Cluster scheduling method and device for algorithm chips

Abstract

The invention discloses a cluster scheduling method and device for algorithm chips, and the method comprises the steps: obtaining a scheduling task, chip function information and chip data information of an algorithm chip cluster, recognizing the algorithm chip cluster according to the scheduling task, determining function demand information, screening and determining a target algorithm chip corresponding to the function demand information according to the chip function information, and carrying out the cluster scheduling of the algorithm chip. And generating and executing chip scheduling information, obtaining feedback execution result information, determining an abnormal reason of the abnormal algorithm chip based on the abnormal scheduling information when determining that the abnormal scheduling information exists, and adjusting the corresponding chip scheduling information according to the abnormal reason. And executing the adjusted chip scheduling information until no abnormal scheduling information exists in a feedback execution result. Through the scheduling method, automatic scheduling of the algorithm chip cluster is realized, the integrity of normal execution of the scheduling information of the algorithm chip cluster is ensured, and the execution efficiency and the execution accuracy of the scheduling task are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of algorithm chip clusters, and in particular to a cluster scheduling method and device for algorithm chips. Background Art

[0002] Traditional algorithm master chip driving algorithm chip technology is generally based on the internal manual provided by the manufacturer, which is a technology for scheduling a single algorithm chip, but cannot schedule the algorithm chip cluster. Therefore, how to directly schedule the algorithm chip cluster is the current research focus.

[0003] The traditional scheduling method of algorithm chip clusters requires a large number of main control chips to schedule each algorithm chip one by one, so that the algorithm chip cluster can be scheduled. However, this method has high resource costs and long scheduling time, resulting in low scheduling efficiency of the algorithm chip cluster. Summary of the invention

[0004] The present invention provides a cluster scheduling method and device for an algorithm chip, so as to achieve the technical effect of improving the efficiency of algorithm chip cluster scheduling and simultaneously improving the efficiency of abnormality elimination in processing abnormal data of abnormal algorithm chips.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a cluster scheduling method for algorithm chips, comprising the following steps:

[0006] Obtaining a scheduling task of an algorithm chip cluster, as well as first chip function information and first chip data information of each algorithm chip, identifying the algorithm chip cluster based on the scheduling task, and obtaining first function requirement information;

[0007] Determine the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information screening, and generate the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip;

[0008] Scheduling the algorithm chip cluster according to the first chip scheduling information, and obtaining execution result information fed back by the algorithm chip cluster;

[0009] When the execution result information contains abnormal scheduling information, identifying the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information, and adjusting the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain second chip scheduling information;

[0010] The first chip scheduling information of the abnormal algorithm chip is updated according to the second chip scheduling information, and the algorithm chip cluster is scheduled according to the second chip scheduling information until the execution result information does not contain the abnormal scheduling information.

[0011] The algorithm chip cluster scheduling method provided by the present invention first obtains the scheduling task of the algorithm chip cluster, as well as the first chip function information and the first chip data information of each algorithm chip, that is, obtains the chip function information and the chip data information of all algorithm chips in the algorithm chip cluster that needs to be cluster scheduled, and identifies the algorithm chip cluster that needs to be scheduled based on the obtained scheduling task, obtains the functional requirement information of each algorithm chip, and provides data reference for the subsequent determination of the target scheduling chip of each algorithm chip to be scheduled.

[0012] After determining the functional requirement information of each algorithm chip, the system will determine the target algorithm chip corresponding to the functional requirement information of each chip based on the functional information of each chip, that is, according to the data information required for scheduling each algorithm chip, that is, according to the functional requirement information, determine the functional information that can realize the requirement, and determine the target scheduling chip that can schedule the algorithm chip based on the determined functional information, that is, the target algorithm chip. And after determining the target algorithm chip that can schedule the algorithm chip cluster, based on the data information of the target algorithm chip, generate chip scheduling information corresponding to the functional requirement information of the scheduling algorithm chip cluster, that is, the first chip scheduling information, to achieve automatic scheduling of the algorithm chip cluster.

[0013] After determining the first chip scheduling information for the algorithm chip cluster, the system will schedule the algorithm chip cluster according to the scheduling information and obtain the execution result information fed back by the chip cluster to provide a data basis for subsequent detection of abnormal scheduling data.

[0014] If it is determined that there is abnormal scheduling information in the execution result information, the system will identify the abnormal cause of the abnormal algorithm chip based on the abnormal scheduling information, that is, determine the specific information content of the abnormality of the algorithm chip, and adjust the scheduling information of the algorithm chip according to the determined abnormal cause to obtain the second chip scheduling information, that is, adjust the scheduling information of each algorithm chip in a timely manner to avoid the execution efficiency of the chip scheduling information being affected by temporary abnormal conditions such as chip operating status abnormality, freeze, operating bugs, etc., thereby ensuring the normal execution of each chip scheduling information, improving the efficiency of eliminating chip abnormal scheduling, and indirectly improving the scheduling efficiency of each algorithm chip.

[0015] After adjusting the scheduling information of the first chip to obtain the corresponding scheduling information of the second chip, the system will schedule the algorithm chip cluster according to the adjusted chip scheduling information until there is no abnormal scheduling information in the feedback information received by the system, thereby ensuring the integrity of the normal execution of the scheduling information of the algorithm chip cluster and improving the execution efficiency of its scheduling tasks.

[0016] As a preferred example, the identifying the algorithm chip cluster based on the scheduling task to obtain the first functional requirement information specifically includes:

[0017] Dividing the scheduling task into a plurality of scheduling subtasks, and determining task target information and task criticality of each scheduling subtask;

[0018] Identify and process the task target information to obtain target data information of each of the scheduled subtasks, and determine initial functional requirement information corresponding to the task target information based on the target data information;

[0019] The initial functional requirement information corresponding to the scheduling subtasks is sorted according to the task criticality to obtain the first functional requirement information.

[0020] Before the system determines the functional requirement information of each algorithm chip in the algorithm chip cluster, the system will also divide the acquired scheduling task into several scheduling subtasks, and determine the task target information and task criticality of each scheduling subtask, and obtain the target data information of each scheduling subtask by identifying the task target information, and then determine the initial functional requirement information corresponding to each task target information according to the target data information, and before outputting the functional requirement information, sort the initial functional requirements according to the task criticality, so that the execution priority of important tasks can be ensured when subsequent scheduling tasks are executed, avoiding the occurrence of subsequent abnormal tasks affecting the execution of critical tasks, and improving the execution efficiency of critical tasks.

[0021] As a preferred example, determining the initial functional requirement information corresponding to the task target information based on the target data information specifically includes:

[0022] Extracting key requirement features of the target task information according to the target requirement information in the target data information, and identifying and determining feature restriction ranges of the key requirement features according to the target condition information in the target data information;

[0023] The key requirement features and the corresponding feature restriction ranges are output as the initial functional requirement information of the task target information.

[0024] The system first extracts the key requirement features of the task target information based on the target requirement information in the target data information, and then identifies the feature restriction range corresponding to each key requirement feature based on the target condition information in the target data information. Then, each key requirement feature and the feature restriction range corresponding to each key requirement feature are output as the functional requirement information corresponding to the task target information, thereby improving the efficiency of identifying the target algorithm chip corresponding to the functional requirement information.

[0025] As a preferred example, the screening and determining the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information includes:

[0026] Extracting features from the first chip function information to obtain first function feature information, and identifying and determining a function operation range corresponding to the first function feature information based on the first chip data information of each algorithm chip, and integrating the first function feature information and the function operation range into first function operation information;

[0027] The corresponding target function information is determined based on the first function requirement information and the first function operation information, and the algorithm chip corresponding to the target function information is associated with the first function requirement information as the target algorithm chip and output.

[0028] After determining the functional requirement information of the algorithm chip cluster, the system will extract the functional feature information from the chip functional information of each algorithm chip, and identify the functional operation range corresponding to the functional feature information of each algorithm chip based on the chip data information of each algorithm chip, and integrate the above two into functional operation information. After integration, the system will screen and determine the scheduling chip corresponding to each algorithm chip based on the matching degree between the functional requirement information and the functional operation information, so as to determine the target algorithm chip.

[0029] As a preferred example, the screening and determining corresponding target function information according to the first function requirement information and the first function operation information includes:

[0030] Screening and acquiring the candidate algorithm chips, and screening and matching the first function operation information of each candidate algorithm chip according to the first function requirement information to obtain a corresponding matching result;

[0031] The first function operation information having the highest matching value with the first function requirement information in the matching results is output as the target function information.

[0032] The system provided by the present invention determines the target algorithm chip for scheduling each algorithm chip by determining the functional operation information with the highest matching degree with each functional requirement information. Since the system performs matching degree screening according to the criticality of the task, the above screening method is adopted to improve the priority screening degree of the functional requirement information with higher criticality, and avoids the functional operation information adapted to the functional requirement information with higher criticality, whose data processing capacity is less compatible with the functional requirement information, thereby affecting the data scheduling requirements of the functional requirement information with higher criticality, and ensuring the demand matching degree of the functional requirement information with higher mission criticality.

[0033] As a preferred example, the generating of the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip includes:

[0034] According to the first chip data information, the target algorithm chip is identified and processed to obtain corresponding second access data information, and according to the first functional requirement information, corresponding scheduling data information is obtained by querying in a preset scheduling database;

[0035] The scheduling data information is adjusted according to the second access data information to obtain the first chip scheduling information.

[0036] Before generating chip scheduling information for the algorithm chip cluster, the system also needs to identify the target algorithm chip through the chip data information of the chips in the algorithm chip cluster, obtain the access data information of the target algorithm chip, and then query the corresponding scheduling data information in the system's scheduling database according to the functional requirement information, and then adjust the scheduling data information according to the access data information of the target algorithm chip to obtain the first chip scheduling information.

[0037] The system adds the access data information and functional requirement information corresponding to the target algorithm chip to the scheduling data information to obtain the chip scheduling information corresponding to the functional requirement information. This not only improves the comprehensiveness and intelligence of the chip scheduling information and its matching accuracy with the target algorithm chip, but also indirectly improves the scheduling efficiency of each algorithm chip.

[0038] As a preferred example, the identifying the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information includes:

[0039] Identify and process the abnormal scheduling information, obtain chip operation status distribution information and data processing tasks, analyze the data processing tasks, and obtain task execution results and task execution time periods;

[0040] Based on the chip operation status distribution information, the chip operation status corresponding to each task execution period is queried and determined, and abnormal data processing tasks are screened out from the data processing tasks according to the task execution results and the chip operation status, and then the abnormal data processing tasks are determined as the abnormal cause of the abnormal algorithm chip.

[0041] When determining the cause of the chip abnormality based on the abnormal scheduling information, the system will first identify the chip operation status distribution information of the chip where the abnormality occurs, as well as the task execution results of each data processing task in the scheduling execution result information, and then query and determine the chip operation status of each task execution period in the chip operation status distribution information based on the task execution period of each data processing task. Subsequently, based on the chip operation status and task execution results, the system will jointly determine the abnormal data processing tasks in the abnormal algorithm chip, thereby improving the accuracy of identifying abnormal data processing tasks.

[0042] As a preferred example, the step of adjusting the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain the second chip scheduling information includes:

[0043] Calculate and obtain the real-time power consumption of the abnormal algorithm chip, compare the real-time power consumption with a preset power consumption threshold, and obtain a corresponding comparison result;

[0044] If the comparison result is that the real-time power consumption is less than or equal to the power consumption threshold, adjusting the first chip scheduling information according to the abnormal cause to obtain the second chip scheduling information;

[0045] If the comparison result is that the real-time power consumption is greater than the power consumption threshold, the abnormal cause is processed by abnormal scheduling according to the preset scheduling mechanism until the real-time power consumption is reduced to within the power consumption threshold range, and the first chip scheduling information after abnormal scheduling is adjusted to obtain the second chip scheduling information.

[0046] Before processing the abnormal data processing tasks in the abnormal algorithm chip, the system will also determine the real-time power consumption of each abnormal algorithm chip and compare it with the power consumption threshold to determine whether the real-time power consumption of each abnormal algorithm chip exceeds the power consumption threshold, that is, whether it is necessary to adjust the abnormal tasks in the abnormal algorithm chip to other chips for processing, so as to avoid the power consumption of the abnormal algorithm chip exceeding the threshold, resulting in the failure of the tasks that need to be executed therein, affecting the accuracy and completeness of the execution of the scheduling task.

[0047] As a preferred example, adjusting the first chip scheduling information according to the abnormal cause to obtain the second chip scheduling information specifically includes:

[0048] Screening the data processing tasks of the abnormal algorithm chip according to the task execution results, determining the data processing task whose task execution result is being executed as the first target data processing task, and determining the data processing task whose task execution result is not executed as the second target data processing task;

[0049] The processing order of the abnormal data processing tasks corresponding to the abnormal cause is adjusted to between the first target data processing task and the second target data processing task, and the second chip scheduling information is generated and output according to the adjusted processing order.

[0050] If it is determined that the remaining power consumption of the abnormal algorithm chip can still meet the processing of the corresponding abnormal data processing task, the processing order of the abnormal data processing task contained therein is adjusted to the next task of the currently being executed task, so as to avoid affecting the execution order of subsequent tasks. At the same time, it also ensures that high-priority abnormal tasks can be executed quickly, thereby improving the abnormal processing efficiency of the abnormal data processing tasks in each abnormal algorithm chip.

[0051] Accordingly, an embodiment of the present invention further provides a cluster scheduling device for an algorithm chip, the cluster scheduling device comprising an acquisition module, an adaptation module, a feedback module, an adjustment module and an iteration module; wherein:

[0052] The acquisition module is used to acquire the scheduling task of the algorithm chip cluster, as well as the first chip function information and the first chip data information of each algorithm chip, identify the algorithm chip cluster based on the scheduling task, and obtain the first function requirement information;

[0053] The adaptation module is used to screen and determine the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information, and generate the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip;

[0054] The feedback module is used to schedule the algorithm chip cluster according to the first chip scheduling information, and obtain the execution result information fed back by the algorithm chip cluster;

[0055] The adjustment module is used for, when the execution result information contains abnormal scheduling information, identifying the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information, and adjusting the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain the second chip scheduling information;

[0056] The iteration module is used to update the first chip scheduling information of the abnormal algorithm chip according to the second chip scheduling information, and schedule the algorithm chip cluster according to the second chip scheduling information until the execution result information does not contain the abnormal scheduling information. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 : A flow chart of an embodiment of the cluster scheduling method of the algorithm chip provided by the present invention;

[0058] Figure 2 : A structural diagram of an embodiment of a cluster scheduling device of the algorithm chip provided by the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Embodiment 1

[0061] Please refer to Figure 1 , which is a flow chart of an embodiment of the cluster scheduling method of the algorithm chip provided by the present invention, including steps 101 to 105, each step is specifically as follows:

[0062] Step 101: Obtain a scheduling task of an algorithm chip cluster, as well as first chip function information and first chip data information of each algorithm chip, identify the algorithm chip cluster based on the scheduling task, and obtain first function requirement information.

[0063] The algorithm chip cluster scheduling method provided in the embodiment of the present invention first obtains the scheduling task of the algorithm chip cluster, as well as the first chip function information and the first chip data information of each algorithm chip, that is, obtains the chip function information and the chip data information of all the algorithm chips in the algorithm chip cluster that needs to be cluster scheduled, and identifies the algorithm chip cluster that needs to be scheduled based on the obtained scheduling task, obtains the functional requirement information of each algorithm chip, and provides data reference for the subsequent determination of the target scheduling chip of each algorithm chip to be scheduled.

[0064] In this embodiment, the system responds to the user's scheduling task upload operation and obtains the scheduling task for the algorithm chip cluster. The scheduling task in this embodiment is the scheduling function requirement information for the algorithm chip cluster and the criticality information of each scheduling function requirement information, wherein the scheduling function requirement information is necessary information, and the criticality information is optional information, that is, the scheduling task must include the scheduling function requirement information, but not necessarily the criticality information.

[0065] The system directly uses the algorithm chip corresponding to each functional requirement information as the target algorithm chip of each functional requirement information, that is, determines the algorithm chip to perform the task according to the functional requirement information. At the same time, the system will also obtain the chip function information, chip specification information and security identification code of each algorithm chip, wherein the chip specification information and security identification code are the chip data information described in this embodiment.

[0066] Among them, the chip function information is the function information that the algorithm chip can run. The function information is the conventional operating function of the chip of the algorithm chip type, and the specific function information that each algorithm chip can run can be adjusted and determined by the user or staff.

[0067] The security identification code in the chip data information is specifically the access identification code of the algorithm chip when accessing each algorithm chip, which is used to limit the data information containing the access verification code of the algorithm chip from being input into the algorithm chip.

[0068] The chip specification information in the chip data information specifically includes data format information that can be processed by the algorithm chip, and the functional operating range of each operating function information of the algorithm chip.

[0069] Specifically, the method of identifying the algorithm chip cluster based on the scheduling task to obtain the first functional requirement information in this embodiment specifically includes:

[0070] Dividing the scheduling task into a plurality of scheduling subtasks, and determining task target information and task criticality of each scheduling subtask;

[0071] Identify and process the task target information to obtain target data information of each of the scheduled subtasks, and determine initial functional requirement information corresponding to the task target information based on the target data information;

[0072] The initial functional requirement information corresponding to the scheduling subtasks is sorted according to the task criticality to obtain the first functional requirement information.

[0073] Before the system determines the functional requirement information of each algorithm chip in the algorithm chip cluster, the system will also divide the acquired scheduling task into several scheduling subtasks, and determine the task target information and task criticality of each scheduling subtask, and obtain the target data information of each scheduling subtask by identifying the task target information, and then determine the initial functional requirement information corresponding to each task target information according to the target data information, and before outputting the functional requirement information, sort the initial functional requirements according to the task criticality, so that the execution priority of important tasks can be ensured when subsequent scheduling tasks are executed, avoiding the occurrence of subsequent abnormal tasks affecting the execution of critical tasks, and improving the execution efficiency of critical tasks.

[0074] Furthermore, the determining of the initial functional requirement information corresponding to the task target information based on the target data information in this embodiment specifically includes:

[0075] Extracting key requirement features of the target task information according to the target requirement information in the target data information, and identifying and determining feature restriction ranges of the key requirement features according to the target condition information in the target data information;

[0076] The key requirement features and the corresponding feature restriction ranges are output as the initial functional requirement information of the task target information.

[0077] The system first extracts the key requirement features of the task target information based on the target requirement information in the target data information, and then identifies the feature restriction range corresponding to each key requirement feature based on the target condition information in the target data information. Then, each key requirement feature and the feature restriction range corresponding to each key requirement feature are output as the functional requirement information corresponding to the task target information, thereby improving the efficiency of identifying the target algorithm chip corresponding to the functional requirement information.

[0078] In this embodiment, the system extracts key demand features of the task target information based on the target demand information, and the specific method may include:

[0079] 1. When the target demand information is text information, the system will extract the key demand features of the target demand information through the text feature recognition network.

[0080] 2. When the target demand information is non-text information, the system will query the identification information of the key demand features in the database, and query the target demand content corresponding to each identification information in the target demand information, and then convert the target demand content into a feature vector to obtain each key demand feature.

[0081] Specifically, the feature restriction range described in this embodiment is specifically the restriction range of the requirement content corresponding to each key requirement feature. For example, if the requirement content is data summation, data classification, or data identification, the restriction range of data summation is the range of the number of input data to be summed simultaneously, the restriction range of data classification is the range of types and numbers that can be classified, the restriction range of data identification is the range of data types that can be identified, etc.

[0082] Step 102: Screen and determine the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information, and generate the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip.

[0083] After determining the functional requirement information of each algorithm chip, the system will determine the target algorithm chip corresponding to the functional requirement information of each chip based on the functional information of each chip, that is, according to the data information required for scheduling each algorithm chip, that is, according to the functional requirement information, determine the functional information that can realize the requirement, and determine the target scheduling chip that can schedule the algorithm chip based on the determined functional information, that is, the target algorithm chip. And after determining the target algorithm chip that can schedule the algorithm chip cluster, based on the data information of the target algorithm chip, generate chip scheduling information corresponding to the functional requirement information of the scheduling algorithm chip cluster, that is, the first chip scheduling information, to achieve automatic scheduling of the algorithm chip cluster.

[0084] Specifically, the method of screening and determining the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information in this embodiment includes:

[0085] Extracting features from the first chip function information to obtain first function feature information, and identifying and determining a function operation range corresponding to the first function feature information based on the first chip data information of each algorithm chip, and integrating the first function feature information and the function operation range into first function operation information;

[0086] The corresponding target function information is determined based on the first function requirement information and the first function operation information, and the algorithm chip corresponding to the target function information is associated with the first function requirement information as the target algorithm chip and output.

[0087] After determining the functional requirement information of the algorithm chip cluster, the system will extract the functional feature information from the chip functional information of each algorithm chip, and identify the functional operation range corresponding to the functional feature information of each algorithm chip based on the chip data information of each algorithm chip, and integrate the above two into functional operation information. After integration, the system will screen and determine the scheduling chip corresponding to each algorithm chip based on the matching degree between the functional requirement information and the functional operation information, so as to determine the target algorithm chip.

[0088] In this embodiment, the system extracts the functional feature information of each chip functional information, and based on the chip specification information of each algorithm chip, identifies and determines the functional operation range of each functional feature information, and then screens the target functional information corresponding to each functional requirement information in order from front to back according to the ranking of each functional requirement information in the functional requirement criticality, and uses the algorithm chip of the target functional information as the target algorithm chip corresponding to the functional requirement information.

[0089] Furthermore, the step of screening and determining corresponding target function information according to the first function requirement information and the first function operation information in this embodiment includes:

[0090] Screening and acquiring the candidate algorithm chips, and screening and matching the first function operation information of each candidate algorithm chip according to the first function requirement information to obtain a corresponding matching result;

[0091] The first function operation information having the highest matching value with the first function requirement information in the matching results is output as the target function information.

[0092] The system provided by the embodiment of the present invention determines the target algorithm chip for scheduling each algorithm chip by determining the functional operation information with the highest matching degree with each functional requirement information. Since the system performs matching degree screening according to the criticality of the task, the above screening method is adopted to improve the priority screening degree of the functional requirement information with higher criticality, and avoids the functional operation information adapted to the functional requirement information with higher criticality, whose data processing capacity is less compatible with the functional requirement information, thereby affecting the data scheduling requirements of the functional requirement information with higher criticality, and ensures the demand matching degree of the functional requirement information with high mission criticality.

[0093] In this embodiment, the system uses algorithm chips that have not been determined as target algorithm chips as candidate algorithm chips, and for each functional requirement information, based on each key requirement feature of the functional requirement information, the feature restriction range corresponding to each key requirement feature, each functional feature information of the chip functional information of each candidate algorithm chip, and the functional operation range of each functional feature information of each chip functional information, the system screens each candidate chip functional information in the chip functional information, that is, determines the matching value between each candidate chip functional information and the functional requirement information.

[0094] Among them, the candidate chip function information described in this embodiment is specifically: functional feature information required to meet all key requirement features of the functional requirement information, and the functional operation range of each functional feature information is not less than the feature restriction range of the key requirement feature information corresponding to the functional feature information.

[0095] Furthermore, the method of identifying the functional feature information required for the key requirement feature is specifically as follows: the similarity between each key requirement feature and each functional feature information is calculated by a similarity recognition algorithm, and the functional feature information with a similarity greater than a similarity threshold preset in the terminal and corresponding to the maximum similarity is used as the functional feature information required for the key requirement feature. The similarity threshold can be adjusted and changed by the user or staff, for example, the similarity is 90%, 95%, 97% or 99%, etc.

[0096] The system can also calculate the feature restriction range corresponding to each key requirement feature of the functional requirement information and the average range deviation value between the functional operation range of each chip functional information, and then screen and determine the chip functional information with the smallest average range deviation value, and use it as the target chip functional information corresponding to the functional requirement information. The system provided in this embodiment screens all chip functional information with similar key requirement features, functional operation ranges, and functional restriction ranges as the target chip functional information corresponding to the functional requirement information, thereby improving the utilization rate of each algorithm chip while ensuring that the requirements of the functional requirement information can be met.

[0097] Specifically, the step of generating the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip in this embodiment includes:

[0098] According to the first chip data information, the target algorithm chip is identified and processed to obtain corresponding second access data information, and according to the first functional requirement information, corresponding scheduling data information is obtained by querying in a preset scheduling database;

[0099] The scheduling data information is adjusted according to the second access data information to obtain the first chip scheduling information.

[0100] Before generating chip scheduling information for the algorithm chip cluster, the system also needs to identify the target algorithm chip through the chip data information of the chips in the algorithm chip cluster, obtain the access data information of the target algorithm chip, and then query the corresponding scheduling data information in the system's scheduling database according to the functional requirement information, and then adjust the scheduling data information according to the access data information of the target algorithm chip to obtain the first chip scheduling information.

[0101] The system adds the access data information and functional requirement information corresponding to the target algorithm chip to the scheduling data information to obtain the chip scheduling information corresponding to the functional requirement information. This not only improves the comprehensiveness and intelligence of the chip scheduling information and its matching accuracy with the target algorithm chip, but also indirectly improves the scheduling efficiency of each algorithm chip.

[0102] In this embodiment, the system first generates an access verification code corresponding to each target algorithm chip based on the security identification code of each target algorithm chip, and identifies the data format information corresponding to each target algorithm chip based on the chip specification information of each target algorithm chip, and then queries the database for the scheduling data corresponding to the functional requirement information and the data processing task of the scheduling data for each functional requirement information, and outputs the scheduling data and the data processing task corresponding to the scheduling data as the scheduling data information of the functional requirement information, and then adjusts the scheduling data information based on the data format information of the target algorithm chip corresponding to the functional requirement information to obtain the initial chip scheduling information, i.e., the scheduling data information, and adds the access verification code corresponding to the target algorithm chip and the sequence position of the functional requirement information in the functional requirement sequence, i.e., the position of the task criticality sorting described in this embodiment, to the initial chip scheduling information to obtain the chip scheduling information corresponding to the functional requirement information.

[0103] Among them, the database stores the scheduling data corresponding to each functional requirement information in the historical scheduling process, as well as the correspondence between the data processing tasks of each scheduling data. Users or staff can update, improve, add and other operations according to needs, thereby increasing the number and types of functional requirement information, the number and types of scheduling data, and the number and types of data processing tasks.

[0104] Based on the data format information of the target algorithm chip corresponding to the functional requirement information, the system adjusts the data characteristics of the scheduling data in the scheduling data information to the data format information of the target algorithm chip, and obtains the initial chip scheduling information, that is, the scheduling data information described in this embodiment, and adds the access verification code corresponding to the target algorithm chip and the sequence position of the functional requirement information in the functional requirement sequence to the initial chip scheduling information, and obtains the chip scheduling information corresponding to the functional requirement information.

[0105] Step 103: Schedule the algorithm chip cluster according to the first chip scheduling information, and obtain execution result information fed back by the algorithm chip cluster.

[0106] After determining the first chip scheduling information for the algorithm chip cluster, the system will schedule the algorithm chip cluster according to the scheduling information and obtain the execution result information fed back by the chip cluster to provide a data basis for subsequent detection of abnormal scheduling data.

[0107] In this embodiment, the system generates chip scheduling information corresponding to each functional requirement information based on the chip specification information and security identification code of each algorithm chip, so that when each chip scheduling information is transmitted to the algorithm chip cluster, there is no need to establish a separate transmission channel for each algorithm chip separately. All chip scheduling information can be directly transmitted to the algorithm chip cluster through a single channel, and access requests are made to each algorithm chip one by one through the access verification code in each chip scheduling information, so that each chip scheduling information is transmitted to the target algorithm chip corresponding to the chip scheduling information.

[0108] This method not only reduces the number of scheduling channels and saves data transmission costs during scheduling, but also improves the accuracy of data transmission through access verification. By generating scheduling information for each algorithm chip based on the chip specification information of each algorithm chip, the chip mobilization efficiency and chip mobilization success rate of the algorithm chip are improved.

[0109] Step 104: When the execution result information contains abnormal scheduling information, the abnormal cause of the corresponding abnormal algorithm chip is identified based on the abnormal scheduling information, and the first chip scheduling information of the abnormal algorithm chip is adjusted according to the abnormal cause to obtain the second chip scheduling information.

[0110] If it is determined that there is abnormal scheduling information in the execution result information, the system will identify the abnormal cause of the abnormal algorithm chip based on the abnormal scheduling information, that is, determine the specific information content of the abnormality of the algorithm chip, and adjust the scheduling information of the algorithm chip according to the determined abnormal cause to obtain the second chip scheduling information, that is, adjust the scheduling information of each algorithm chip in a timely manner to avoid the execution efficiency of the chip scheduling information being affected by temporary abnormal conditions such as chip operating status abnormality, freeze, operating bugs, etc., thereby ensuring the normal execution of each chip scheduling information, improving the efficiency of eliminating chip abnormal scheduling, and indirectly improving the scheduling efficiency of each algorithm chip.

[0111] Specifically, the method of identifying the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information in this embodiment includes:

[0112] Identify and process the abnormal scheduling information, obtain chip operation status distribution information and data processing tasks, analyze the data processing tasks, and obtain task execution results and task execution time periods;

[0113] Based on the chip operation status distribution information, the chip operation status corresponding to each task execution period is queried and determined, and abnormal data processing tasks are screened out from the data processing tasks according to the task execution results and the chip operation status, and then the abnormal data processing tasks are determined as the abnormal cause of the abnormal algorithm chip.

[0114] When determining the cause of the chip abnormality based on the abnormal scheduling information, the system will first identify the chip operation status distribution information of the chip where the abnormality occurs, as well as the task execution results of each data processing task in the scheduling execution result information, and then query and determine the chip operation status of each task execution period in the chip operation status distribution information based on the task execution period of each data processing task. Subsequently, based on the chip operation status and task execution results, the system will jointly determine the abnormal data processing tasks in the abnormal algorithm chip, thereby improving the accuracy of identifying abnormal data processing tasks.

[0115] Among them, this embodiment does not further limit the task execution order of each data processing task. The user can determine it according to the data processing priority strategy of each algorithm chip, and the data processing priority strategy can be the algorithm chip's own strategy, or it can be a different data processing priority strategy for each algorithm chip set manually, and each algorithm chip can only execute a single data processing task in the same time period.

[0116] Furthermore, in this embodiment, adjusting the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain the second chip scheduling information includes:

[0117] Calculate and obtain the real-time power consumption of the abnormal algorithm chip, compare the real-time power consumption with a preset power consumption threshold, and obtain a corresponding comparison result;

[0118] If the comparison result is that the real-time power consumption is less than or equal to the power consumption threshold, adjusting the first chip scheduling information according to the abnormal cause to obtain the second chip scheduling information;

[0119] If the comparison result is that the real-time power consumption is greater than the power consumption threshold, the abnormal cause is processed by abnormal scheduling according to the preset scheduling mechanism until the real-time power consumption is reduced to within the power consumption threshold range, and the first chip scheduling information after abnormal scheduling is adjusted to obtain the second chip scheduling information.

[0120] Before processing the abnormal data processing tasks in the abnormal algorithm chip, the system will also determine the real-time power consumption of each abnormal algorithm chip and compare it with the power consumption threshold to determine whether the real-time power consumption of each abnormal algorithm chip exceeds the power consumption threshold, that is, whether it is necessary to adjust the abnormal tasks in the abnormal algorithm chip to other chips for processing, so as to avoid the power consumption of the abnormal algorithm chip exceeding the threshold, resulting in the failure of the tasks that need to be executed therein, affecting the accuracy and completeness of the execution of the scheduling task.

[0121] In this embodiment, the scheduling mechanism includes hot migration or locking, and the preset scheduling mechanism is: hot migration of abnormal scheduling tasks with a general importance level in the algorithm chip whose real-time power consumption exceeds the preset power consumption threshold to the algorithm chip with the lowest current power consumption in the non-overloaded algorithm chip, and monitoring the real-time power consumption of the algorithm chip whose real-time power consumption exceeds the preset power consumption threshold. The importance level may include multiple levels such as unimportant, general, important, and very important.

[0122] If the real-time power consumption of the algorithm chip whose real-time power consumption exceeds the preset power consumption threshold has been reduced to within the corresponding preset power consumption threshold range, hot migration of the abnormal data processing task is stopped;

[0123] Otherwise, the system first transfers the abnormal scheduling tasks with an importance level of important from the algorithm chip whose real-time power consumption exceeds the preset power consumption threshold to the algorithm chip that is not overloaded, and executes the scheduling tasks in the algorithm chip with the lowest current power consumption, and continues to monitor the real-time power consumption of the algorithm chip whose real-time power consumption exceeds the preset power consumption threshold;

[0124] Secondly, among the algorithm chips whose real-time power consumption exceeds the preset power consumption threshold, the system will lock the GPU frequencies of the chips with the most abnormal scheduling tasks with the importance level of "very important" one by one from high to low according to the corresponding real-time power consumption, until the real-time power consumption of the algorithm chips whose real-time power consumption exceeds the preset power consumption threshold is reduced to the corresponding preset power consumption threshold range.

[0125] Furthermore, in this embodiment, adjusting the first chip scheduling information according to the abnormal cause to obtain the second chip scheduling information specifically includes:

[0126] Screening the data processing tasks of the abnormal algorithm chip according to the task execution results, determining the data processing task whose task execution result is being executed as the first target data processing task, and determining the data processing task whose task execution result is not executed as the second target data processing task;

[0127] The processing order of the abnormal data processing tasks corresponding to the abnormal cause is adjusted to between the first target data processing task and the second target data processing task, and the second chip scheduling information is generated and output according to the adjusted processing order.

[0128] If it is determined that the remaining power consumption of the abnormal algorithm chip can still meet the processing of the corresponding abnormal data processing task, the processing order of the abnormal data processing task contained therein is adjusted to the next task of the currently being executed task, so as to avoid affecting the execution order of subsequent tasks. At the same time, it also ensures that high-priority abnormal tasks can be executed quickly, thereby improving the abnormal processing efficiency of the abnormal data processing tasks in each abnormal algorithm chip.

[0129] In this embodiment, for each abnormal algorithm chip, the system screens and determines the data processing task whose task execution result is not completing the current task in the task execution results of the corresponding data processing task, and uses it as the target data processing task, and then arranges the abnormal data processing task after the execution sequence of each target data processing task, that is, arranges the abnormal data processing task after the task currently being processed, that is, before the task that has not been processed. The system adjusts the execution order of the abnormal data processing task, so as to re-run the data processing task without affecting the normal execution of other data processing tasks, thereby ensuring that all data processing tasks are executed normally.

[0130] Step 105: Update the first chip scheduling information of the abnormal algorithm chip according to the second chip scheduling information, and schedule the algorithm chip cluster according to the second chip scheduling information until the execution result information does not contain the abnormal scheduling information.

[0131] After adjusting the scheduling information of the first chip to obtain the corresponding scheduling information of the second chip, the system will schedule the algorithm chip cluster according to the adjusted chip scheduling information until there is no abnormal scheduling information in the feedback information received by the system, thereby ensuring the integrity of the normal execution of the scheduling information of the algorithm chip cluster and improving the execution efficiency of its scheduling tasks.

[0132] In order to better illustrate the working principle and step flow of a cluster scheduling method and device for an algorithm chip of the present invention, reference may be made to, but is not limited to, the relevant records above.

[0133] Accordingly, see Figure 2 , Figure 2 This is a structural diagram of an embodiment of the cluster scheduling device of the algorithm chip provided by the present invention. Figure 2 As shown, the cluster scheduling device includes an acquisition module 201, an adaptation module 202, a feedback module 203, an adjustment module 204 and an iteration module 205; wherein:

[0134] The acquisition module 201 is used to acquire the scheduling task of the algorithm chip cluster, as well as the first chip function information and the first chip data information of each algorithm chip, identify the algorithm chip cluster based on the scheduling task, and obtain the first function requirement information.

[0135] Furthermore, the acquisition module 201 identifies the algorithm chip cluster based on the scheduling task to obtain the first functional requirement information, which specifically includes:

[0136] Dividing the scheduling task into a plurality of scheduling subtasks, and determining task target information and task criticality of each scheduling subtask;

[0137] Identify and process the task target information to obtain target data information of each of the scheduled subtasks, and determine initial functional requirement information corresponding to the task target information based on the target data information;

[0138] The initial functional requirement information corresponding to the scheduling subtasks is sorted according to the task criticality to obtain the first functional requirement information.

[0139] Furthermore, the acquisition module 201 determines the initial functional requirement information corresponding to the task target information based on the target data information, specifically including:

[0140] Extracting key requirement features of the target task information according to the target requirement information in the target data information, and identifying and determining feature restriction ranges of the key requirement features according to the target condition information in the target data information;

[0141] The key requirement features and the corresponding feature restriction ranges are output as the initial functional requirement information of the task target information.

[0142] The adaptation module 202 is used to screen and determine the target algorithm chip corresponding to each first functional requirement information based on the first chip functional information, and generate the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip.

[0143] Furthermore, the adaptation module 202 screens and determines the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information, including:

[0144] Extracting features from the first chip function information to obtain first function feature information, and identifying and determining a function operation range corresponding to the first function feature information based on the first chip data information of each algorithm chip, and integrating the first function feature information and the function operation range into first function operation information;

[0145] The corresponding target function information is determined based on the first function requirement information and the first function operation information, and the algorithm chip corresponding to the target function information is associated with the first function requirement information as the target algorithm chip and output.

[0146] Furthermore, the adaptation module 202 screens and determines corresponding target function information according to the first function requirement information and the first function operation information, including:

[0147] Screening and acquiring the candidate algorithm chips, and screening and matching the first function operation information of each candidate algorithm chip according to the first function requirement information to obtain a corresponding matching result;

[0148] The first function operation information having the highest matching value with the first function requirement information in the matching results is output as the target function information.

[0149] Further, the adaptation module 202 generates first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip, including:

[0150] According to the first chip data information, the target algorithm chip is identified and processed to obtain corresponding second access data information, and according to the first functional requirement information, corresponding scheduling data information is obtained by querying in a preset scheduling database;

[0151] The scheduling data information is adjusted according to the second access data information to obtain the first chip scheduling information.

[0152] The feedback module 203 is used to schedule the algorithm chip cluster according to the first chip scheduling information, and obtain the execution result information fed back by the algorithm chip cluster.

[0153] The adjustment module 204 is used to identify the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information when the execution result information contains abnormal scheduling information, and adjust the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain the second chip scheduling information.

[0154] Further, the adjustment module 204 identifies the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information, including:

[0155] Identify and process the abnormal scheduling information, obtain chip operation status distribution information and data processing tasks, analyze the data processing tasks, and obtain task execution results and task execution time periods;

[0156] Based on the chip operation status distribution information, the chip operation status corresponding to each task execution period is queried and determined, and abnormal data processing tasks are screened out from the data processing tasks according to the task execution results and the chip operation status, and then the abnormal data processing tasks are determined as the abnormal cause of the abnormal algorithm chip.

[0157] Further, the adjustment module 204 adjusts the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain the second chip scheduling information, including:

[0158] Calculate and obtain the real-time power consumption of the abnormal algorithm chip, compare the real-time power consumption with a preset power consumption threshold, and obtain a corresponding comparison result;

[0159] If the comparison result is that the real-time power consumption is less than or equal to the power consumption threshold, adjusting the first chip scheduling information according to the abnormal cause to obtain the second chip scheduling information;

[0160] If the comparison result is that the real-time power consumption is greater than the power consumption threshold, the abnormal cause is processed by abnormal scheduling according to the preset scheduling mechanism until the real-time power consumption is reduced to within the power consumption threshold range, and the first chip scheduling information after abnormal scheduling is adjusted to obtain the second chip scheduling information.

[0161] Furthermore, the adjusting module 204 adjusts the first chip scheduling information according to the abnormal reason to obtain the second chip scheduling information, which specifically includes:

[0162] Screening the data processing tasks of the abnormal algorithm chip according to the task execution results, determining the data processing task whose task execution result is being executed as the first target data processing task, and determining the data processing task whose task execution result is not executed as the second target data processing task;

[0163] The processing order of the abnormal data processing tasks corresponding to the abnormal cause is adjusted to between the first target data processing task and the second target data processing task, and the second chip scheduling information is generated and output according to the adjusted processing order.

[0164] The iteration module 205 is used to update the first chip scheduling information of the abnormal algorithm chip according to the second chip scheduling information, and schedule the algorithm chip cluster according to the second chip scheduling information until the execution result information does not contain the abnormal scheduling information.

[0165] In summary, the embodiment of the present invention provides a cluster scheduling method and device for algorithm chips, which obtains the scheduling tasks, chip function information and chip data information of the algorithm chip cluster, and identifies the algorithm chip cluster according to the scheduling tasks to determine the function requirement information, and screens and determines the target algorithm chip corresponding to the function requirement information according to the chip function information, and then generates and executes the chip scheduling information, obtains the feedback execution result information, and when it is determined that there is abnormal scheduling information, determines the abnormal cause of the abnormal algorithm chip based on the scheduling abnormality information, adjusts the corresponding chip scheduling information according to the abnormal cause, and executes the adjusted chip scheduling information until there is no abnormal scheduling information in the feedback execution result. The above scheduling method realizes the automatic scheduling of the algorithm chip cluster, ensures the integrity of the normal execution of the scheduling information of the algorithm chip cluster, and improves the execution efficiency of the scheduling task and the accuracy of the scheduling task execution.

[0166] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cluster scheduling method for algorithm chips, characterized in that: The following steps are involved: Obtaining a scheduling task of an algorithm chip cluster, as well as first chip function information and first chip data information of each algorithm chip, identifying the algorithm chip cluster based on the scheduling task, and obtaining first function requirement information; Screening and determining the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information, and generating the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip; Scheduling the algorithm chip cluster according to the first chip scheduling information, and obtaining execution result information fed back by the algorithm chip cluster; When the execution result information contains abnormal scheduling information, identifying the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information, and adjusting the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain second chip scheduling information; The first chip scheduling information of the abnormal algorithm chip is updated according to the second chip scheduling information, and the algorithm chip cluster is scheduled according to the second chip scheduling information until the execution result information does not contain the abnormal scheduling information.

2. The cluster scheduling method of an algorithm chip according to claim 1, characterized in that: The identifying the algorithm chip cluster based on the scheduling task to obtain the first functional requirement information specifically includes: Dividing the scheduling task into a plurality of scheduling subtasks, and determining task target information and task criticality of each scheduling subtask; Identify and process the task target information to obtain target data information of each of the scheduled subtasks, and determine initial functional requirement information corresponding to the task target information based on the target data information; The initial functional requirement information corresponding to the scheduling subtasks is sorted according to the task criticality to obtain the first functional requirement information.

3. The cluster scheduling method of an algorithm chip according to claim 2, characterized in that: The determining, based on the target data information, the initial functional requirement information corresponding to the task target information specifically includes: Extracting key requirement features of the target task information according to the target requirement information in the target data information, and identifying and determining feature restriction ranges of the key requirement features according to the target condition information in the target data information; The key requirement features and the corresponding feature restriction ranges are output as the initial functional requirement information of the task target information.

4. The cluster scheduling method of an algorithm chip according to claim 1, characterized in that: The screening and determining the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information includes: Extracting features from the first chip function information to obtain first function feature information, and identifying and determining a function operation range corresponding to the first function feature information based on the first chip data information of each algorithm chip, and integrating the first function feature information and the function operation range into first function operation information; The corresponding target function information is screened and determined according to the first function requirement information and the first function operation information, and the algorithm chip corresponding to the target function information is associated with the first function requirement information as the target algorithm chip and output.

5. The cluster scheduling method of an algorithm chip according to claim 4, characterized in that: The screening and determining corresponding target function information according to the first function requirement information and the first function operation information includes: Screening and acquiring candidate algorithm chips, and screening and matching the first function operation information of each candidate algorithm chip according to the first function requirement information to obtain corresponding matching results; The first function operation information having the highest matching value with the first function requirement information in the matching results is output as the target function information.

6. The cluster scheduling method of an algorithm chip according to claim 1, characterized in that: The generating the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip includes: According to the first chip data information, the target algorithm chip is identified and processed to obtain corresponding second access data information, and according to the first functional requirement information, corresponding scheduling data information is obtained by querying in a preset scheduling database; The scheduling data information is adjusted according to the second access data information to obtain the first chip scheduling information.

7. The cluster scheduling method of an algorithm chip according to claim 1, characterized in that: The identifying the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information includes: Identify and process the abnormal scheduling information, obtain chip operation status distribution information and data processing tasks, analyze the data processing tasks, and obtain task execution results and task execution time periods; Based on the chip operation status distribution information, the chip operation status corresponding to each task execution period is queried and determined, and abnormal data processing tasks are screened out from the data processing tasks according to the task execution results and the chip operation status, and then the abnormal data processing tasks are determined as the abnormal cause of the abnormal algorithm chip.

8. The cluster scheduling method of an algorithm chip according to claim 1, characterized in that: The step of adjusting the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain second chip scheduling information includes: Calculate and obtain the real-time power consumption of the abnormal algorithm chip, compare the real-time power consumption with a preset power consumption threshold, and obtain a corresponding comparison result; If the comparison result is that the real-time power consumption is less than or equal to the power consumption threshold, adjusting the first chip scheduling information according to the abnormal cause to obtain the second chip scheduling information; If the comparison result is that the real-time power consumption is greater than the power consumption threshold, the abnormal cause is processed by abnormal scheduling according to the preset scheduling mechanism until the real-time power consumption is reduced to within the power consumption threshold range, and the first chip scheduling information after abnormal scheduling is adjusted to obtain the second chip scheduling information.

9. The cluster scheduling method of an algorithm chip according to claim 8, characterized in that: The adjusting the first chip scheduling information according to the abnormal cause to obtain the second chip scheduling information specifically includes: Screening the data processing tasks of the abnormal algorithm chip according to the task execution results, determining the data processing task whose task execution result is being executed as the first target data processing task, and determining the data processing task whose task execution result is not executed as the second target data processing task; The processing order of the abnormal data processing tasks corresponding to the abnormal cause is adjusted to between the first target data processing task and the second target data processing task, and the second chip scheduling information is generated and output according to the adjusted processing order.

10. A cluster scheduling device for algorithm chips, characterized in that: The cluster scheduling device includes an acquisition module, an adaptation module, a feedback module, an adjustment module and an iteration module; wherein: The acquisition module is used to acquire the scheduling task of the algorithm chip cluster, as well as the first chip function information and the first chip data information of each algorithm chip, identify the algorithm chip cluster based on the scheduling task, and obtain the first function requirement information; The adaptation module is used to screen and determine the target algorithm chip corresponding to each of the first functional requirement information based on the first chip functional information, and generate the first chip scheduling information corresponding to the first functional requirement information based on the second chip data information of the target algorithm chip; The feedback module is used to schedule the algorithm chip cluster according to the first chip scheduling information, and obtain the execution result information fed back by the algorithm chip cluster; The adjustment module is used for, when the execution result information contains abnormal scheduling information, identifying the abnormal cause of the corresponding abnormal algorithm chip based on the abnormal scheduling information, and adjusting the first chip scheduling information of the abnormal algorithm chip according to the abnormal cause to obtain the second chip scheduling information; The iteration module is used to update the first chip scheduling information of the abnormal algorithm chip according to the second chip scheduling information, and schedule the algorithm chip cluster according to the second chip scheduling information until the execution result information does not contain the abnormal scheduling information.

Images