27 results about "Tertiary storage" patented technology

The invention relates to the technical field of computers, in particular to a database hot loading method, device and equipment based on a persistent memory and a medium, and the method comprises the following steps: constructing a three-level storage structure comprising a dynamic random access storage layer, a persistent memory layer and a solid state disk storage layer; when the current database is in a starting stage, loading the pre-write log from the persistent memory layer, and starting to execute a redo operation and / or a cancel operation from a preset check point so as to recover the state of the current database; and when the current database is in a running stage, obtaining a current data access frequency and a current resource state of each layer in the three-level storage structure, and carrying out cross-layer migration on data in the three-level storage structure based on a preset dynamic scheduling strategy according to the current data access frequency and the current resource state. Therefore, the problems that a traditional database system is long in starting time, high in memory cost, large in resource occupation and the like are solved, and the overall performance and the resource utilization efficiency of the database system are improved.

The invention relates to the technical field of computer data writing, in particular to a computer data writing system and method.The computer data writing system comprises a multi-level storage module, a data receiving and analyzing module, a heat evaluation and layering module, a differential writing module, a dynamic migration module, a metadata management module and a monitoring and scheduling module; according to the corresponding method, a hot, warm and cold three-level storage architecture is constructed, an initial data storage hierarchy is determined based on a hot degree prediction model, and differential write-in operation is executed; establishing a'migration forecast-write optimization 'cooperation mechanism, a'write performance data-storage hierarchy parameter' closed-loop optimization mechanism and a'multi-scene write strategy-hierarchy-migration 'parameter mapping system to ensure data access transparency; the cold data storage cost is reduced while the high write-in performance of the hot data is ensured, the problem that the storage performance and the cost are difficult to balance in a time sequence big data scene is solved, and the method is suitable for high-frequency time sequence data storage scenes such as power monitoring and financial transactions.

The invention discloses a local data security storage method for a non-networked vehicle, which comprises the following steps of: setting a vehicle-mounted MP5 (Mobile Pentium 5), and storing local data through the vehicle-mounted MP5; setting a three-level storage architecture for the vehicle-mounted MP5, and storing vehicle static information into a configuration data area; storing the operation parameters into a runtime data area according to a time sequence; storing the fault event into an event data area; implementing a hierarchical encryption strategy on data, performing full disk encryption on storage equipment, performing independent file encryption on sensitive data files, and implementing additional character encryption on key fields; an encryption key is derived based on a vehicle VIN code and a user PIN, a circular buffer management mechanism is established, and a secure data export interface is provided. According to the invention, the confidentiality, integrity and availability of vehicle local data can be improved, and the safety and reliability of vehicle data storage are improved.

The application discloses a multi-competition face searching system and method based on GPU parallel acceleration and Milvus, and relates to the technical field of face recognition and vector retrieval. The system is divided into an access layer, a service scheduling layer, a GPU inference engine layer and a three-level storage layer; GPU batch parallel face detection and 512-dimensional feature extraction are realized by relying on CUDA; video frame feature batch deduplication is completed by using high-speed operation of the GPU; Milvus adopts competition_id as a partition key to realize multi-competition data partition isolation, and is matched with Milvus-GPU to accelerate partition vector retrieval; the service scheduling layer dynamically divides GPU computing power priority, and preferentially guarantees real-time retrieval resources; the system configures hierarchical API keys and IP white lists to realize safety management and control, and is compatible with watermark storage and extra_data parameter transparent transmission functions. The application overcomes the defects of low efficiency of traditional CPU serial processing, large video storage redundancy and high full-database retrieval delay, has the advantages of fast processing speed, high hardware utilization rate, small storage cost and strong safety, and is suitable for competition personnel verification, park visitor retrieval, security blacklist control and other scenes.

The present disclosure provides a graphics card storage resource management method and device, electronic equipment and readable medium, and relates to the technical field of graphics cards. The graphics card storage resource management method comprises: obtaining an access history of a virtual address block; and according to the access history, performing step-by-step migration of the virtual address block between a graphics card, a host memory and a solid state disk in a UVM framework. Through the embodiment of the present disclosure, the solid state disk is organically integrated into the UVM system as the third level of storage, the addressable space of the graphics card is expanded, the access frequency and the characteristics of the storage medium are balanced, and the probability of hot data residing in the high-speed layer is improved. While optimizing the utilization rate of the graphics card storage resource, the graphics card access efficiency and reliability are ensured.

The invention discloses an efficient and parallel autoregressive diffusion model hardware acceleration system based on an FPGA (Field Programmable Gate Array), and relates to the technical field of FPGA and machine learning. The system comprises a host end and an FPGA hardware end, the host end is responsible for preprocessing and post-processing of data, and the FPGA hardware end comprises a system control scheduling layer, a three-level storage structure layer and an autoregressive diffusion parallel architecture. According to the method, a collaborative pipeline design is adopted, deep parallelism of an autoregression Token generation process and a diffusion picture generation process is realized through a Token grouping adding mechanism, and a diffusion denoising process can be started without waiting for the completion of complete sequence generation. Besides, a universal attention processing hardware module is designed in the system, QKV projection and matrix multiplication are realized by using a systolic array, and calculation resources are optimized in cooperation with a pipelined Softmax calculation and delay normalization strategy. Through collaborative design of software and hardware, the problems of high reasoning delay and large resource overhead of the autoregression diffusion model in an edge computing scene are effectively solved, and the throughput and the energy efficiency ratio are remarkably improved while the generation quality is ensured.

The invention discloses a data offline retention time calculation method and device, and relates to the technical field of data offline time calculation, and the method comprises the steps: scanning a single physical page of a three-stage storage unit block through return information obtained by decoding a low-density check code in a solid state disk and the threshold voltage distribution characteristic of a flash memory, and obtaining a data offline retention time value of the three-stage storage unit block; the offline time of the solid state disk is estimated, so that the data retention time of the solid state disk and the scanned real-time voltage axis are updated, and the problems that in the related technology, the time cost and the resource overhead of voltage axis scanning on each physical block are large, and due to the fact that the distance between the erasing state and the programming state of the physical block of a single-stage storage unit is long, the time consumption is low are solved. The technical problem that distribution information cannot be obtained in a threshold voltage scanning interval opened by a manufacturer is solved, and the technical effect that the data retention time of the whole disk and the scanned real-time voltage axis are updated in time, so that the accuracy and reliability of subsequent data reading operation are powerfully ensured is achieved.

The application discloses a vector library model heterogeneous scheduling method and system facing asymmetric hardware resources, and belongs to the technical field of hardware scheduling. The vector library model heterogeneous scheduling method facing asymmetric hardware resources extracts a computation memory ratio to construct a global asymmetric feature matrix by monitoring physical indexes of nodes in real time, divides the nodes into computation advantage type nodes and capacity advantage type nodes; decouples a vector processing request into reasoning subtasks and retrieval subtasks, and schedules the subtasks to corresponding type nodes based on the global asymmetric feature matrix; for extreme asymmetric nodes, adopts double-buffer asynchronous scheduling based on a three-level storage architecture to completely overlap computation operation and data transfer; monitors resource saturation in real time, and performs subtask cross-node migration and asynchronous compensation computation based on a redirection strategy. The application significantly improves resource utilization and response real-time performance of a heterogeneous cluster under mixed loads of AI reasoning and vector retrieval, reduces transmission delay, and expands model deployment capacity.

The invention discloses a data backup management method and system for an automobile data recorder, relates to the technical field of data backup management, solves the technical problems of data storage structuralization and safety deficiency and passive and low-efficiency storage space optimization, and aims to calculate the use frequency through a quantitative standard and accurately divide high-frequency / intermediate-frequency / low-frequency scenes. A differential backup strategy is matched, high-frequency scene data backup lagging is avoided, low-frequency scene redundancy is reduced, backup efficiency is improved, a three-level storage system is established, core event data are classified independently, retrieval time is shortened, meanwhile, three-level permission of an administrator, a common user and a temporary authorized user is set, unauthorized access is effectively prevented, and data backup efficiency is improved. The core data privacy security guarantee capability is improved, low-frequency non-important data is transferred to cold storage according to data access frequency and importance dynamic optimization, high-frequency data is encoded by H.265, and the local storage utilization rate is improved.

The invention discloses a wide area network flow management method and system based on an SD-WAN architecture. The method comprises the following steps: constructing a flow table level-to-level management mechanism based on use frequency, and dividing flow table items into a high-priority flow table set and a low-priority flow table set according to use frequency values; constructing a flow table association prediction mechanism, and establishing a prediction model; establishing a multi-stage flow table storage structure at a network node; when the data packet arrives at the network node, flow table entries matched with the data packet are queried in the first-level storage area, the third-level storage area and the second-level storage area in sequence; if the matching is successful, forwarding the data packet by using the matched flow table item, and sending a prediction request to the controller to trigger associated flow table prediction; if the matching fails, requesting to generate a new flow table item from the controller; in order to solve the problem that passive flow table management causes new flow first packet forwarding delay and seriously affects the service quality of delay-sensitive applications, the first packet forwarding delay is reduced to be close to zero delay.

The application provides a data processing method and an electronic device; the method comprises: initializing a memory pool of a first file; wherein the memory pool corresponds to a three-level storage structure, the three-level storage structure comprises a first-level cache ring stored by using a ring queue, a second-level cache unit stored by using a continuous storage space, and a third-level cache slice stored by using a continuous storage space; the first-level cache ring comprises a plurality of second-level cache units; the second-level cache unit comprises a plurality of third-level cache slices; writing data into the memory pool, reading data in the memory pool, or performing remote synchronization on the data in the memory pool. Through the application, the memory utilization and the data access efficiency can be improved.

A tiered storage arrangement is contemplated that provides high speed tier 1 storage for rapid access to recalled data, mid-speed tier 2 storage for data that is not frequently recalled and tier 3 storage for archive purposes that may include tape library storage and even cold storage in vaults. Embodiments contemplate a server linked to all tiered storage devices on one end and a host computer on another. Data files can be migrated between the tiered storage systems based on migration policies, such as time elapsed from being filed or last accessed. Also, embodiments contemplate a time bar displayed at the host computer that provides accurate time predictions of data recall including information of the kind of storage on which a target file is stored and the pathway to recall the target file / s locally to the host computer.

This invention provides an audio storage method, electronic device, and storage medium, relating to the field of voice recognition technology. In the method, the audio to be stored is divided into several sub-audio segments, and the emotional intensity value corresponding to each sub-audio segment is determined using its corresponding acoustic and physiological feature vectors. Then, by combining keyword density and topic novelty scores, an importance score is obtained for each sub-audio segment, achieving a comprehensive and accurate quantitative assessment of its importance. Furthermore, a dynamic three-level storage strategy based on importance scores is used to store the sub-audio segments: low, medium, and high importance levels are respectively discarded, compressed, and stored without loss. This achieves refined dynamic configuration of storage resources, eliminating redundant data while ensuring the integrity and availability of the core audio data of the audio to be stored. It significantly reduces storage usage, avoids resource waste, and improves the utilization rate of storage resources.

The application provides a door lock low-power consumption video linkage recording device, and belongs to the technical field of smart home. The device comprises a door lock main body and a door lock companion. Through the linkage of a heterogeneous control module, a layered wake-up mechanism and a three-level storage unit, and the design of business sharing, the energy consumption of the door lock main body is significantly reduced under the premise of guaranteeing the video recording quality and transmission efficiency. The device has a reasonable structure design, realizes the balance between the door lock endurance and the video monitoring function, and is suitable for various battery-powered smart door lock scenes.

The invention provides a heterogeneous multi-core system off-chip communication system supporting cluster expansion, belongs to the technical field of integrated circuit design, and realizes full-flow hardware fault-tolerant control of CRC (Cyclic Redundancy Check) and type analysis, configuration / data double FIFO (First In First Out) hierarchical caching, retransmission RAM (Random Access Memory) data backup and feedback-based automatic retransmission through a sending end module. According to the system, a three-level storage architecture composed of a configuration information FIFO, a data information FIFO and a retransmission RAM is adopted at a sending end, control information and service data are effectively isolated, a backup copy is created for each data packet by using the retransmission RAM, and hardware-level rapid detection and automatic retransmission of transmission errors are realized in combination with feedback signals of a receiving end, so that the transmission efficiency is improved. And the accuracy and continuity of data transmission are guaranteed from the source.

The invention discloses a historical data analysis method and system based on a space-time dimension. The method comprises the following steps: constructing a multi-level space-time association network through GPS point locations and time slices of a historical scene; constructing a historical data slice storage system based on the multi-level space-time association network; historical data analysis is realized through at least one of the following modes based on a historical data slice storage system: a single GPS point location time axis mode; a multi-GPS point simultaneous association mode is adopted; a linear series time axis mode; and a planar matrix correlation mode. Through construction of a space-time association network, a three-level storage system and multi-mode analysis, the problems of insufficient space-time integration, data fragmentation and the like in the prior art are solved, structured management and deep mining of historical data are realized, support is provided for cultural heritage protection, document travel innovation and academic research, and historical data processing efficiency and cultural value are improved.

The invention provides a data processing method and electronic equipment. The method comprises the steps of initializing a memory pool of a first file; wherein the memory pool corresponds to a three-stage storage structure, and the three-stage storage structure comprises a first-stage cache ring stored by adopting an annular queue, a second-stage cache unit stored by adopting a continuous storage space and a third-stage cache chip stored by adopting the continuous storage space; the first-level cache ring comprises a plurality of second-level cache units; the second-level cache unit comprises a plurality of third-level cache pieces; and writing data into the memory pool, or reading the data in the memory pool, or performing remote synchronization on the data in the memory pool. Through the application, the memory utilization rate and the data access efficiency can be improved.

The application relates to a pointer allocation device of a linked list, a linked list structure and a chip, which comprises a first-stage FIFO module, a second-stage cache module, a pointer control logic module and a third-stage storage module; the first-stage FIFO module is used for storing empty pointers; the second-stage cache module is used for storing first use state information of the pointers corresponding to part of storage units of a storage entity; the third-stage storage module is used for storing second use state information of the pointers corresponding to all storage units of the storage entity; the pointer control logic module is used for obtaining empty pointers from the first-stage FIFO module according to a write request, and when the number of empty pointers in the first-stage FIFO module is insufficient, the second-stage cache module is used to supplement empty pointers to the first-stage FIFO module, and when resources in the second-stage cache module are insufficient, the third-stage storage module is used to supplement resources to the second-stage cache module. Through the application, the hardware resources of the linked list can be reduced, and the hardware logic of the linked list can be simplified.

The invention relates to a pointer distribution device of a linked list, a linked list structure and a chip. The pointer distribution device comprises a first-stage FIFO module, a second-stage cache module, a pointer control logic module and a third-stage storage module, the first-stage FIFO module is used for storing null pointers; the second-level cache module is used for storing first use state information of pointers corresponding to partial storage units of the storage entity; the third-level storage module is used for storing second use state information of pointers corresponding to all storage units of the storage entity; and the pointer control logic module is used for acquiring null pointers from the first-stage FIFO module according to the write request, supplementing null pointers to the first-stage FIFO module by utilizing the second-stage cache module when the number of the null pointers in the first-stage FIFO module is insufficient, and supplementing resources to the second-stage cache module by utilizing the third-stage storage module when the resources in the second-stage cache module are insufficient. Through the application, hardware resources of the linked list can be reduced, and hardware logic of the linked list can be simplified.

The invention discloses an unattended processing system for multi-source remote sensing data, and relates to the technical field of remote sensing data processing. The system comprises an automatic production module, an intelligent management module, a cooperative control module and a three-stage storage pool, the automatic production module executes end-to-end large-scale processing from preprocessing to product generation on the multi-source remote sensing data based on a workflow template capable of being arranged, and supports parallel task level and data level; the intelligent management module realizes state awareness and self-adaptive resource scheduling through metadata driving and event monitoring; the cooperative control module serves as a center, provides a unified interface and manages configuration in a centralized mode. And the three-stage storage pool realizes automatic data circulation under the control of the system. Through event driving and intelligent scheduling, the whole-process unattended automatic processing of optical and radar remote sensing data from access to standard product output is realized, and the production efficiency, the system manageability and the expandability are remarkably improved.

The application discloses an energy consumption use data management system and method based on edge computing, and relates to the technical field of data management.The method comprises the following steps: collecting energy consumption data by using an edge device, binding geographical coordinates and time stamps for each piece of data, superimposing a device-level feature code, and generating a unique digital fingerprint; when it is detected that the energy consumption value deviates from an adaptive threshold and the feature matching degree is lower than a feature matching degree threshold, triggering a high-confidence abnormality marking mechanism automatically, and generating a hot spot event record; managing the collected data according to a three-level storage strategy, wherein the three-level storage comprises device-end storage, edge-end storage and cloud-end storage; based on space-time coding, quickly positioning geographical adjacent edge nodes, aggregating the running states of geographical adjacent devices, generating a local optimization instruction through a preset strategy rule library, generating an optimization strategy package by the cloud based on multi-dimensional information, and after virtual verification, directing the optimization strategy package to the target device end.

The invention relates to the technical field of biological sample storage and transportation, and discloses a biological sample transfer and storage system and method. The biological sample transfer and storage system comprises a first-stage storage device, a second-stage storage device and a third-stage storage device, the first-stage storage device is used for temporarily storing biological samples in a collection place, the second-stage storage device is used for temporarily storing and transferring the biological samples in batches, and the third-stage storage device is used for storing the biological samples for a long time. According to the transfer and storage system for the biological samples, the biological samples can be transferred from a collection place to a long-term storage place in a step-by-step, batch and whole-course deep hypothermia mode, and the biological samples are in a deep hypothermia or vacuum ideal environment in the whole course in the transfer process; the condition that the biological sample is in contact with water vapor, oxygen and microorganisms in the air or the temperature rises can be effectively avoided, so that the activity of the biological sample can be effectively prevented from being damaged in the transfer process, the operation is simple, the transfer efficiency is high, and high-throughput and high-efficiency sample turnover can be realized.

The application discloses a multi-level encryption and intelligent clustering storage optimization method and system for electronic invoices, and relates to the technical field of data information security. The application comprises the following steps: extracting multi-dimensional features of electronic invoices and quantifying, establishing a feature vector of the electronic invoices, and initializing a dynamic weight matrix to distribute the initial influence weight of each feature; based on the feature vector and the dynamic weight matrix, the dynamic encryption intensity coefficient of each invoice is calculated through an activation function to implement progressive encryption processing; the encrypted invoice data is reconstructed into a new feature vector, and the storage space is pre-distributed according to historical data to optimize the physical layout; a three-level storage mapping structure is constructed through the clustering result, the storage block capacity is monitored in real time, and the storage space is dynamically expanded to form a circular linked list structure; according to the value range of the encryption intensity coefficient, a key fragmentation strategy is selected differently, and the key fragments are stored in the local or distributed nodes through a threshold scheme.

This invention relates to the field of electrical data management technology, and more particularly to a method and system for remote monitoring and data management of an electricity metering box. The method includes the following steps: deploying an electrical sensor array in the electricity metering box to collect electrical parameters and output electrical sequence packets; performing digital value assessment on the electrical sequence packets to generate data value classification results; constructing a three-level hierarchical storage architecture based on the data value classification results and performing initial hierarchical storage deployment; performing rolling calculations of sequence access heat based on the three-level hierarchical storage architecture to generate a sequence heat list; and performing dynamic migration processing on the three-level hierarchical storage architecture based on the sequence heat list to complete the dynamic bidirectional flow of electrical sequences between the three storage layers. This invention improves the response performance and storage utilization efficiency of electrical monitoring data access while reducing overall storage and maintenance costs.

The application discloses a data backup management method and system for a driving recorder, and relates to the technical field of data backup management.The application solves the technical problems of data storage structure and security loss, passive and inefficient storage space optimization, calculates the frequency of use through a quantitative standard, accurately divides high-frequency / middle-frequency / low-frequency scenarios, and matches differentiated backup strategies, thereby avoiding backup lag of high-frequency scenario data, reducing redundancy of low-frequency scenario, improving backup efficiency, establishing a three-level storage system, separately classifying core event data, shortening search time, setting three levels of permissions of administrators, ordinary users and temporary authorized users, effectively preventing unauthorized access, improving the privacy and security protection capability of core data, dynamically optimizing low-frequency non-important data according to data access frequency and importance, and transferring the low-frequency non-important data to cold storage, and using H.265 coding for high-frequency data to improve the utilization rate of local storage.

The invention provides a display card storage resource management method and device, electronic equipment and a readable medium, and relates to the technical field of display cards, and the display card storage resource management method comprises the following steps: obtaining an access history of a virtual address block; according to the access history, in the UVM framework, the virtual address blocks are migrated step by step among a graphics card, a host memory and a solid state disk. According to the embodiment of the invention, the solid state disk is organically integrated into the UVM system as the third-level storage, the addressable space of the graphics card is expanded, the access frequency and the storage medium characteristics are balanced, the probability that the hotspot data resides in the high-speed layer is improved, and the access efficiency and reliability of the graphics card are guaranteed while the utilization rate of the storage resources of the graphics card is optimized.

The present disclosure provides techniques for optical inspection systems and methods for moving objects. In some embodiments, an optical inspection system includes: an image capturing device configured to acquire images from moving objects; a first-stage storage system; a second-stage processor; a second-stage storage system; a third-stage processor; and a third-stage storage system. In some embodiments, an optical inspection system, includes: an image capturing device; a volatile memory system; a first and a second second-stage processor; and a third-stage storage system. The second-stage processor or the first second-stage processor can be configured to analyze the images from the image capturing device. The third-stage processor or the second second-stage processor can be configured to process information from a processor and / or storage system and produce a report.