67results about How to "Avoid performance bottlenecks" patented technology

The invention discloses a multi-bus decoupling distributed embedded hardware rendering architecture, system and method, and relates to the technical field of ultra-high-definition video hardware processing, and the key points of the technical scheme are that the architecture comprises a multi-bus decoupling module which comprises a plurality of hardware buses which are physically isolated from one another, each hardware bus is an exclusive hardware transmission link bearing a single-type manufacturing task; the modular processing unit is an independent hardware board card and is integrated with an external signal input port, an upstream bus signal input port and a downstream bus signal output port; the plurality of modular processing units are cascaded in sequence through a hardware bus in the multi-bus decoupling module to form a distributed assembly line rendering link; when the modular processing units are newly added, the signal processing capability and the rendering synthesis capability are linearly improved along with the increase of the number of the modular processing units. According to the invention, the problem of performance bottleneck of the traditional architecture in processing 4K / 8K ultra-high-definition and multichannel rendering is fundamentally solved.

The invention discloses a hospital-oriented software and hardware collaborative medical data hierarchical processing method and device. According to the method, in a hospital intranet, statistics is carried out on the flow of medical data flow based on a source IP and a target IP two-tuple through a hardware side, a large flow and a small flow of the medical data flow are distinguished according to a dynamic threshold value, and an identifier is added to the large flow. The hardware flow table module carries out short-period quintuple statistics on the identified large-flow medical data message and reports a result with a timestamp; and the software flow table module performs long-period statistics on the small-flow medical data messages, synchronously fuses results of the two parties through timestamps, generates a full-amount medical network flow table, and further optimizes a dynamic threshold value and a processing strategy. According to the invention, through layered flow establishment and double-cycle cooperation, high performance of hardware processing and flexible large capacity of software processing are both considered, the utilization rate of hospital network resources and the consistency of flow data are effectively improved, and accurate support can be provided for hospital intranet monitoring, service awareness, safety protection and strategy automation.

The invention discloses a semi-supervised adaptive unstructured scene segmentation method, and relates to the technical field of computer vision. The method comprises the following steps: acquiring label-free image data of an unstructured scene; inputting the weakly enhanced image into a teacher model to generate a first prediction result; inputting the strong enhancement image into the student model to generate a second prediction result; calculating the JS divergence between the first prediction result and the second prediction result, and dynamically adjusting the attenuation coefficient of the index moving average according to the JS divergence; updating teacher model parameters according to the adjusted index moving average attenuation coefficient to obtain an unstructured scene segmentation model; and inputting a to-be-predicted unstructured scene image into the trained unstructured scene segmentation model, and outputting an unstructured scene segmentation result. According to the method, the teacher model weight updating rate is reduced in the model divergence significant stage such as the initial training stage or the difficult sample processing stage so as to suppress noise propagation, and the prediction precision of unstructured scene segmentation is effectively improved.

The embodiment of the invention provides a node graph rendering method and related equipment. The node graph rendering method comprises the following steps: monitoring the number of nodes contained in a node graph; if the number of the nodes does not exceed a preset node number threshold value, using a real-time rendering mode to render the node graph; and if the number of the nodes exceeds a preset node number threshold value, rendering the node graph by using a background rendering mode. According to the technical scheme, the number of the nodes in the node graph is monitored, real-time rendering is forbidden when the number of the nodes exceeds the preset node number threshold value, background rendering is used instead, the final layout of all the nodes is calculated through a background program, and after calculation is completed, the layouts of all the nodes are rendered into the page at a time, so that the layout of all the nodes is rendered into the page at a time. According to the method, frequent calculation and rendering are reduced, and the performance bottleneck of real-time rendering is avoided, so that the page loading speed and the user experience are remarkably improved.

The invention provides an energy efficiency optimization algorithm for a low-power wide area network (LPWAN), and aims to reduce the energy consumption of LPWAN equipment in a data transmission process and prolong the service life through adaptive data rate adjustment and a micro machine learning (TinyML) technology. The algorithm core comprises a self-adaptive data rate adjustment mechanism, and the transmission rate is dynamically adjusted according to the real-time link quality index and the signal strength. A TinyML model is integrated for local data processing and intelligent decision making, and the data uploading amount is reduced; according to the intelligent dormancy awakening strategy, energy consumption use of equipment is predicted and optimized according to an equipment use mode; and a multi-parameter optimization scheduling algorithm is adopted, and signal quality, equipment electric quantity and environmental factors are integrated for scheduling. In addition, the security of data transmission and user privacy protection are also considered, and a lightweight encryption technology is adopted. The practical application test in various LPWAN technologies verifies the effectiveness of the algorithm, realizes the obvious reduction of energy consumption and the prolonging of the service life of equipment, and has high expandability and adaptability.

The invention relates to an image processing method, a map generation method, electronic equipment and a storage medium. The image processing method comprises the following steps: extracting image features from an environment image to obtain a feature map of the environment image; segmenting the feature map through at least two different types of geometric element models to obtain different types of geometric element maps; fusing the geometric element diagrams of different categories to obtain a first fused diagram; and performing vectorization processing on geometric elements in the first fusion image to obtain a structured vector graph corresponding to the environment image.

The application relates to the computer technical field and discloses a modeling method of a statistical mixed model in a big data distributed scene. The method comprises the following steps: distributing and storing data shards; initializing model parameters; iteratively performing an expectation step and a maximization step; the expectation step is scheduled to a GPU node to perform parallel calculation on posterior probability; the maximization step is scheduled to a CPU node to aggregate statistics and update parameters, perform component merging / deletion, and perform convergence judgment; meanwhile, a memory reuse mechanism based on reference counting and scope analysis is adopted to reduce redundant data transmission. Through heterogeneous task scheduling and memory collaborative optimization, the training speed, resource utilization rate, and model self-adaptation capability are improved.

The application discloses an intelligent examination data processing system and method based on permission integration and asynchronous statistics, and relates to the field of computer software.The system comprises a teacher terminal, a student terminal and a server end;the server end realizes atomicity verification of commercial permission through a permission control subsystem;the whole process safety is guaranteed through a dynamic token chain;high-concurrent requests and data processing are decoupled through a message queue and an asynchronous processing engine;real-time dynamic statistics of question granularity are realized through atomic operation of an in-memory database and streaming calculation.The application solves the problems of poor safety, low high-concurrent performance, lagging analysis feedback and difficulty in reliable integration of commercial logic of the existing online examination system, and realizes safe, efficient, real-time and commercial intelligent examination data processing.

The invention provides a cross-cloud platform virtual private cloud interconnection system based on a unified cloud management platform and a construction method, and relates to the technical field of computer networks. The system comprises a unified cloud management platform and at least two independent cloud platforms. Corresponding virtual private clouds run on the independent cloud platform, and the virtual private clouds are in network isolation; the unified cloud management platform is based on the VPN technology, a VPN tunnel network is constructed between a VPN server and a VPN client, and the virtual private cloud client serves as a gateway node to achieve communication between virtual private clouds on different cloud platforms. The configuration information of the VPN server and the virtual private cloud client is issued in a mode of virtual machine data injection and VPN server interface; the problem of data and information islands among cloud platforms in a multi-cloud platform scene can be solved, the resource sharing capability among the cloud platforms is enhanced, and the cross-cloud interconnection of applications and the cross-domain sharing capability of data are realized.

The invention relates to a DPU-based fog cloud collaborative Internet of Things intrusion detection system and method. The system comprises an edge node used for receiving target data sent by Internet of Things equipment, encoding the target data through a pre-trained stacked auto-encoder to obtain a plurality of encoding features, determining a prediction capability index of each encoding feature, and determining a target encoding feature according to each prediction capability index; the cloud server is used for receiving the target coding features sent by the edge node, inputting the target coding features into a pre-trained Internet of Things intrusion identification model, extracting target global features, target local features and target time features associated with the target coding features and Internet of Things intrusion through different branch networks in the Internet of Things intrusion identification model, and sending the extracted target global features, target local features and target time features to the edge node; and outputting an Internet of Things intrusion detection result according to the target global feature, the target local feature and the target time feature. By adopting the system, Internet of Things intrusion detection with low resource occupation, high precision and high efficiency can be realized.

The application discloses an Internet of Things mass service bill processing system, method and device and a storage medium, and comprises an application software service layer, a platform environment service layer and an infrastructure service layer.The application software service layer is used for collecting first service bills, and performing format analysis, deduplication, merging and distribution processing on the first service bills to obtain first bill files and first bill messages, and then uploading the first bill files and the first bill messages.The platform environment service layer is deployed with a distributed file system, a distributed coordination service, a data database and a streaming processing framework, is used for receiving the first bill files and the first bill messages, and performing full-amount synchronization on the data database according to the first bill files and incremental synchronization on the data database according to the first bill messages.The infrastructure service layer comprises a bottom cloud host, and is used for providing a computing resource pool, a storage resource pool and a network resource pool.The application improves the processing efficiency of service bills, optimizes the use perception of users, and can be widely applied to the technical field of data processing.

The invention discloses a method and a device for guiding and optimizing data traffic across virtual network domains. The method comprises the following steps: edge gateway equipment of a first communication branch receives data traffic sent from the first communication branch to a second communication branch; according to the invention, through the architecture of combining centralized strategy control and distributed data forwarding, the single-point fault risk in the traditional scheme is eliminated. The data transmission path is directly established between the edge gateway devices, the performance bottleneck caused by the fact that data flow passes through a single agent node is avoided, when any transmission path breaks down, the system can be automatically switched to the standby path within millisecond time, and the reliability and continuity of enterprise cross-branch network communication are remarkably improved.

The application discloses a user behavior analysis method, which can deeply understand the operation mode and behavior characteristics of users by modeling the user behavior sequence using an RNN model, thereby optimizing the user interaction experience; the ARIMA model is used to model the page performance data, which can accurately capture the time sequence characteristics of the page performance, identify potential performance bottlenecks and change trends; the RNN model is introduced into the ARIMA model, the page performance data is taken as an input of the sequence model, and the user behavior data is analyzed together, so that comprehensive analysis of the user behavior and the page performance can be realized, and the prediction accuracy and practicability of the model are improved; the trained model parameters are used to predict the optimization scheme and optimization cost of a future time point, so that the possible performance problems and optimization requirements in the future can be predicted, the optimization scheme can be prepared and implemented in advance, and potential performance bottlenecks can be avoided.

The invention provides a cross-working-condition fault diagnosis method and system for a gearbox of a mining shovel loading robot. The method comprises the steps that S1, vibration signals of the gearbox of a lifting mechanism of the mining shovel loading robot under different working conditions are collected; s2, constructing a diagnosis network and performing adaptive feature extraction on the vibration signals; the diagnosis network comprises a feature extractor and a fault classifier of the one-dimensional deformable convolutional network; s3, training the diagnosis network by adopting a dual-stage domain adaptive strategy fused with triple loss, and outputting a diagnosis model; and S4, performing fault diagnosis on the vibration signal under the target working condition according to the diagnosis model, and outputting a diagnosis result. According to the method, a double-stage domain self-adaptive strategy is adopted, feature optimization is carried out on the source domain in the first stage, data enhancement and domain invariant feature learning are carried out through mixed source domain and target domain labeling samples in the second stage, the domain offset problem caused by working condition changes is effectively relieved, and cross-working-condition efficient fault diagnosis is achieved.

The invention relates to a mail system single copy storage system and method based on object storage. The system comprises a mail index service module, an object storage service module, a single copy storage management service module and a mail client. When a mail is sent, the hash value of a mail body file is calculated, only a unique copy is stored in the object storage, and the reference count of the unique copy is managed through the metadata table; when the mail is read, the file in the object storage is directly accessed through the mail index; when the mail is deleted, the reference count is decreased progressively, and the file is physically deleted only when the count is zero. According to the method and the system, single copy storage of the mail attachments is realized, the storage space occupation in a mass mail sending scene is remarkably reduced, and meanwhile, the consistency and the security of data are ensured through a reference counting mechanism.

Embodiments of the present application relate to a data synchronization method and a computing device. The method comprises: in a first management node and a second management node, deploying a controller component in the first management node or the second management node, determining the management node in which the controller component is deployed as a master node, and determining the management node in which the controller component is not deployed as a standby node; deploying a management service component on the master node, deploying a data synchronization component on the standby node, and synchronizing data in the master node to the standby node and a computing node through the data synchronization component; when the master node fails, migrating the controller component and the management service component to the standby node, and determining the standby node as a new master node; and when the standby node fails, migrating the data synchronization component to the master node, and continuing data synchronization by the migrated data synchronization component. In this way, high-availability and high-performance synchronization of data in a small-scale AI cluster can be achieved at a low cost without relying on expensive shared storage devices.

The application discloses a substation optical circuit topology dynamic reconstruction and incremental consistency control method and system, and relates to the technical field of intelligent substation electrical secondary auxiliary design. The method comprises three core methods of connection creation and incremental synchronization, topology reversible conversion and selective rendering. The connection creation and incremental synchronization method responds to port dragging and multi-selection operation, and only performs local connection chain reconstruction and incremental rendering on the affected port neighborhood; the topology reversible conversion method responds to optical cable node insertion or deletion operation, and realizes bidirectional reversible conversion of the connection between two ports and the connection between two ports and the optical cable node; the selective rendering method only renders the connection with both ends of the port being valid in the current window, and responds to window scrolling to dynamically add or delete. The application avoids full-station recalculation through local chain reconstruction, guarantees that the connection semantics are not lost in design change through reversible conversion, solves the large-scale scene performance bottleneck through partition rendering, and realizes efficient incremental consistency maintenance of optical circuit design.

This application provides a user mining method and apparatus, electronic device, program product, and storage medium. The method includes: acquiring user data through public API interfaces of multiple social networking platforms and storing the acquired user data in an intelligent computing performance-enhanced file storage service deployed on a cloud computing platform; using a data processing cluster on the cloud computing platform to read the user data from the intelligent computing performance-enhanced file storage service, reconstructing features based on the user data, and storing the reconstructed feature data back to the intelligent computing performance-enhanced file storage service; constructing and training a reinforcement learning model on the cloud computing platform based on the reconstructed feature data; storing historical experience data of the agent in the intelligent computing performance-enhanced file storage service during the training process; deploying the trained reinforcement learning model to a production environment and using the reinforcement learning model to perform user mining based on real-time acquired user data.

Embodiments of the present application provide a channel training method and program product, the method comprising: in response to a first update instruction and first equalizer state information sent by a first device, receiving end adaptive equalization is carried out on equalization coefficient through a software interface; according to the receiving end adaptive equalization result, the second update instruction and the second equalizer state information are sent to the first device, so that the first device modifies the specified equalization coefficient, receives the third equalizer state information and the third update instruction sent by the first device after updating; in response to the third equalizer state information and the third update instruction, hardware adaptation is carried out on the receiving end equalization coefficient, and the fourth equalizer state information is sent to the first device according to the hardware adaptation result, wherein the fourth equalizer state information is used to indicate that the channel composed of the first device and the second device is trained. The problem that only hardware automatic training is used for channel training in the related art leads to unstable channel training and long time consumption.

The invention discloses an event-assisted unmanned aerial vehicle night precise landing system and method. The system comprises a high-frequency dynamic infrared sign array module, an event camera module, a pose resolving module and a landing control module. The system takes a high-frequency dynamic infrared mark array as a positioning reference, an event camera carried on the unmanned aerial vehicle captures an optical event flow with unique time sequence characteristics by accurately controlling the independent flicker frequency of each infrared light source, then the pose of the unmanned aerial vehicle is calculated by using an N-point perspective method, and finally accurate landing is realized by a landing control module. According to the invention, the event camera and the high-frequency dynamic infrared sign array are innovatively combined, the cooperative working mechanism of the event camera and the high-frequency dynamic infrared sign array is utilized, sub-millisecond-level positioning feedback is realized in a complex environment, the limitation of a traditional static optical sign is broken through, the positioning precision and reliability are improved, and the method is suitable for accurate take-off, landing and positioning of the unmanned aerial vehicle at night and in the complex environment.

The invention relates to an annotation data processing method and device, equipment, a medium and a product, and the method comprises the steps: obtaining corresponding incremental annotation data in response to an annotation operation of a user on a current page; storing the incremental annotation data to a cache position corresponding to the current page, and rendering the incremental annotation data to the current page; and under the condition that the user ends the annotation operation for the current page, exporting each piece of incremental annotation data in the cache position corresponding to the current page. According to the method, every time the user performs the annotation operation on the current page, the annotation data generated by the annotation operation serves as the increment annotation data to be cached, and after the user ends the annotation operation for the current page, the caching work of the increment annotation data of the current page is completed; therefore, the exporting efficiency of the annotation data is greatly improved.

The application discloses a data read-write processing method and an electronic device in a storage system, relates to the technical field of computer storage, and realizes the asynchronization and isolation of read-write operations by constructing a three-layer resource processing architecture, decoupling host read-write requests into independent controllable processing stages. The first resource focuses on recording metadata, the second resource serves as a temporary carrier for data transfer, and the third resource serves as a persistent cache management center. This architecture enables the write operation to return a response to the host as soon as the data reaches the second resource, greatly reducing the write delay; the read operation realizes non-blocking access through the concurrent control mechanism of the third resource. Overall, the response speed and throughput of the storage system are improved, and the performance bottleneck of mutual blocking of read and write due to competition for the same data resource in the related architecture is solved.

The application discloses an IO optimization method and device based on distributed storage, equipment and a storage medium, and relates to the technical field of distributed storage. The method comprises the following steps: a distributed metadata index layer is constructed between a client and a metadata server, and a dynamic mapping relationship between a path prefix and the metadata server is maintained, so that when the client initiates a file access request, the target metadata server can be directly located based on the mapping relationship, and whether the request is routed to a temporary metadata server for load balancing is dynamically determined according to real-time load information of the target metadata server. Therefore, the problems of metadata request amplification caused by tree directory level-by-level analysis and uneven load of metadata servers caused by static directory binding in the related art are solved, and the technical effects of significantly reducing metadata interaction delay, improving the throughput of the system in the massive small file and burst IO scene, and avoiding a single metadata server from becoming a performance bottleneck are achieved.

The invention provides a SmartNIC network card multi-dimensional data acquisition method for heterogeneous computing, and relates to the technical field of network management, the method comprises the following steps: taking a SmartNIC network card as a data hub, and carrying out active query of a running state to obtain M heterogeneous component state data streams; carrying out cross-calculation component alignment segmentation, and constructing K associated behavior portraits; performing performance mode mining, and constructing a cross-component behavior graph; after interference mode recognition is carried out, interference root cause quantitative analysis is carried out according to the K component calling chains, and an interference root component and an interference propagation quantitative link are obtained; and carrying out dynamic balancing on closed-loop resources of the SmartNIC network card. According to the method and the device, the technical problems that the resource scheduling and task scheduling efficiency is low and the overall performance of the system is influenced due to the fact that the prior art generally focuses on monitoring of a single computing resource and neglects the synergistic effect among a plurality of computing components in a heterogeneous computing environment are solved.

The invention relates to the technical field of computer networks, and discloses a multistage CDN (Content Delivery Network) traffic scheduling method based on linear constraint, which comprises the following steps of: receiving an access request analyzed by a local DNS (Domain Name System) to form a scheduling instance to be optimized; carrying out complexity discrimination on the scheduling instance, and determining an applicable scheduling category of the scheduling instance; establishing a performance upper limit model related to server hardware resources; functional linear constraints are introduced, and Big-M parameters related to the constraints are finely set; and server cluster information, server information, quality constraint information and region hijacking information are obtained, joint solving is performed on the performance upper limit model and the functional constraint, a target cache node IP address meeting all constraint conditions is determined and fed back to a local DNS, and response and scheduling of an access request are completed. According to the invention, through a multi-stage scheduling framework and refined Big-M parameter setting, the computational efficiency and resource utilization precision of CDN scheduling in a complex network environment are improved.

The embodiment of the invention provides an acoustic horn immersion type sound field rendering method and system, and relates to the technical field of sound field rendering, and the method comprises the steps: obtaining a user position and a head posture in a virtual space; acquiring the position and intensity of a virtual space sound source; identifying a focus sound source currently concerned by the user according to the user position, the head posture, the sound source position and the intensity; identifying sound sources except the focus sound source as environment sound sources; performing sound field rendering on the environmental sound source to obtain a shared environmental sound field signal, the computing resource of the environmental sound source being lower than the computing resource of the focus sound source; performing convolution operation on the original audio signal of the focus sound source and the head acoustic transmission function to obtain a personalized high-resolution signal; and superposing the shared environment sound field signal and the personalized high-resolution signal to generate a binaural signal, and playing the binaural signal through an acoustic horn array. According to the method, the immersion and the actual effect of virtual reality training can be improved.

The application belongs to the technical field of databases, and specifically discloses a multi-mode time series data processing method and system. The method comprises the following steps: establishing a row storage table and a column storage table according to a table creation request, and establishing an associated mapping relationship between the row storage table and the column storage table; receiving time series data to be inserted and extracting label data and index data, determining a target data node based on label data hash calculation, writing the label data into the row storage table, and writing the index data into a target time series data block of the column storage table; performing partition pruning and scanning on the stored time series data based on a screening condition, and generating a query result by parallel execution of split query tasks and merging of results based on cross-mode query requirements; determining the affected time series data block based on an operation request, and updating the label data and the index data or deleting invalid data according to the type of the operation request. The application can realize low-redundancy storage, high-concurrency read-write and fast query analysis of time series data.

The invention discloses a Yarn application process tree construction method and device and a storage medium, and relates to the technical field of Hadoop YARN resource management and monitoring. A method for constructing a Yarn application process tree comprises the steps that a monitoring program is started, and a monitor marks monitored application processes as different codes according to configuration; executing corresponding monitoring logic according to the code to obtain an application process tree construction result; the monitoring logic comprises a traditional global scanning construction mode, a single-process mode, a sub-process interface construction mode and an incremental updating mode. The method is simple in implementation mode and suitable for process monitoring and resource metering optimization of the node side of the YARN node-level resource manager, self-adaptive selection for different types of containers is achieved by introducing a multi-mode switchable monitoring framework, and resource waste and performance bottleneck caused by traditional fixed full scanning are avoided.

The present application belongs to the field of language processing, and specifically relates to a multi-causal relationship extraction method and system for intelligent evaluation, aiming to solve the problems of performance bottleneck, significant "bucket effect", and insufficient adaptability and integration caused by relying on multiple independent models. The present application comprises: constructing a causal analysis prompt template for generating step-by-step reasoning as a thinking chain prompt word, combining the original corpus to generate a fine-tuning corpus, inputting into a large language model for training, and adopting low-rank adaptation technology to fine-tune the large language model during the training process to obtain a causal relationship extraction model; inputting the corpus to be extracted into the causal relationship extraction model, and outputting the causal relationship extraction result; generating a comprehensive evaluation result and decision suggestion. The present application improves the efficiency and quality of the multi-causal relationship extraction task, enhances the stability and adaptability of the system, and at the same time reduces the labor cost and provides scientific and reliable evaluation results and decision suggestions.

The invention discloses a method and system for reducing KV storage and loading expenditure in a retrieval enhancement generation system, and the method comprises the steps: S1, compressing a key value cache block of a pre-calculated external document to reduce the storage and loading expenditure of the key value cache block; s2, when the requested key value cache block does not hit the cache, searching and replacing the key value cache block with similar semantics in the cache according to the similarity of retrieval results; and S3, performing comprehensive value evaluation on the key value cache blocks in the cache based on the frequency, the recency and the first retrieval score, and executing cache replacement according to an evaluation result to maximize a cache hit rate. Through an innovative two-stage compression technology, storage occupation and data loading overhead of mass external document pre-calculation key value cache are remarkably reduced. Based on a key value cache reuse strategy of semantic similarity, high-delay disk input / output access is converted into rapid memory access, and the performance bottleneck of data loading is effectively relieved.