation |
| CN116521380A (en) * | 2023-07-05 | 2023-08-01 | 之江实验室 | Resource self-adaptive collaborative model training acceleration method, device and equipment |
| CN116909758A (en) * | 2023-09-13 | 2023-10-20 | 中移(苏州)软件技术有限公司 | Processing method and device of calculation task and electronic equipment |
| CN117055818A (en) * | 2023-08-25 | 2023-11-14 | 南京合宜信息技术有限公司 | Client information storage management method and system based on block chain |
| CN117520345A (en) * | 2023-11-15 | 2024-02-06 | 北京宝联之星科技股份有限公司 | Data processing method, system and medium based on cloud computing and blockchain |
| CN117687798A (en) * | 2024-02-01 | 2024-03-12 | 浪潮通信信息系统有限公司 | Management and control method, system and storage medium for original application of computing power network |
-
2022
- 2022-08-18 CN CN202210991170.5A patent/CN115658282A/en active Pending
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116361006A (en) * | 2023-03-31 | 2023-06-30 | 山东省计算中心（国家超级计算济南中心） | Method and system for collaborative scheduling of computing network resources oriented to memory computing separation |
| CN116361006B (en) * | 2023-03-31 | 2023-11-21 | 山东省计算中心（国家超级计算济南中心） | Method and system for collaborative scheduling of computing network resources oriented to memory computing separation |
| CN116521380A (en) * | 2023-07-05 | 2023-08-01 | 之江实验室 | Resource self-adaptive collaborative model training acceleration method, device and equipment |
| CN117055818A (en) * | 2023-08-25 | 2023-11-14 | 南京合宜信息技术有限公司 | Client information storage management method and system based on block chain |
| CN117055818B (en) * | 2023-08-25 | 2024-05-10 | 南京合宜信息技术有限公司 | Client information storage management method and system based on block chain |
| CN116909758A (en) * | 2023-09-13 | 2023-10-20 | 中移(苏州)软件技术有限公司 | Processing method and device of calculation task and electronic equipment |
| CN116909758B (en) * | 2023-09-13 | 2024-01-26 | 中移(苏州)软件技术有限公司 | Processing method and device of calculation task and electronic equipment |
| CN117520345A (en) * | 2023-11-15 | 2024-02-06 | 北京宝联之星科技股份有限公司 | Data processing method, system and medium based on cloud computing and blockchain |
| CN117520345B (en) * | 2023-11-15 | 2024-04-23 | 北京宝联之星科技股份有限公司 | Data processing method, system and medium based on cloud computing and blockchain |
| CN117687798A (en) * | 2024-02-01 | 2024-03-12 | 浪潮通信信息系统有限公司 | Management and control method, system and storage medium for original application of computing power network |
| CN117687798B (en) * | 2024-02-01 | 2024-05-10 | 浪潮通信信息系统有限公司 | Management and control method, system and storage medium for original application of computing power network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115658282A (en) | Server computing power management distribution method, system, network device and storage medium | |
| CN108280670B (en) | Seed crowd diffusion method and device and information delivery system | |
| CN110462607B (en) | Identifying reason codes from gradient boosters | |
| CN110659311B (en) | Topic pushing method and device, electronic equipment and storage medium | |
| CN107909038B (en) | Social relationship classification model training method and device, electronic equipment and medium | |
| CN107516102B (en) | Method, device and system for classifying image data and establishing classification model | |
| CN110969243B (en) | Method and device for training countermeasure generation network for preventing privacy leakage | |
| CN111243601A (en) | Voiceprint clustering method and device, electronic equipment and computer-readable storage medium | |
| CN111931809A (en) | Data processing method and device, storage medium and electronic equipment | |
| CN113962800A (en) | Model training and overdue risk prediction method, device, equipment and medium | |
| CN111444930B (en) | Method and device for determining prediction effect of two-classification model | |
| CN114268839A (en) | Video publishing account screening method and device, electronic equipment and storage medium | |
| CN113642652A (en) | Method, device and equipment for generating fusion model | |
| US8918406B2 (en) | Intelligent analysis queue construction | |
| CN117475253A (en) | Model training method and device, electronic equipment and storage medium | |
| CN109145207B (en) | Information personalized recommendation method and device based on classification index prediction | |
| CN116563604A (en) | End-to-end target detection model training, image target detection method and related equipment | |
| CN115600818A (en) | Multi-dimensional scoring method and device, electronic equipment and storage medium | |
| CN109308565B (en) | Crowd performance grade identification method and device, storage medium and computer equipment | |
| CN111737371B (en) | Data flow detection classification method and device capable of dynamically predicting | |
| CN115063858A (en) | Video facial expression recognition model training method, device, equipment and storage medium | |
| CN111753519B (en) | Model training and identifying method and device, electronic equipment and storage medium | |
| KR102078541B1 (en) | Issue interest based news value evaluation apparatus and method, storage media storing the same | |
| US20200265270A1 (en) | Mutual neighbors | |
| JP7010383B2 (en) | Judgment method and judgment program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |