35results about How to "Avoid communication bottlenecks" patented technology

The invention discloses a cell array calculation system and a communication method between cells. The cell array calculation system comprises a main control CPU, a cell array, a cell array bus and at least one memory unit array; the cell array is a two-dimension cell array consisting of at least one cell having functions of calculation and storage; the main control CPU performs communication with each cell in the cell array through the cell array bus; a communication interface is arranged between the adjacent cells in the cell array for transmitting data; the cells participating the communication between cells in the cell array comprises starting point cells, terminal point cells and transfer cells; the starting point cell transmits data to the terminal point cell; the terminal point cell finally receives the data transmitted from the starting point; and the forwarding cells are adjacent along the communication path between the cells and transfers the data transmitted by the starting point cell through the communication cell. The invention can overcome the communication bottleneck between the CPU and the memory and storage in the current computer architecture and improves the integral performance of the system.

The invention disclose a cell array calculation system and a communication method wherein, comprising a main control CPU, a cell array and a cell array bus; the cell array is a three-dimension cell array obtained through stacking more than one two-dimension cell array; the two-dimension cell array consists of more than one cell having functions of calculation and storage, wherein each cell comprises a microprocessor and a nonvolatile random storage device; the nonvolatile random storage device is used for random access of the data related to the calculation of the microprocessor and is used for storing the software code and the data needed to be stored permanently; each cell stores the position of itself in the cell array as an ID for the software or the hardware in the cell to read; the main control CPU performs communication with each cell in the cell array through the cell array bus; and a communication interface is arranged between the adjacent cells in the cell array for transmitting data. The invention can overcome the communication bottleneck between the CPU and the memory and storage in the current computer architecture and improves the integral performance of the system.

The invention relates to a cell array calculation system. The cell array calculation system comprises a main-control central processing unit (CPU), a cell array, a cell array bus and at least one storage unit array, wherein the cell array is a two-dimensional array comprising more than one cell compatible with functions of calculation and storage, each cell comprises a micro-processor and a non-volatile random access memory, the storage unit array is a two-dimensional array comprising more than one storage unit, the cell array and all storage unit arrays are laminated to form a three-dimensional structure, the storage unit in each storage unit array is connected with each cell of the cell array in one-to-one correspondence, the storage units and the non-volatile random access memory are matched and are jointly used for storing an instruction code of software and data required to be permanently stored, the main-control CPU is in communication with each cell in the cell array through the cell array bus, and a communication interface is arranged between adjacent cells and can be used for sending data. By the cell array calculation system, the communication bottleneck existing between the CPU and the memory and storage of a conventional computer framework can be overcome, and the comprehensive performance of the system is improved.

The invention relates to a fine-grained cloud platform security access control method based on user identity capabilities, belonging to the technical field of resource access; the invention divides files uploaded by data owners into control fields and ciphertext fields, and stores the ciphertext fields in Cloud physical nodes, the cloud server saves the control field, and the cloud server agent performs intervention control, reasonably allocates network resources, reduces the burden on data owners, and only needs to update the control list locally in the process of access authorization redirection , associate each globally unique file number identifier with a globally unique user identity identifier, and bind the user's operation authority to achieve fine-grained access control, and encrypt the uploaded identity-permission-ciphertext list , to ensure the confidentiality and integrity of user information and file information, and reduce the risk of information leakage on the third-party cloud platform.

The invention discloses a graph-based large-scale embedding model training method and system for click-through rate prediction. The system includes a dense parameter module and a client-side module, adopts a hybrid communication architecture, and distributes the click-through rate prediction input data set to different jobs. Each worker node maintains a client, and the local model parameters are directly stored in the GPU memory; each worker node holds a copy of the model parameters and synchronizes them during training. The present invention adopts the Embedding model parameter to represent the importance of the feature value of the corresponding category of the input data of click-through rate prediction, expresses the click-through rate prediction data and the embedding model vector as a binary graph model, and uses the graph locality and degree skew characteristics to perform parallel training of the model; Design graph-based partitioning and bounded synchronization to improve scalability and parallel computing efficiency for training large embedding models.