Computing power scheduling method

By analyzing the influence coefficient and word count characteristics of keywords and non-keywords in the task request, the total resource requirements of the task request are determined, and the available resources of the computing power node are matched, the problem of resource mismatch is solved and the efficient utilization of resources is achieved and efficient scheduling of computing power nodes.

CN118916165APending Publication Date: 2024-11-08CHINA MOBILE INTERNET CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411028636.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the existing computing power node scheduling scheme, the problem that the resources required for the request do not match the resources of the computing power node, resulting in shortage of resources or vacant resources.

Method used

By obtaining the task request for the pending task and the available resources of each computing power node, the total resource requirements of the task request are determined based on the resource requirements corresponding to the keywords in the task request, the coefficient of influence of non-keywords on the task request and the number of words characteristics. And allocate the target computing power nodes according to the principle that available resources match the needs.

Benefits of technology

Optimize the utilization of resources, realize the matching of resources required for task requests with the supply resources of computing power nodes, and improve the efficient scheduling of computing power nodes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916165A_ABST
    Figure CN118916165A_ABST
Patent Text Reader

Abstract

The invention discloses a computing power scheduling method. The method comprises the steps of obtaining a task request of a to-be-processed task and available resources of each computing power node; determining a total resource demand corresponding to the task request according to a resource demand corresponding to a keyword in the task request, an influence coefficient of a non-keyword in the task request on the task request, and a word number feature of the task request; allocating a target computing power node to the to-be-processed task according to a total resource demand corresponding to the task request and available resources of each computing power node; wherein the available resources of the target computing power node are matched with the total resource demand corresponding to the task request.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of communication technology, and specifically relates to a computing power scheduling method. Background Art

[0002] In related technologies, computing node scheduling schemes mainly include scheduling appropriate computing nodes based on the idle resources of each computing node in the cluster, such as the idleness of the computing node's central processing unit (CPU) and remaining memory. Alternatively, the appropriate computing node can be scheduled based on information carried in the request, such as the token, Internet Protocol (IP) address, and port number.

[0003] However, the above solution still has the problem of mismatch between the resources required for the request and the resources of the computing power nodes. For example, some computing power nodes are short of resources due to carrying too many requests, or some computing power nodes have idle resources because the resources required for the requests are too few. Summary of the Invention

[0004] The embodiments of the present application provide a computing power scheduling method that can solve the problem in related technologies that the resources required for a request do not match the resources of the computing power node.

[0005] In the first aspect, an embodiment of the present application provides a computing power scheduling method, which includes: obtaining a task request for a task to be processed and the available resources of each computing power node; determining the total resource demand corresponding to the task request based on the resource demand corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count characteristics of the task request; allocating a target computing power node to the task to be processed based on the total resource demand corresponding to the task request and the available resources of each computing power node; wherein the available resources of the target computing power node match the total resource demand corresponding to the task request.

[0006] In the second aspect, an embodiment of the present application provides a computing power scheduling system, which includes: a computing power scheduling center, a resource estimation gateway, and a resource management center; wherein the computing power scheduling center is used to obtain task requests for tasks to be processed; the resource management center is used to obtain available resources of each computing power node; the resource estimation gateway is used to determine the total resource demand corresponding to the task request based on the resource demand corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count characteristics of the task request; the computing power scheduling center is also used to allocate target computing power nodes to the tasks to be processed based on the total resource demand corresponding to the task request and the available resources of each computing power node; wherein the available resources of the target computing power node match the total resource demand corresponding to the task request.

[0007] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

[0008] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0009] In a fifth aspect, an embodiment of the present application provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the steps of the method described in the first aspect.

[0010] In a sixth aspect, an embodiment of the present application provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes a program or instructions, which, when executed, implement the steps of the method described in the first aspect.

[0011] In an embodiment of the present application, the task request of the task to be processed and the available resources of each computing power node are obtained; the total resource demand corresponding to the task request is determined according to the resource demand corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count characteristics of the task request; the target computing power node is allocated to the task to be processed according to the total resource demand corresponding to the task request and the available resources of each computing power node; wherein the available resources of the target computing power node match the total resource demand corresponding to the task request, thereby allocating a suitable computing power node to the task to be processed based on the estimation of the resources required for the task request and the resource supply situation of the computing power node, so that the resources required for the task request match the supply resources of the computing power node, optimize resource utilization, and achieve efficient scheduling of the computing power nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of a computing power scheduling method provided in an embodiment of the present application;

[0013] Figure 2 This is a schematic diagram of the structure of a computing power scheduling system provided in an embodiment of the present application;

[0014] Figure 3 This is a flowchart of another computing power scheduling method provided in an embodiment of the present application;

[0015] Figure 4 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0017] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0018] The computing power scheduling method provided in the embodiment of the present application is described in detail below with reference to the accompanying drawings through specific embodiments and their application scenarios.

[0019] Figure 1 A flow chart of a computing power scheduling method provided by an embodiment of the present application is shown, which can be executed by an electronic device. Figure 1 , the method may include the following steps.

[0020] Step 102: Obtain the task request of the task to be processed and the available resources of each computing node.

[0021] Among them, the task request includes keywords and non-keywords. By analyzing the resource requirements corresponding to the keywords in the task request and the influence coefficient of non-keywords on the task request, the task request is analyzed and the resource requirements are estimated to determine the total resource requirements corresponding to the task request.

[0022] Step 104 : determining the total resource requirement corresponding to the task request based on the resource requirement corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count feature of the task request.

[0023] The resource requirements corresponding to the keywords in a task request reflect the primary resource requirements of the task request. Non-keywords in a task request may also affect the entire task request or keywords. By using the influence coefficient of non-keywords on the task request, we can improve the accuracy of estimating the total resource requirements corresponding to the task request. The word count feature of a task request can include the total word count of the task request and the word count of the keywords.

[0024] Optionally, the resource requirements corresponding to each keyword can be stored in a database. When estimating the total resource requirements corresponding to a task request, the resource requirements corresponding to each keyword in the task request are queried from the database, and the sum of the resource requirements corresponding to each keyword is determined as the resource requirements corresponding to the keyword in the task request.

[0025] Step 106: Allocate a target computing node for the task to be processed based on the total resource demand corresponding to the task request and the available resources of each computing node.

[0026] The available resources of the target computing power node match the total resource requirements corresponding to the task request. Optionally, the matching of the available resources of the target computing power node with the total resource requirements corresponding to the task request may include a single condition matching, such as the graphics processing unit (GPU) video memory of the target computing power node meeting the GPU video memory required for the task request. It may also include multiple conditions matching, such as the GPU video memory and double-precision floating-point computing power meeting the GPU video memory and double-precision floating-point computing power required for the task request, and comprehensive conditions such as the performance of the central processing unit (CPU), memory conditions, and disk conditions.

[0027] In an embodiment of the present application, by obtaining the task request of the task to be processed and the available resources of each computing power node, the total resource demand corresponding to the task request is determined according to the resource demand corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count characteristics of the task request, and the target computing power node is allocated to the task to be processed according to the total resource demand corresponding to the task request and the available resources of each computing power node; wherein the available resources of the target computing power node match the total resource demand corresponding to the task request, so that based on the estimation of the resources required for the task request and the resource supply situation of the computing power node, a suitable computing power node is allocated to the task to be processed, so that the resources required for the task request match the supply resources of the computing power node, the resource utilization is optimized, and the computing power nodes are efficiently scheduled.

[0028] In one implementation, the above-mentioned step 104 determines the total resource requirements corresponding to the task request based on the resource requirements corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count characteristics of the task request, and may include the following steps.

[0029] Step 1041: perform word segmentation processing on the task request to obtain keywords and non-keywords.

[0030] Among them, keywords are separated from the task request to determine the resource requirements corresponding to the keywords, and non-keywords are separated to analyze the impact of non-keywords on the task request or keywords, so as to accurately determine the total resource requirements corresponding to the task request.

[0031] Step 1042 : Determine the influence coefficient of the non-keyword on the task request according to the basic influence coefficient of the non-keyword on the task request and the supplementary influence coefficient of the non-keyword on the keyword.

[0032] The impact coefficient of non-keywords on task requests includes a basic impact coefficient and a supplementary impact coefficient. The basic impact coefficient is related to the characteristics of the non-keyword itself, while the supplementary impact coefficient is related to the correlation between the non-keyword and the keyword.

[0033] Step 1043 : Determine the total resource requirement corresponding to the task request based on the resource requirement corresponding to the keyword, the impact coefficient, and the word count feature of the task request.

[0034] The word count feature of the task request may include the total word count of the task request and the word count of the keyword.

[0035] In an embodiment of the present application, by performing word segmentation processing on the task request, the keywords and non-keywords in the task request are obtained, and then the keywords and non-keywords are analyzed separately. The analysis of non-keywords includes the basic influence coefficient of non-keywords on the task request based on their own characteristics, and the supplementary influence coefficient of non-keywords on keywords, thereby combining the basic influence coefficient of the overall influence of non-keywords on the task request and the supplementary influence coefficient of the key local influence of non-keywords on keywords to determine the total influence coefficient of non-keywords on the task request, so as to determine the total resource demand corresponding to the task request based on the resource demand corresponding to the keyword, the influence coefficient, and the word count characteristics of the task request, which can effectively improve the estimation accuracy of the total resource demand corresponding to the task request.

[0036] In one embodiment, the total resource requirement QR corresponding to the task request can be determined by the following formula:

[0037] Among them, QL is the total number of words in the task request, CL is the number of words in the keyword, QCR is the resource requirement corresponding to the keyword in the task request, and QCNF is the influence coefficient of non-keywords on the task request.

[0038] In one implementation, the above step 1042 determines the influence coefficient of the non-keyword on the task request based on the basic influence coefficient of the non-keyword on the task request and the supplementary influence coefficient of the non-keyword on the keyword, which may include the following steps.

[0039] Step 21 : determining a basic influence coefficient of the non-keyword on the task request according to the category of the non-keyword.

[0040] Among them, the basic influence coefficient corresponding to the non-keyword can be determined based on the characteristics of the non-keyword itself and the preset correlation between the characteristics and the basic influence coefficient. For example, the category of non-keywords can include nouns, directional verbs, adjectives, numerals, quantifiers, willing pronouns, adverbs, prepositions, conjunctions, auxiliary words, onomatopoeia, and interjections. Different categories of non-keywords correspond to different basic influence coefficients. Table 1 shows an exemplary non-keyword category and basic influence coefficient provided in an embodiment of the present application. Among them, the basic influence coefficients corresponding to the non-keyword categories in different task requests may be different, and Table 1 is only used as an example.

[0041] Table 1.

[0042]

[0043] Step 22: According to the category of the non-keyword, the distance between the non-keyword and the keyword, and the basic influence coefficient, the supplementary influence coefficient and the harmonic coefficient of the non-keyword on the keyword are determined through the trained neural network model.

[0044] The supplementary influence coefficient may include the supplementary influence coefficient of the non-keyword on the keyword adjacent to the non-keyword, or may include the supplementary influence coefficient of the non-keyword on all keywords in the task request. The trained neural network model is used to process the category of the non-keyword, the distance between the non-keyword and the keyword, and the basic influence coefficient to determine the supplementary influence coefficient and the harmonic coefficient of the non-keyword on the keyword.

[0045] Step 23: Determine the influence coefficient of the non-keyword on the task request based on the basic influence coefficient, the supplementary influence coefficient, and the harmonic coefficient.

[0046] In an embodiment of the present application, the basic influence coefficient of the non-keyword is determined according to the category of the non-keyword and the different basic influence coefficients corresponding to different categories, and then feature extraction processing is performed through a trained neural network model based on the category of the non-keyword, the distance between the non-keyword and the keyword, and the basic influence coefficient to determine the supplementary influence coefficient and the harmonization coefficient of the non-keyword on the keyword, which not only covers the characteristics of the non-keyword itself, but also integrates the correlation between the non-keyword and the keyword, so that the impact of the non-keyword on the task request can be determined more accurately, thereby improving the estimation accuracy of the total resource demand corresponding to the task request.

[0047] In one embodiment, according to the category of the non-keyword, the basic influence coefficient of the i-th non-keyword is determined as KTi; through the trained neural network model, the supplementary influence coefficient KAi of the i-th non-keyword on the keyword is determined, and the influence coefficient QNCF of the non-keyword on the task request can be determined by the following formula: where h is the harmonic coefficient. Assume the task request is: How many schools are there in place A? If only the keywords "place A" and "school" are considered, the resource requirements for this task request cannot be accurately estimated. By introducing non-keywords, including the basic influence coefficient, the category of the non-keyword "piece" is a quantifier, and the coefficient is 1, which is higher than the basic influence coefficients of other non-keywords, so as to accurately reflect the importance of this non-keyword to the task request. And including the supplementary influence coefficient of the non-keyword on the keyword, especially introducing the influence of the distance between the non-keyword and the keyword, can determine the degree of influence of each non-keyword on the keyword. For example, the distance between the non-keyword "piece" and the keyword "school" is closer than the distance between the non-keyword "piece" and the keyword "place A", it can be determined that this non-keyword "piece" acts on the keyword "school", and it is "schools in place A", thus more accurately reflecting the influence of this non-keyword on the entire task request.

[0048] In one implementation, before step 22 above, according to the category of the non-keyword, the position and order in the task request, through the trained neural network model, to determine the supplementary influence coefficient of the non-keyword on the keyword and the harmonic coefficient, the method may further include the following steps.

[0049] Step 24, generate a character sequence corresponding to the training statement according to the sample keyword, the category of the sample non-keyword, and the distance between the sample non-keyword and the sample keyword in the training statement.

[0050] Among them, before training the initial neural network model, determine the training samples, and generate a character sequence corresponding to the training statement by extracting the sample keyword, the category of the sample non-keyword, and the distance between the sample non-keyword and the sample keyword in the training statement.

[0051] For example, the training statement is: Flight from place A to place B. Among them, "from" is a verb, represented by the attribute value "a", "to" is a preposition, represented by the attribute value "b", and "of" is an adjective, represented by the attribute value "c". Then the character sequence corresponding to this training statement can be expressed as: {a; m1; b; m2; c; m3}, where mi is the i-th keyword.

[0052] Step 25: Train the initial neural network model based on the training sentence, the character sequence, the distance between the sample keywords and the sample non-keywords, the basic influence coefficient of the sample non-keywords on the training sentence, and the resource requirements corresponding to the sample keywords to obtain the trained neural network model.

[0053] Among them, the training sentences, character sequences, the distance between sample keywords and sample non-keywords, the basic influence coefficient of sample non-keywords on the training sentences, and the resource requirements corresponding to the sample keywords are input into the initial neural network model for model training. The convolution layer in the initial neural network model performs feature extraction according to the input content. By obtaining the coefficients, number of iterations and loss function constraints corresponding to each convolution layer, the initial neural network model is trained to obtain the trained neural network model, and the supplementary influence coefficient and harmonic coefficient of non-keywords on keywords are determined based on the trained neural network model.

[0054] In one implementation, the above step 1041 of performing word segmentation processing on the task request to obtain keywords and non-keywords may include the following steps.

[0055] Step 31: pre-process the task request to obtain a first text.

[0056] Preprocessing the task request may include removing irrelevant characters such as spaces and punctuation marks.

[0057] Step 32 : performing forward maximum matching and reverse maximum matching on the first text respectively, and obtaining a first word segmentation result corresponding to the forward maximum matching and a second word segmentation result corresponding to the reverse maximum matching.

[0058] The forward maximum match and reverse maximum match are implemented based on a pre-built dictionary containing keywords. The forward maximum match is performed by traversing the first text from left to right and segmenting the words in a maximum match manner. The reverse maximum match is performed by traversing the first text from right to left and segmenting the words in a maximum match manner.

[0059] In one implementation, performing a forward maximum match on the first text in step 32 to obtain a first word segmentation result corresponding to the forward maximum match may include:

[0060] Step 32a, based on a pre-built word segmentation dictionary, traverse each character of the first text in order from left to right, and match them with words in the word segmentation dictionary;

[0061] Step 32b: Determine the longest matching word that matches a word in the word segmentation dictionary as the first word segmentation result. Wherein, the word segmentation dictionary includes at least one preset keyword.

[0062] In an embodiment of the present application, a method for determining a first word segmentation result by forward maximum matching is provided, which can be implemented by the following steps: S1, setting a pointer to point to the starting position of the first text; S2, starting from the pointer position, taking characters backward one by one, matching them with words in the dictionary, and finding the longest matching word; S3, taking the matched word as a word segmentation result, and moving the pointer to the end position of the matching word; S4, repeating the above steps until the pointer points to the end position of the first text to obtain the first word segmentation result.

[0063] In one implementation, performing reverse maximum matching on the first text in step 32 to obtain a second word segmentation result corresponding to the reverse maximum matching may include:

[0064] Step 32c, based on a pre-built word segmentation dictionary, traverse each character of the first text in order from right to left, and match them with words in the word segmentation dictionary;

[0065] Step 32d: Determine the longest matching word that matches a word in the word segmentation dictionary as the second word segmentation result; wherein the word segmentation dictionary includes at least one preset keyword.

[0066] In an embodiment of the present application, a method for determining a first word segmentation result by reverse maximum matching is provided, which can be implemented by the following steps: S1, setting a pointer to point to the end position of the first text; S2, starting from the pointer position, taking characters forward one by one, matching them with words in the dictionary, and finding the longest matching word; S3, taking the matched word as a word segmentation result, and moving the pointer to the starting position of the matching word; S4, repeating the above steps until the pointer points to the starting position of the first text to obtain a second word segmentation result.

[0067] Step 33: Determine the keywords and the non-keywords according to the first word segmentation result and the second word segmentation result.

[0068] In an embodiment of the present application, the first word segmentation result and the second word segmentation result are obtained by forward maximum matching and reverse maximum matching, and the first word segmentation result and the second word segmentation result are compared. For each pair of similar word segmentation results, the result with the longer length is selected as the final word segmentation result, and the keywords can be obtained. The keywords are used as the segmentation, and the remaining words are non-keywords.

[0069] Optionally, the word segmentation algorithm may also include a Hidden Markov Model (HMM) and the like.

[0070] In one implementation, the total resource requirement corresponding to the task request includes at least one type of total resource requirement; and the available resources of the computing nodes include at least one type of available resources. Step 106, which allocates a target computing node to the pending task based on the total resource requirement corresponding to the task request and the available resources of each computing node, may include the following steps.

[0071] Step 1061: Determine at least one candidate computing power node that matches the total resource demand of the at least one type based on the total resource demand of the at least one type.

[0072] The at least one type of total resource requirement may include GPU resource requirements, CPU resource requirements, double-precision floating-point computing power resource requirements, memory requirements, and disk requirements, etc. The at least one type of available resource may include GPU resources, CPU resources, double-precision floating-point computing power resources, memory capacity, and disk capacity, etc.

[0073] Step 1062: According to the priority of the total resource demand of the at least one type, the candidate computing power node corresponding to the total resource demand with the highest priority is determined as the target computing power node.

[0074] In an embodiment of the present application, based on the total resource demand of at least one type, at least one candidate computing power node that matches the total resource demand of at least one type is determined respectively, and then based on the demand priority of the total resource demand of at least one type, the candidate computing power node corresponding to the total resource demand with the highest demand priority is determined as the target computing power node. Based on the demand priority, the target computing power node that better meets the task requirements can be allocated to the task to be processed, which also improves the flexibility of computing power scheduling. For example, the total resource demand corresponding to the task request includes the GPU resource demand of x1 and the CPU resource demand of x2. The candidate computing power node y1 is determined, whose GPU resources meet the GPU resource demand of x1, and the candidate computing power node y2 is determined, whose CPU resources meet the CPU resource demand of x2. Then, based on the demand priority of the total resource demand of the task request: the demand priority of the CPU resource demand is higher than the demand priority of the GPU resource demand, the candidate computing power node y2 corresponding to the CPU resource demand with a higher demand priority is determined as the target computing power node, so that the target computing power node is called to process the task to be processed, so that the CPU resources of the target computing power node can give priority to meeting the CPU resource demand in the task request.

[0075] In one implementation, the available resources of the target computing power node match the total resource requirements corresponding to the task request, including at least one of the following:

[0076] (1) The GPU resources of the target computing node match the GPU resource requirements corresponding to the task request;

[0077] (2) The CPU resources of the target computing node match the CPU resource requirements corresponding to the task request;

[0078] (3) The double-precision floating-point computing power resources of the target computing power node match the double-precision floating-point computing power resource requirements corresponding to the task request;

[0079] (4) The memory capacity of the target computing node matches the memory requirement corresponding to the task request;

[0080] (5) The disk capacity of the target computing node matches the disk requirement corresponding to the task request.

[0081] The matching of the available resources of the target computing power node with the total resource requirements corresponding to the task request can be a single condition or a multiple condition. In addition, the matching of the available resources of the target computing power node with the total resource requirements corresponding to the task request is not limited to the above five conditions.

[0082] In one implementation, after the above-mentioned step 104 determines the total resource requirements corresponding to the task request based on the resource requirements corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count characteristics of the task request, the method may further include the following steps.

[0083] Step 108 : determining the actual resource requirement of each keyword according to the total resource requirement corresponding to the task request and the number of keywords in the task request.

[0084] The actual resource requirement of each keyword can be obtained by averaging the total resource requirements corresponding to the task request.

[0085] Step 110 : updating the resource requirements corresponding to the keywords in the task request according to the actual resource requirements of each keyword, the historical resource requirements of each keyword, and the weight of each keyword in the historical task requests.

[0086] In an embodiment of the present application, the resource requirements corresponding to each keyword are stored in a database. In order to improve the accuracy of the resource requirements corresponding to each keyword, the resource requirements corresponding to each keyword stored in the database are updated after each estimation of the total resource requirements corresponding to the task request. The design resource requirements of each keyword are determined by the total resource requirements corresponding to the current task request, and then the resource requirements corresponding to each keyword are recalculated based on the historical resource requirements of each keyword and the weight of each keyword in the historical task request to achieve the update of the keyword. When the number of times increases to a certain value, the resource requirements corresponding to each keyword can be closer to the actual application situation, thereby improving the estimation accuracy of the total resource requirements corresponding to the task request.

[0087] Figure 2 This is a schematic diagram of the structure of a computing power scheduling system provided according to an embodiment of the present application. Figure 2 As shown, the system 200 includes: a computing power scheduling center 201, a resource estimation gateway 202, and a resource management center 203.

[0088] The computing power scheduling center 201 is used to obtain task requests for tasks to be processed; and the resource management center 203 is used to obtain available resources of each computing power node.

[0089] The resource estimation gateway 202 is used to determine the total resource requirement corresponding to the task request based on the resource requirement corresponding to the keywords in the task request, the influence coefficient of the non-keywords in the task request on the task request, and the word count feature of the task request.

[0090] The computing power scheduling center 201 is also used to allocate target computing power nodes to the tasks to be processed based on the total resource requirements corresponding to the task request and the available resources of each computing power node; wherein the available resources of the target computing power node match the total resource requirements corresponding to the task request.

[0091] In one implementation, the above-mentioned resource estimation gateway 202 can be used to perform word segmentation processing on the task request to obtain keywords and non-keywords; determine the influence coefficient of the non-keywords on the task request based on the basic influence coefficient of the non-keywords on the task request and the supplementary influence coefficient of the non-keywords on the keywords; determine the total resource demand corresponding to the task request based on the resource demand corresponding to the keywords, the influence coefficient, and the word count characteristics of the task request.

[0092] In one implementation, the above-mentioned resource estimation gateway 202 can be used to determine the basic influence coefficient of the non-keyword on the task request based on the category of the non-keyword; determine the supplementary influence coefficient and the harmonic coefficient of the non-keyword on the keyword through a trained neural network model based on the category of the non-keyword, the distance between the non-keyword and the keyword, and the basic influence coefficient; determine the influence coefficient of the non-keyword on the task request based on the basic influence coefficient, the supplementary influence coefficient, and the harmonic coefficient.

[0093] In one implementation, the resource estimation gateway 202 can also be used to generate a character sequence corresponding to the training sentence based on the sample keywords, the category of the sample non-keywords, and the distance between the sample non-keywords and the sample keywords in the training sentence; and train the initial neural network model based on the training sentence, the character sequence, the distance between the sample keywords and the sample non-keywords, the basic influence coefficient of the sample non-keywords on the training sentence, and the resource requirements corresponding to the sample keywords to obtain the trained neural network model.

[0094] In one implementation, the resource estimation gateway 202 can be used to preprocess the task request to obtain a first text; perform forward maximum matching and reverse maximum matching on the first text respectively to obtain a first word segmentation result corresponding to the forward maximum matching and a second word segmentation result corresponding to the reverse maximum matching; determine the keywords and the non-keywords based on the first word segmentation result and the second word segmentation result.

[0095] In one implementation, the above-mentioned computing power scheduling center 201 can also be used to determine the actual resource demand of each keyword based on the total resource demand corresponding to the task request and the number of keywords in the task request; and update the resource demand corresponding to the keyword in the task request based on the actual resource demand of each keyword, the historical resource demand of each keyword and the weight of each keyword in the historical task request.

[0096] Figure 3 This is a flow chart of another computing power scheduling method provided by an embodiment of the present application. This method can be applied to Figure 2 The computing power scheduling system shown in Figure 2. Figure 3 , the method may include the following steps.

[0097] Step 301: The computing power scheduling center receives a task request from a user / terminal and sends the task request to a resource estimation gateway.

[0098] Step 302: The resource estimation gateway analyzes the task request and determines the word count feature of the task request.

[0099] The word count feature includes the total word count QL of the task request.

[0100] In step 303, the resource estimation gateway determines the keywords, the number of keywords, and non-keywords for the task request through a word segmentation algorithm.

[0101] Among them, word segmentation can specifically include the following steps: Step 303a, preprocessing: preprocess the text to be segmented, including removing spaces, punctuation marks and other irrelevant characters. Step 303b, dictionary construction: based on the existing word segmentation dictionary, construct a dictionary containing keywords for matching word segmentation. Step 303c, forward maximum matching: traverse the text to be segmented from left to right, and perform word segmentation in a maximum matching manner. Step 303d, reverse maximum matching: traverse the text to be segmented from right to left, and perform word segmentation in a maximum matching manner. Step 303e, merging results: merge the results of forward maximum matching and reverse maximum matching, select the result with the longer matching length as the final word segmentation result, and obtain keywords, keyword word count CL, and non-keywords.

[0102] In step 304, the resource estimation gateway queries and obtains the resource requirements of each keyword from the keyword resource estimation database.

[0103] The keyword resource requirement QCR of the task request is obtained by accumulating the resource requirements of all keywords.

[0104] Step 305 : The resource estimation gateway determines the influence coefficient QNCF of each non-keyword on the task request based on the basic influence coefficient of each non-keyword and the supplementary influence coefficient of each keyword on the keyword.

[0105] The basic influence coefficient of each non-keyword is obtained by querying and acquiring from a preset non-keyword category and coefficient relationship table, and the supplementary influence coefficient of each non-keyword on the keyword is acquired through a trained neural network model.

[0106] In step 306, the resource estimation gateway determines the total resource demand QR corresponding to the task request based on the keyword resource demand QCR of the task request, the influence coefficient QNCF of each non-keyword on the task request, the total number of words in the task request QL, and the number of keyword words CL, and sends QR to the computing power scheduling center.

[0107] Among them, the total resource requirements corresponding to the task request

[0108] Step 307: The resource management center regularly collects the available resources NR of all computing nodes and sends the NR to the computing power scheduling center.

[0109] In step 308, the computing power scheduling center allocates a target computing power node to the user / terminal based on the total resource demand QR corresponding to the task request and the available resources NR of all computing power nodes, and forwards the task request to the target computing power node.

[0110] In step 309, the target computing power node receives the task request, executes the task, and returns the task execution result, duration, etc. to the computing power scheduling center.

[0111] In step 310 , the computing power scheduling center updates the resource requirements of the keywords in the keyword resource estimation database according to the total resource requirements QR corresponding to the task request and the task execution results.

[0112] In step 311, the computing power scheduling center returns the task execution result to the user / terminal.

[0113] Through the embodiments of the present application, the total resource demand corresponding to the task request is determined by the resource demand corresponding to the keywords of the task request and the influence coefficient of non-keywords on the task request, and based on the resource supply situation of the computing power node, a suitable computing power node is allocated to the task to be processed, so that the resources required for the task request are matched with the supply resources of the computing power node, the utilization of resources is optimized, and efficient scheduling of the computing power nodes is achieved.

[0114] Based on the same technical concept, an embodiment of the present application further provides an electronic device for executing the above-mentioned computing power scheduling method. Figure 4 A schematic diagram of the structure of an electronic device for implementing various embodiments of the present application. Electronic devices may have relatively large differences due to different configurations or performances, and may include a processor (processor) 401, a communication interface (Communications Interface) 402, a memory (memory) 403 and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404. The processor 401 can call a computer program stored in the memory 403 and can be run on the processor 401 to execute the various steps of the above-mentioned computing power scheduling method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0115] The above electronic device structure does not constitute a limitation of the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. For example, the input unit may include a graphics processing unit (GPU) and a microphone, and the display unit may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. to configure the display panel. The user input unit includes a touch panel and at least one of other input devices. The touch panel is also called a touch screen. Other input devices may include but are not limited to a physical keyboard, function keys (such as volume control buttons, switch buttons, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.

[0116] The memory can be used to store software programs and various data. The memory may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, applications or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory may include a volatile memory or a non-volatile memory, or the memory may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct rambus random access memory (DRRAM).

[0117] The processor may include one or more processing units; optionally, the processor may integrate an application processor and a modem processor, wherein the application processor primarily handles operations related to the operating system, user interface, and application programs, and the modem processor primarily processes wireless communication signals, such as a baseband processor. It is understood that the modem processor may not be integrated into the processor.

[0118] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned computing power scheduling method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0119] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0120] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned computing power scheduling method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0121] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0122] An embodiment of the present application also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes a program or instructions. When the program or instructions are executed, the various processes of the above-mentioned computing power scheduling method embodiment are implemented and can achieve the same technical effect. To avoid repetition, they will not be repeated here.

[0123] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0124] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can 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 a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0125] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A computing power scheduling method, It is characterized in that include: Obtain task requests for pending tasks and available resources of each computing node; Determine the total resource demand corresponding to the task request according to the resource demand corresponding to the keyword in the task request, the influence coefficient of the non-keyword in the task request on the task request, and the word count feature of the task request; Allocate a target computing node for the task to be processed according to the total resource demand corresponding to the task request and the available resources of each computing node; Among them, the available resources of the target computing power node match the total resource requirements corresponding to the task request.

2. The method according to claim 1, It is characterized in that The determining of the total resource requirement corresponding to the task request according to the resource requirement corresponding to the keyword in the task request, the influence coefficient of the non-keyword in the task request on the task request, and the word count feature of the task request includes: Perform word segmentation processing on the task request to obtain keywords and non-keywords; Determining an influence coefficient of the non-keyword on the task request according to a basic influence coefficient of the non-keyword on the task request and a supplementary influence coefficient of the non-keyword on the keyword; The total resource requirement corresponding to the task request is determined according to the resource requirement corresponding to the keyword, the influence coefficient, and the word count feature of the task request.

3. The method according to claim 2, It is characterized in that The determining the influence coefficient of the non-keyword on the task request according to the basic influence coefficient of the non-keyword on the task request and the supplementary influence coefficient of the non-keyword on the keyword includes: Determining a basic influence coefficient of the non-keyword on the task request according to the category of the non-keyword; According to the category of the non-keyword, the distance between the non-keyword and the keyword, and the basic influence coefficient, the supplementary influence coefficient and the harmonic coefficient of the non-keyword on the keyword are determined by a trained neural network model; The influence coefficient of the non-keyword on the task request is determined according to the basic influence coefficient, the supplementary influence coefficient, and the harmonic coefficient.

4. The method according to claim 3, It is characterized in that Before determining the supplementary influence coefficient and the harmonic coefficient of the non-keyword on the keyword through a trained neural network model according to the category of the non-keyword, the position and the order in the task request, the method further includes: Generate a character sequence corresponding to the training sentence according to the sample keywords, the categories of the sample non-keywords, and the distance between the sample non-keywords and the sample keywords in the training sentence; The initial neural network model is trained according to the training sentence, the character sequence, the distance between the sample keyword and the sample non-keyword, the basic influence coefficient of the sample non-keyword on the training sentence, and the resource demand corresponding to the sample keyword to obtain the trained neural network model.

5. The method according to claim 2, It is characterized in that The word segmentation process of the task request to obtain keywords and non-keywords includes: Preprocessing the task request to obtain a first text; Performing a forward maximum match and a reverse maximum match on the first text respectively, to obtain a first word segmentation result corresponding to the forward maximum match and a second word segmentation result corresponding to the reverse maximum match; The keywords and the non-keywords are determined according to the first word segmentation result and the second word segmentation result.

6. The method according to claim 5, It is characterized in that The performing a forward maximum match and a reverse maximum match on the first text respectively to obtain a first word segmentation result corresponding to the forward maximum match and a second word segmentation result corresponding to the reverse maximum match includes: Based on a pre-built word segmentation dictionary, traverse each character of the first text in order from left to right, and match them with words in the word segmentation dictionary; Determine the longest matching word that matches the word in the word segmentation dictionary as the first word segmentation result; Wherein, the word segmentation dictionary includes at least one preset keyword.

7. The method according to claim 5, It is characterized in that The performing a forward maximum match and a reverse maximum match on the first text respectively to obtain a first word segmentation result corresponding to the forward maximum match and a second word segmentation result corresponding to the reverse maximum match includes: Based on a pre-built word segmentation dictionary, traverse each word of the first text in order from right to left, and match them with words in the word segmentation dictionary; Determine the longest matching word that matches the word in the word segmentation dictionary as the second word segmentation result; Wherein, the word segmentation dictionary includes at least one preset keyword.

8. The method according to claim 1, It is characterized in that The total resource demand corresponding to the task request includes at least one type of total resource demand; the available resources of the computing power node include at least one type of available resources; and allocating a target computing power node for the task to be processed according to the total resource demand corresponding to the task request and the available resources of each computing power node, including: Determine, according to the at least one type of total resource demand, at least one candidate computing power node that matches the at least one type of total resource demand; According to the demand priority of the at least one type of total resource demand, the candidate computing power node corresponding to the total resource demand with the highest demand priority is determined as the target computing power node.

9. The method according to claim 1, It is characterized in that The available resources of the target computing power node match the total resource requirements corresponding to the task request, including at least one of the following: The GPU resources of the target computing power node match the GPU resource requirements corresponding to the task request; The CPU resource of the target computing power node matches the CPU resource requirement corresponding to the task request; The double-precision floating-point computing power resources of the target computing power node match the double-precision floating-point computing power resource requirements corresponding to the task request; The memory capacity of the target computing power node matches the memory requirement corresponding to the task request; The disk capacity of the target computing power node matches the disk requirement corresponding to the task request.

10. The method according to claim 1, It is characterized in that After determining the total resource demand corresponding to the task request according to the resource demand corresponding to the keyword in the task request, the influence coefficient of the non-keyword in the task request on the task request, and the word count feature of the task request, the method further includes: Determining the actual resource requirement of each keyword according to the total resource requirement corresponding to the task request and the number of keywords in the task request; The resource requirements corresponding to the keywords in the task request are updated according to the actual resource requirements of each keyword, the historical resource requirements of each keyword, and the weight of each keyword in the historical task requests.