16375 results about "Parallel computing" patented technology

The invention discloses a big language model reasoning optimization method and system based on dynamic video memory compression and memory isomerism, and intelligent management of video memory resources is realized by integrating a dynamic compression strategy of KV Cache and a memory parallel architecture. The method comprises the following steps: 1) analyzing the spatial-temporal characteristics of the KV Cache in real time, adaptively selecting a quantization compression algorithm, a rarefaction algorithm or a low-rank decomposition algorithm, performing hierarchical storage based on attention head importance scores, keeping high precision of a core head, and implementing low-bit quantization on a secondary head; (2) the compressed inactive data are divided into a plurality of data blocks to be stored in a system memory, a parallel data channel group is established according to the number of physical channels, the compressed blocks are concurrently read through multiple channels during loading, and parallel decompression of a sparse matrix is accelerated through a GPU tensor core; and 3) constructing a KV Cache multiplexing mechanism and a parallel channel, and parallelizing a compression / decompression process and model calculation by adopting a hardware acceleration compression and asynchronous pipeline mechanism.

The invention relates to the technical field of cloud computing, and discloses a cluster load balancing processing method based on cloud computing, which comprises the following steps of: S1, initializing a cluster state: determining computing power, power consumption parameters and a task allocation model of each node in a cloud computing cluster; s2, acquiring a node operation state: providing dynamic input data for construction of a load distribution model; s3, constructing a load distribution model: defining a global optimization target; s4, generating a preliminary task allocation scheme: satisfying an optimization target; s5, aiming at the dynamic change of the task arrival rate and the node state; s6, task migration is executed, and the state is updated; and task migration between the nodes is implemented. Through a closed-loop process of node state acquisition, task allocation model construction, dynamic adjustment and feedback optimization, efficient matching of a task request and a computing node is realized. The method can adapt to dynamic changes in a cloud computing environment, the resource utilization rate is remarkably improved, and task processing delay is effectively reduced.

The invention discloses a cloud computing parallel task optimization scheduling method based on a priority dependency graph, which mainly solves the problem of parallel task scheduling in a cloud computing environment, and constructs the priority dependency graph (PDG) of cloud computing tasks by analyzing the dependency relationship and priority among the cloud computing tasks, so as to optimize the scheduling of the parallel tasks in the cloud computing environment. Nodes in the PDG represent cloud computing tasks, edges in the PDG represent dependency relationships among the tasks, the load condition of each computing node is continuously monitored on the basis of the PDG, the tasks are distributed to the most suitable nodes in combination with task priorities and the dependency relationships in the PDG, and if the load of a certain node is too high or the task running time exceeds the expectation, the tasks are distributed to the most suitable nodes. And automatically adjusting the resource allocation of the task. According to the method, the dependency and the priority among the tasks can be more accurately captured, so that the scheduling strategy is dynamically adjusted, the optimal arrangement of task execution is realized, the execution efficiency of the tasks is improved, and the overall resource utilization rate is maximized.

The invention discloses a report verification batch parallel execution optimization method based on a pluggable rule engine framework, and relates to the technical field of report verification, and the method comprises the following steps: receiving compliance rules, converting the compliance rules into a uniform data structure model, analyzing to generate metadata indexes, and supporting business rule set configuration; loading the matched business rule set based on the meta-information of the to-be-verified report, and registering the matched business rule set to an execution container; s3, receiving a to-be-verified report, analyzing a report structure, and generating a report field index table; performing field dependency analysis based on the report field index and the loaded business rule set, and constructing a rule field dependency graph; graph partition optimization and topological sorting are executed according to the dependency graph, and rule batches capable of being executed in parallel are generated; and uniformly collecting and outputting all rule execution results, hit states, abnormal records and performance indexes, and generating a structured verification report and an audit chain log. The financial statement compliance check execution efficiency can be improved.

The invention relates to the technical field of server resource scheduling, in particular to a server resource scheduling system integrating AI and edge computing. The server resource scheduling system integrating the AI and the edge computing comprises a computing node module, a resource calling module, a task traversal module, a demand prediction module, a resource allocation module and a load scheduling module, the computing node module is used for monitoring the real-time running state of a server computing node and performing distributed node architecture fitting, and the load scheduling module is used for scheduling the load of the server computing node; constructing a distributed server node network; the resource calling module is used for acquiring a server running log; performing server resource calling behavior analysis on the server running log, performing resource calling time sequence characteristic evolution, and generating a server resource calling time sequence characteristic; and the task traversal module is used for carrying out unprocessed task traversal identification on the server running log and constructing an unprocessed task state mapping sequence. According to the invention, efficient and load-balanced server resource scheduling is realized.

The invention relates to the technical field of GPU resource dynamic scheduling, in particular to a cloud environment GPU resource dynamic scheduling method, which comprises the following steps: monitoring operation indexes of GPU instances in a cloud environment in real time, collecting GPU video memory allocation records, GPU video memory fragmentation degree magnitude and page migration activity count, and predicting task resource demand total quantity through statistics. And generating a video memory state evaluation result. According to the method, the resource pool adjustment decision and the task matching degree list are jointly optimized, the instance models and the number are screened under the double constraints of the instance starting time parameter and the release cost parameter, and operation delay and extra cost caused by blind capacity expansion and contraction are reduced. Based on a strong binding mechanism started by a video memory pre-allocation instruction and a task process, in combination with continuous tracking of a resource consumption trajectory, a video memory peak deviation and an execution time deviation are fed back in real time, and a closed-loop scheduling optimization link is formed.

The invention discloses a big data stream-oriented memory cache storage system, relates to the technical field of memory cache storage, and is used for solving the problem of poor memory cache storage management. According to the method, the access frequency, the timeliness and the service importance of the data are evaluated, the priority ranking model is constructed, the cache strategy is optimized based on the ranking result, the data storage and scheduling efficiency is improved, the write-in and read operation of the memory cache is adjusted in real time in combination with the flow prediction model, memory overload is effectively avoided, and the data storage and scheduling efficiency is improved. According to the method, non-emergency data writing is delayed by adopting a delayed writing and lazy loading strategy, data is loaded as required, meanwhile, a cache elimination mechanism is designed according to data priority, access frequency, survival time and size, data migration is carried out when a cache space is insufficient, reasonable resource allocation is ensured, and through a multi-level and dynamic cache management scheme, the cache management efficiency is improved. Therefore, the system performance, the resource utilization rate and the response speed are remarkably improved, the data processing efficiency is optimized, and cache conflicts are reduced.

The invention discloses a method for automatically drawing an OpenGL (Open Graphics Library) program by using Vulkan. The method comprises the following steps of: creating a context used by the Vulkan, initializing each module, processing an OpenGL instruction related to texture and data buffering, and managing storage of texture and data buffering resources in a video memory; a shader program used by the OpenGL is preprocessed into a format acceptable to Vulkan, and an OpenGL shader program instruction is created and destroyed; processing an OpenGL (Open Graphics Library) instruction related to frame buffering to generate structural body information required by Vulkan dynamic rendering; an OpenGL instruction of the sampler is also created; processing an OpenGL (Open Graphics Library) instruction for creating a vertex input format and managing a vertex data buffer area, and maintaining vertex input information, a vertex buffer area and an index buffer area required by Vulkan; and finally, drawing or calculating, distributing and calling Vulkan on the basis of all the instructions.

The invention discloses a heterogeneous AI computing power resource scheduling method and system, and the method comprises the steps: constructing a heterogeneous AI computing power resource pool, wherein the heterogeneous AI computing power resource pool integrates the computing resources of a plurality of heterogeneous AI acceleration chips; obtaining a scheduling demand of the AI task, wherein the scheduling demand comprises a task type, a resource request quantity, a priority identifier and a task group association relationship; generating a multi-dimensional scheduling strategy according to task requirements, wherein the scheduling strategy comprises a priority scheduling rule, an affinity scheduling rule and a resource preemption rule; based on a multi-dimensional scheduling strategy, the AI tasks are dynamically allocated to target computing power nodes of the heterogeneous AI computing power resource pool, and the task execution state and the resource utilization rate are monitored in real time; and dynamically adjusting computing resource allocation according to the resource utilization rate. Through the heterogeneous AI computing power resource pool, the resource utilization rate is remarkably improved, dynamic resource allocation is realized through a multi-dimensional scheduling strategy, and meanwhile, a communication path is optimized through an affinity scheduling strategy, so that the problem of task starvation caused by resource fragmentation is avoided.

The present application discloses a task management method, system, device, and medium based on a graph structure, and relates to the field of computer technology. The method includes: decomposing a preset target task into at least two subtasks, creating a task node for each of the subtasks, and mapping the task node to a node set of a preset directed acyclic graph to obtain a task graph; updating the edge set of the task graph according to the dependency relationship between each of the task nodes to obtain a task dependency graph; topologically sorting all the task nodes in the task dependency graph, and determining the execution order corresponding to the independent task nodes and the dependent task nodes according to the topological sorting result; if there are multiple independent task nodes, executing the subtasks corresponding to the multiple independent task nodes in parallel, and executing all the dependent subtasks in sequence according to the execution order. The present application improves the processing efficiency of the task processing framework for complex tasks.

The invention relates to the technical field of data storage and processing, in particular to an RAID card static cache management method and device based on data popularity, and the method comprises the steps that the physical position of a data block needing to be accessed in an SSD array is acquired according to a read-write request; historical data access information of the RAID card is collected, a historical data set is generated, and a neural network model is trained by using the historical data set to obtain a cache management model; predicting and outputting data blocks which are possibly accessed in the future and the access probability of the data blocks through the cache management model; the RAID controller dynamically adjusts a cache strategy in combination with an LRU strategy according to a prediction result output by the cache management model and a read-write request of a file system, and optimizes a storage position and an updating mechanism of a data block in a cache space; and when the hit rate or the data access efficiency does not reach the preset value, adjusting the parameters of the cache strategy and updating the cache strategy. According to the method, cache resources can be more effectively distributed, the cache hit rate is improved, the cache replacement overhead is reduced, and therefore the performance of a storage system is optimized.

A system for adaptively caching network communication protocols provides improved efficiency and performance through a multi-level cache architecture. The system monitors performance metrics and synchronizes cache contents across distributed nodes using a hierarchical structure. Protocol cache requirements are predicted based on usage patterns, enabling proactive cache management. The system compresses cached protocols using existing codebooks to optimize storage and transmission efficiency. Integration with event-driven data communication systems enables seamless protocol selection and translation. The caching system works in conjunction with transaction managers and protocol predictors to enhance network communication performance. The system maintains cache coherency across local, regional, and global cache levels while minimizing network overhead and optimizing protocol availability. By predicting and pre-caching likely needed protocols, the system reduces protocol negotiation latency and improves overall network communication efficiency.

A federated distributed AI reasoning and action platform utilizing decentralized, partially observable hierarchical computing for neuro-symbolic reasoning. It features a federated Distributed Computational Graph (DCG) system integrating core components like pipeline orchestration, transformers, and marketplaces. The platform enables privacy-preserving dynamic resource allocation, intelligent task scheduling, and variable information sharing across diverse computing environments. By coordinating with an AI-based operating system and analyzing performance metrics, environmental conditions, and resource availability, the system optimizes efficiency across AI workloads and decision-making processes. This results in an adaptive, power-efficient, and scalable AI-enabled data processing system capable of handling complex tasks while maintaining peak performance under various operating conditions.

The embodiments of the present application disclose a speculative decoding optimization method, an electronic device and a storage medium, wherein a speculative decoding optimization method is used for a draft model and a target verification model, the method comprising: inputting the draft content generated by the draft model in the previous round into the target verification model; performing parallel verification on the tokens in the draft content in the target verification model; using a lightweight draft length predictor to analyze the latent vector of the last received token position and the last token generated, and predicting the optimal draft length for the next round; and autoregressively generating a corresponding number of drafts based on the optimal draft length predicted by the lightweight draft length predictor through the draft model.

The embodiment of the invention provides a task scheduling method, system and device based on cloud computing and a storage medium, and belongs to the technical field of computers. According to the method, task priorities are initialized according to task resource requirements, task types and task timeliness requirements, task scheduling strategies are determined according to task metadata and computing resource information, the task metadata comprise the task priorities and the task resource requirements, and the task scheduling strategies are used for representing task scheduling sequences and task allocation objects; task scheduling is carried out according to the task scheduling strategy, load fluctuation information and task demand change information of each computing node are obtained, the task scheduling strategy is adjusted according to the load fluctuation information and the task demand change information, and the step of task scheduling according to the task scheduling strategy is repeatedly executed, so that the task scheduling efficiency is improved in the task execution process. Task scheduling and allocation are adaptively adjusted based on load changes and task demand changes, and the task execution efficiency and the system resource utilization rate are improved.

The invention discloses a dynamic data scheduling method and system of a computing power cluster, and belongs to the field of computer science and technical field, and the method comprises the following steps: reading node resources of the computing power cluster, synchronously obtaining a task scale, initializing a dynamic granularity distribution network, and determining splitting granularity; task splitting is carried out according to the splitting granularity, and task priorities are configured; performing load and task execution end prediction on each node in the computing power cluster through the prediction model; performing resource adaptation optimization of the M sub-tasks according to the prediction result; establishing a dynamic optimization channel, performing dynamic optimization on an initial allocation result, and performing task preemption optimization based on task priorities; and performing dynamic data scheduling management of the computing power cluster. According to the method and the device, the technical problems of resource waste, low task execution efficiency and slow system response caused by inflexible resource scheduling and unreasonable task allocation of the computing power cluster under the condition of dynamic load and task scale change in the prior art are solved.

The invention relates to the technical field of data analysis, and discloses a GPU computing power resource scheduling method and system, and the method comprises the steps: collecting node hardware parameters and dynamic load indexes of a GPU cluster to construct a multi-dimensional resource feature vector of the GPU cluster, and constructing a resource portrait of the GPU cluster; establishing a node health degree scoring model of the GPU cluster, and generating a health degree score of a cluster node corresponding to the GPU cluster; analyzing a video memory demand of the GPU task request, and calculating an intensive identifier and a communication dependency relationship; determining the SLA weight of the GPU task request, calculating the resource shortage sensitivity of the GPU task request based on the video memory demand, and calculating the target task priority of the GPU task request in combination with the SLA weight; and determining a resource scheduling node group requested by the GPU task in the resource portrait, generating resource scheduling parameters of the resource scheduling node group, and executing scheduling of computing power resources of the GPU cluster based on the resource scheduling parameters. According to the method, the scheduling efficiency of the GPU computing power resources can be improved.

The invention discloses a large model data distributed management method and device, equipment and a storage medium, and relates to the technical field of data processing, and the method comprises the steps: segmenting large model training data to obtain a plurality of data blocks, and determining a predicted access frequency based on a long short-term memory network model and the historical access frequency of the data blocks; caching the large model training data corresponding to the data block to a corresponding data cache layer by utilizing the predicted access frequency; constructing a resource portrait by using the static attribute and the dynamic index of the GPU node, and allocating the large model training task to a target GPU node by using a preset hybrid strategy and the resource portrait based on the predicted access frequency and the storage position corresponding to the data block in the data cache layer; when it is monitored that the large model training task on the target GPU node is executed, periodic snapshot is conducted on the large model training task through a distributed snapshot algorithm, and the obtained complete data state is stored in a distributed storage center. And the resource vacancy rate is reduced.

A key-value cache management system processes inference requests of a generative model. An inference request requests a starting virtual memory address and a number of layers in the generative model. In response to receiving an inference request, contiguous virtual memory space is reserved for the key-value cache in accordance with the inference request by assigning the starting virtual memory address in the key-value cache and calculating memory pointers for each layer in the model and each block so that the one or more blocks are written sequentially. The generated memory pointers are outputted to the generative model so that each self-attention layer writes computed key-value pairs at physical addresses in the key-value cache specified by the memory pointers.

A method for performing a bootstrapping operation in cryptographic application, includes receiving in an accelerator one or more input ciphertexts used in the cryptographic application to be bootstrapped and iteratively processing one or more accumulator variables in function of portions of the input ciphertexts; multiplying, within each iteration, the processed accumulator variables with a bootstrapping key element belonging to a bootstrapping key comprising a plurality of bootstrapping key elements, said bootstrapping key element taken from a bootstrapping cache memory in the accelerator; while performing the multiplying for each of the one or more accumulator variables in turn, loading into the bootstrapping cache memory from an external memory a next bootstrapping key element of the plurality of bootstrapping key elements to be used in a next iteration of the bootstrapping operation.

The invention relates to the technical field of geographic information systems, in particular to a map tile data efficient access method and system based on hybrid storage and intelligent layering, and the method comprises the following steps: constructing a multi-level storage architecture, carrying out dynamic heat analysis, carrying out intelligent layering scheduling, preloading and space prediction, and carrying out transmission and access optimization. The method has the beneficial effects that the multi-level storage pool is constructed by combining the advantages of object storage and a distributed file system; adopting a dynamic tile popularity analysis model to realize hot data cache acceleration and cold data hierarchical archiving; and meanwhile, a tile request preloading mechanism is introduced, and an access path is optimized in combination with a spatial locality prediction algorithm. According to the method, the storage cost can be remarkably reduced, the tile request delay is reduced, quick response in a high-concurrency scene is supported, and the method is suitable for application scenes such as Internet map services and smart cities.

A distributed indexing and resolution architecture is disclosed for decentralized systems requiring modular, trust-scoped mutation control and dynamic alias governance. The system comprises a plurality of index entries arranged in a parent-child hierarchy, each associated with a structured alias and governed by one or more anchor nodes. Anchors perform localized resolution, mutation validation, and restructuring operations under deterministic policy constraints, enabling semantic scope enforcement and entropy-sensitive adaptation without requiring global consensus or centralized control. The architecture supports scoped alias traversal, asynchronous mutation proposals, and elastic anchor registration based on system state metrics. The indexing substrate may be integrated into heterogeneous infrastructures, including systems comprising distributed software agents, semantic execution platforms, or pseudonymous identity frameworks. Anchors coordinate within defined trust domains to ensure lineage continuity, dynamic rekeying, and semantic integrity across independently governed segments of a decentralized namespace.

The invention relates to the technical field of resource scheduling, in particular to a cloud computing resource management system. In the cloud computing resource management process, the execution data of the computing tasks are analyzed, the task combinations with relevance are screened, and the computing path task set is established, so that the scheduling of the computing tasks is more intelligent, and the resource utilization rate is improved. And resource allocation is performed according to the matching degree of the calculation demand and the computing power pool, the adaptability of the calculation resources is dynamically adjusted, and accurate scheduling of the calculation task is realized. And in combination with the load state of the calculation node, the CPU utilization rate, the memory occupancy rate, the I / O load and other key indexes are monitored in real time, so that the resource allocation is more flexible, the calculation pressure of the high-load node is reduced, and the stability of the whole system is ensured. The load conditions of the computing power pools are evaluated, the task allocation strategy is adjusted, the computing tasks are reasonably distributed among the multiple computing power pools, the task migration and resource expansion strategy is optimized, and the balanced utilization rate of computing resources is improved.

A processing device sets a first flag that indicates whether a first critical security parameter (CSP) file exists. The first CSP file includes a first set of CSPs for a memory device. The processing device sets a second flag that indicates whether the first CSP file is valid. The processing device sets a third flag that indicates whether a second CSP file exists. The second CSP file includes a second set of CSPs for the memory device. The processing device sets a fourth flag that indicates whether the second critical security parameter file is valid. The processing device selects one of the first or second CSP file as an active CSP file based on an evaluation of the first, second, third, and fourth flags.

The invention relates to the technical field of chip development, and discloses a chip post-simulation verification acceleration method, system and device and a storage medium, and the method comprises the steps: collecting multi-dimensional data of a large-scale chip; collecting all module names of a given chip; establishing a mapping relation between a redundant module RTL and a corresponding gate-level netlist, ensuring that interfaces before and after replacement are completely consistent, and designing an automatic script identification and processing parameterization module; establishing a post-simulation verification environment, verifying the consistency of output responses before and after replacement, analyzing the time delay change of a key path, calculating the dynamic power consumption and static power consumption of each functional module before and after replacement, and generating a simulation verification result. According to the method, the hybrid simulation model is constructed, and part of the gate-level netlist irrelevant to the current simulation function is replaced with the RTL, so that the actual simulation scale of a chip is reduced, the problem of overlong simulation time after a large-scale integrated circuit is remarkably solved, and the simulation efficiency is effectively improved.

The invention discloses a clock tree synthesis method and related equipment, and the method comprises the steps: dividing a plurality of time sequence units which need to receive clock signals from a clock source into clustering clusters of different levels, and enabling the load capacitance of each clustering cluster of the same level to be balanced; according to a hierarchical sequence of clusters starting from the clock source, taking each cluster as a node of a clock tree, recursively constructing the clock tree, and inserting a target buffer at the position of each node for driving the corresponding node; analyzing the delay and offset of the constructed clock tree to obtain an analysis result; and if the analysis result indicates that the delay and offset of the clock tree do not meet the preset constraint condition, adjusting the clock tree until the adjusted delay and offset of the clock tree meet the preset constraint condition. According to the technical scheme, power consumption and offset can be effectively reduced, calculation efficiency and design reliability are improved, and the method is suitable for high-performance chip design under the advanced technology.

The invention discloses a permanent incremental backup method based on an S3 bucket-level virtual snapshot, and relates to the technical field of data storage and backup, the method comprises the following steps: initializing a backup system and starting a backup agent program, an index management service and an S3 object storage service; complete backup is executed, and a VID and S3 bucket-level virtual snapshoot are generated; efficiently managing a mapping relation between a file path and a snapshot identifier through a B + tree and an inverted index by utilizing a MongoDB index service; incremental backup is executed, and file changes are accurately recognized through Hash comparison and modification time comparison; a storage position is optimized by adopting an incremental storage and deduplication storage mechanism and an IncDSI optimization algorithm; executing backup chain integration and cleaning; and performing data recovery based on the VID of the file and the directory. According to the method, the backup efficiency is remarkably improved, the storage overhead is reduced, the data recovery process is optimized, and the method is suitable for data backup and management of a large-scale file system and has important application value and prospect.

The invention discloses a compatible operation method for an Android camera HAL in a container based on memory access virtualization, and the method comprises the steps: taking a DMA-Buf memory heap of a Linux kernel host system as a target memory heap, taking an ION memory heap of an Android container system as a source memory heap, building a process exclusive memory management context, an FD cache table and a synchronous fence pool, and carrying out the execution of the process exclusive memory management context, the FD cache table and the synchronous fence pool; kernel registration and node binding of virtual ION equipment, pre-allocation of a first memory pool and access hook registration of kernel layer equipment are completed, and when an ION file descriptor is obtained in an HAL process, legality of a container process is verified, and context binding is initialized to the file descriptor; intercepting a memory allocation request of the HAL process, analyzing and adapting parameters, and preferentially multiplexing first memory pool resources to obtain an ION handle; when a data sharing request is processed, matching an FD cache or generating a new FD through an ION handle; when the HAL process releases the memory, resources are recycled according to a memory source, and when the HAL process exits, the context is cached or destroyed, so that cross-system memory operation compatibility is realized.

The invention discloses a data processing method and device, a processor, electronic equipment and a storage medium. The data processing method comprises the following steps: executing multiple first operations to obtain a plurality of weight gradient sub-tensors, wherein each weight gradient sub-tensor is a part divided by the weight gradient tensor according to 1 * n granularity; the first operation comprises the steps that a plurality of first elements of the input gradient tensor and a plurality of second elements of the activation tensor are loaded to a second cache from a first cache, the second elements are the elements corresponding to the target sub-tensor in the activation tensor, and the weight gradient tensor is obtained through the arrangement of the multiple weight gradient sub-tensors. Reading a plurality of first elements from the second cache to a tensor calculation module; and the tensor calculation module multiplexes the plurality of second elements to execute n times of second operations to obtain a target sub-tensor, wherein each second operation is used for obtaining a target element in the target sub-tensor. The method can reduce the access of the cache and improve the utilization rate of the cache.

The invention relates to the technical field of cloud computing and distributed storage, and discloses a distributed metadata management method and storage system based on cloud computing, and the method comprises the following steps: S1, collecting the access frequency of a plurality of metadata objects in a set time period; s2, performing weighting processing on the access frequency, and constructing access heat distribution under multiple time scales; s3, obtaining the available cache capacity of a plurality of cache nodes, and constructing cache resource weight distribution; s4, assessing the access cost and migration cost between each metadata object and each cache node; and S5, generating a scheduling strategy between the metadata object and the cache node based on a preset optimization target, wherein the optimization target is access cost minimization. According to the method, a heat, resource and cost joint modeling mechanism is constructed, and optimal scheduling and feedback adaptive control are combined, so that accurate decision, dynamic adjustment and migration cost controllability of copy deployment are realized, and the performance and stability of distributed metadata management are effectively improved.