180 results about "Network processing unit" patented technology

A system and method of moving information units from an output flow control toward a data transmission network in a prioritized sequence which accommodates several different levels of service. The present invention includes a method and system for scheduling the egress of processed information units (or frames) from a network processing unit according to service based on a weighted fair queue where position in the queue is adjusted after each service based on a weight factor and the length of frame, a process which provides a method for and system of interaction between different calendar types is used to provide minimum bandwidth, best effort bandwidth, weighted fair queuing service, best effort peak bandwidth, and maximum burst size specifications. The present invention permits different combinations of service that can be used to create different QoS specifications. The "base" services which are offered to a customer in the example described in this patent application are minimum bandwidth, best effort, peak and maximum burst size (or MBS), which may be combined as desired. For example, a user could specify minimum bandwidth plus best effort additional bandwidth and the system would provide this capability by putting the flow queue in both the NLS and WFQ calendar. The system includes tests when a flow queue is in multiple calendars to determine when it must come out.

A novel network architecture that integrates the functions of an internet protocol (IP) router into a network processing unit (NPU) that resides in a host computer's chipset such that the host computer's resources are perceived as separate network appliances. The NPU appears logically separate from the host computer even though, in one embodiment, it is sharing the same chip.

In one embodiment, a network security appliance includes a logic circuit, a network processing unit, and a general purpose processor to protect a computer network from malicious codes, unauthorized data packets, and other network security threats. The logic circuit may include one or more programmable logic devices configured to scan incoming data packets at different layers of a multi-layer protocol, such as the OSI-seven layer model. The network processing unit may work in conjunction with the logic circuit to perform protocol parsing, to form higher layer data units from the data packets, and other network communications-related tasks. The general purpose processor may execute software for performing functions not available from the logic circuit or the network processing unit. For example, the general purpose processor may remove malicious code from infected data or perform malicious code scanning on data when the logic circuit is not configured to do so.

The invention discloses a unidirectional transmission internal and external network secure isolating gateway applicable to an industrial control network, belongs to the technical field of the industrial control network. The isolating gateway comprises an external network processing unit, a data ferry unit and an internal network processing unit; the external network processing unit is connected with the internal network processing unit through the data ferry unit; and the software and the hardware of the internal and external networks are isolated at the same time; the external network processing unit, the data ferry unit and the internal network processing unit are completely independent and communicate through a self-defined protocol. The isolating gateway is advantaged by that the isolating gateway is used for realizing unidirectional data transmission by the industrial control network and an upper computer public network under a physical isolating condition; identity authentication and content filter are controlled through a plug-in mode; and the reliability and the security of the industrial control network are ensured to the maximum extent.

The invention provides a neural network processing unit and a processing system comprising the same. The processing unit comprises a multiplying unit module which comprises a multilevel structure forming an assembly line, and is used for executing the multiplying operation of to-be-calculated nerve cells and weight values in a neural network, wherein each level of the structure of the multiplyingunit module complete the suboperation of the multiplying operation of to-be-calculated nerve cells and weight values; and a self-accumulator module which carries out the accumulation of the multiplying operation result of the multiplying unit module based on a control signal, or outputs the accumulation result. The processing unit and the processing system can improve the calculation efficiency ofthe neural network and the resource utilization rate.

A data processing apparatus for registering image data includes a network processing unit configured for acquiring image data through a network, an image correction unit for analyzing the image data, extracting a background image and a subject image from the image data, and correcting the extracted background image according to a correction condition, and a database unit for storing the corrected image data such that the image data is associated with the corrected image data.

The invention provides a device and method for executing neural network calculation. The device comprises an on-chip interconnection module and a plurality of neural network processing modules which are in communication with an on-chip interconnection unit, and the neural network processing modules can read and write data from other neural network processing modules through the on-chip interconnection module. In multi-core multi-layer artificial neural network calculation, neural network calculation of each layer is required to be divided, then the neural network processing modules calculate,calculation result data of each layer is obtained, and multiple neural network processing units further conduct data exchange on the calculation result data of each layer.

An efficient conversion method for a deep learning model provided by the embodiment of the invention is used to solve the technical problem that the development efficiency and operation efficiency of a deep learning model are low. The method includes the following steps: building a data standardization framework corresponding to an NPU (Neural-Network Processing Unit) model according to a general deep learning framework; using the data standardization framework to convert the parameters of a deep learning model into the standard parameters of the data standardization framework; and converting the standard parameters into the parameters of the NPU model. According to the invention, a unified data standardization framework is built for a specific processor according to the parameter structures of general deep learning frameworks. Standard data can be formed using the unified data structure of the data standardization framework according to the parameters of a deep learning model formed by a general deep learning framework. Thus, the process of data analysis by the processor depends much less on the structure of the deep learning model, and the development of the processing process of the processor and the development of the deep learning model can be separated. A corresponding efficient conversion device is also provided.

The invention relates to an intelligent tracking flight system and method of a UAV, and belongs to the field of intelligent autonomous navigation of the UAV. The flight system comprises an image acquisition unit for image acquisition, a flight attitude measurement unit for detecting a flight inertia, a neural network processing unit for neural network processing and a control unit for control flight motion of the UAV; an image collected by a camera in the front of the UAV serves as input, the input is identified by the convolutional neural network, and an obstacle avoidance strategy is given.In the flight process, the alight attitude measurement unit based on a tunnel magnetic resistance effect characterized by high response is used to obtain key change position of the flight attitude ofthe UAV, the image acquisition unit is started to capture key images, a stereoscopic model of the flight environment is formed, key position images are provided for the UAV tracking flight path, and the integrated safety of the UAV tracking flight is improved.

The invention discloses an anomaly detection method and a system for a network processing unit. The method comprises the following steps: anomaly detection operating codes are added into execution processes of each thread of the network processing unit, and the network processing unit makes first marks on mark positions corresponding to the current thread in an anomaly protection mark data area in a shared memory when the network processing unit runs the anomaly detection operating codes of the current thread; when the cycle of a timer is up, a coprocessing unit detects all the mark positions in the anomaly protection mark data area, and the thread not corresponding to the mark positions of the first marks is abnormal if the fact that not all the mark positions are the first marks is detected; and the coprocessing unit can make second marks on all the mark positions if the fact that all the mark positions are the first marks is detected. Based on the characteristics of the process of the network processing unit software, the invention can effectively detect the anomaly of the thread in time, thereby greatly improving the failure detection capability of network equipment using the network processing unit as the core unit in the operating period.