56 results about "High performance network" patented technology

A switch is provided for routing packets in an interconnection network. The switch includes a plurality of egress ports to transmit packets and one or more ingress ports to receive packets. The switch also includes a network capacity circuit for obtaining network capacity for transmitting packets via the plurality of egress ports. The switch also includes a port sequence generation circuit configured to generate a port sequence that defines a pseudo-randomly interleaved sequence of a plurality of path options via the plurality of egress ports, based on the network capacity. The switch also includes a routing circuit for routing one or more packets, received from the one or more ingress ports, towards a destination, based on the port sequence.

The invention provides an efficient AI model management and safe loading system and method oriented to a Serverless framework, and aims to solve the high concurrency performance bottleneck and potential safety hazards of AI model loading in a Serverless environment. According to the system, an AI model management module is introduced, efficient semantic cutting is carried out on an uploaded AI model, and fine-grained encryption model blocks are generated; a high-performance P2P network is constructed by utilizing a Serverless cluster, and distributed storage and fast routing of model blocks are realized; a function instance loads a model block in a high-concurrency mode through a P2P network, a multi-stage intelligent cache is constructed locally, and dynamic caching and prefetching are achieved. According to the method, the model loading parallelism degree is improved by adopting a heuristic cutting algorithm, lightweight model updating is realized by utilizing incremental updating transactions, and the loading overhead is minimized. According to the method, the AI model loading performance and efficiency in the Serverless scene are effectively improved, meanwhile, the flexibility and safety of model management are considered, and the method has remarkable technical innovation and practical value.

The invention relates to the technical field of network supervision, and discloses an extensible high-performance network boundary flow distribution system and method, and the method comprises the steps: a configuration agent module analyzes network configuration information, and transmits the network configuration information of a collection port and a forwarding port to a flow collection module and a flow filtering and forwarding module; the strategy agent module analyzes the network strategy information, and sends a filtering strategy and a forwarding strategy to the message identification module and the flow filtering and forwarding module; the flow acquisition module sets an acquisition port based on the network configuration information of the acquisition port, acquires a data message, and forwards the flow to the message identification module or the flow filtering and forwarding module based on the start and stop condition of the message identification module; the message identification module labels the traffic after filtering the traffic based on the filtering strategy; and the traffic filtering and forwarding module filters the traffic based on the filtering strategy and forwards the traffic according to the forwarding strategy and the network configuration of the forwarding port. And the flow forwarding fine granularity can be accurately controlled.

The invention discloses an RDMA high-performance network multi-service multiplexing method and system on a virtualized host. The method comprises the following operation steps that network service data programmed through a Verbs interface by using an RDMA protocol and network service data programmed through a socket interface by using a TCP / IP protocol are transmitted at the same time; and performing dynamic balance operation of each level on the network service data, transmitting the network service data subjected to the dynamic balance operation of each level to a physical network card, packaging the network service data to obtain an Ethernet message, and sending the Ethernet message to a physical network.

The invention relates to the technical field of computing centers, in particular to a high-performance RoCE network based on optical connection, which comprises a storage module, a model training module and a model reasoning module which are mutually connected through an optical network, model training data are pre-stored in the storage module; the model training module adopts the model training data to train a to-be-trained model to obtain a business model; and the model reasoning module generates a corresponding processing result by adopting a business model after receiving an external processing demand and returns the processing result. In order to solve the problems that in the prior art, an intelligent computing center lacks scheduling during application, and the same computing task is possibly dispersed on clusters at different positions, so that time delay is increased, in the scheme, a server cluster is planned, three modules are divided according to different storage requirements of an artificial intelligence model, and the time delay is increased; the tasks of the same type are arranged and scheduled in a centralized manner respectively corresponding to three parts of model labeling, training and application, so that the transmission distance is shortened, and the efficiency is improved.

The invention discloses an end-network cooperative reliable multicast method and system, and relates to the technical field of artificial intelligence distributed computing and high-performance network communication. The method specifically comprises the following steps that: a sending end analyzes a target expert set output by a hybrid expert model in a gating manner, constructs a layered multicast bitmap comprising a network node bitmap and a local expert bitmap, and packages the layered multicast bitmap in a message extension header for sending; and the switching equipment analyzes the network node bitmap in the message extension header, executes parallel bitwise AND operation with a locally preset port bitmap mask, determines a target port set needing to be forwarded, and executes message copying and path pruning. The invention aims to solve the problem of highly dynamic target set distribution in a hybrid expert model scene through source end bitmap coding and a switch hardware pruning mechanism, and significantly reduce system overhead and long-tail delay while ensuring high-reliability transmission.

The invention provides a user mode network equipment driving framework under a Linux system and an implementation method. The system comprises network equipment hardware; driving a user mode of the network equipment; and the kernel module comprises a network equipment kernel driver, a memory area and a character equipment kernel driver. And when the character device kernel driver is loaded, creating a virtual character string device, providing an interface for the user mode driver to use, and storing the DMA allocated to the network device and the information of the register storage area. When the user mode driver is loaded, relevant information of the network equipment can be obtained, the network equipment is separated from the kernel driver of the network equipment, and the transceiving function of the kernel driver of the inner network equipment or the transceiving function of a part of queues of the kernel driver of the inner network equipment is stopped; and meanwhile, the DMA and register storage area information of the network equipment is mapped to the memory space of the user mode driver, so that the register and the DMA area are directly operated in the user mode driver, and the network equipment can provide a high-performance network communication function.

The invention provides a high-performance network communication system under a skyvein system and an implementation method thereof, a network communication scheme that a main partition is integrated with a third-party protocol stack bypass default protocol stack and a similar Socket interface is provided in the partition is designed under a system framework of a skyvein operating system, an open source protocol stack is utilized for implementation, and the network communication system can be implemented by reducing memory copies. A hardware unloading function is integrated, the data communication performance is greatly improved, the unidirectional UDP communication performance under the same hardware configuration is more than 5Gbps, and the performance can be improved by 5 times; the support of network services is expanded through a customized interface, and the requirements of new network communication characteristics are met. Coexistence of a third-party protocol stack and a system protocol stack is supported at the same time, when multiple network devices exist, the network device supporting the third-party protocol stack can be selected to use the third-party protocol stack, and other devices still use the system protocol stack; when the application of the third-party protocol stack is not used, the management interface can be automatically switched back to the system protocol stack, and flexible switching is achieved.

The invention discloses an ASIC chip for high-performance AOIP network audio stream transmission, and the interior of the chip comprises a function simulation circuit subsystem, a sensor controller, a clock, and a reset control module. The high-speed peripheral interface subsystem is arranged on the first bus, the hardware acceleration subsystem is arranged on the second bus, the low-speed peripheral interface subsystem is arranged on the third bus, the second bus is connected with the first bus and the third bus, and the function simulation circuit subsystem is connected with the third bus through the sensor controller; the hardware acceleration subsystem comprises a plurality of processor cores, two gigabit or 100M bit medium independent interfaces and a real-time transport protocol RTP transceiving engine; the two gigabit or 100M bit medium independent interfaces are compatible with the simplified medium independent interface, one of the two gigabit or 100M bit medium independent interfaces is used as a main network interface, and the other one is used as a standby network interface. The reliability problem of AOIP network audio stream transmission is solved.

The invention discloses a mixed domain AI high-performance network flow congestion suppression control method, which relates to the technical field of computer networks, and comprises the following steps: a congestion detection step: introducing data packet transmission round-trip time and a switch queue length as congestion detection signals, and a switch queue length management coefficient lambda and a data packet transmission round-trip time management coefficient theta are set. The method comprises the following specific steps: acquiring the current queue length Q of the switch, the data packet round-trip time RTT and the reference round-trip time baseRTT; and when the switch queue length Q is less than the queue length lower limit Kmin or greater than the queue length upper limit Kmax, calculating the data packet congestion marking probability P according to a formula. According to the invention, on the basis of the ECN, RTT indexes which can be collected without special support of network equipment are introduced as congestion detection signals together, and Q and RTT management coefficients are set, so that the response sensitivity of the network to congestion is further improved.

The invention discloses a message forwarding method and system, electronic equipment and a storage medium. The method comprises the steps that an initial message sent by forwarding hardware is received, the initial message is a message meeting a preset shunting condition, and the preset shunting condition is used for shunting a message needing to be processed by a first virtual switch from a target message, the target message is a message of which a forwarding path cannot be determined based on a flow table of the forwarding hardware; generating a first message corresponding to the initial message, and sending the first message to the forwarding hardware, so that the forwarding hardware forwards the first message according to the identification information in the first message; the identification information is used for representing a forwarding path of the initial message. According to the embodiment of the invention, the technical problem that a network card in the related technology is difficult to meet the message processing requirement in a high-performance network scene can be solved.

The present invention relates to a method, system, storage medium and electronic device for building a container network. The system includes a Kubernetes cluster, a CNI plug-in, a resource management system and a cloud platform. The resource management system includes a resource management module, a resource proxy module and an IaaS provider module connected in sequence. The Kubernetes cluster is connected to the CNI plug-in. The resource management module is also connected to the CNI plug-in. The IaaS provider module is connected to the cloud platform. The IaaS provider module is used to connect to the upper cloud platform module through the built-in IaaS provider resource management protocol. Various resource management protocols only provide general description information of resources that are not related to the cloud platform. The capability self-description protocol indicates the network resource management capabilities that the IaaS provider module can provide. In the present invention, the network plug-in is decoupled from the specific cloud platform through the IaaS provider resource management protocol. Only a small amount of corresponding IaaS provider modules are developed to quickly realize the use of the cloud platform VPC network capability Underlay high-performance network plug-in and quickly connect to different cloud platforms.

The invention relates to the technical field of magnetic devices, and discloses a high-performance network transformer and electronic equipment, and the high-performance network transformer comprises a shell which is provided with a containing cavity with an opening in the lower end; the at least one first magnetic ring is configured in the accommodating cavity, a winding coil is wound on the first magnetic ring, and a plurality of lead connectors are led out from the winding coil; the plurality of PINs are designed in a C-shaped structure and are connected in the side wall of the shell, one end of each PIN is provided with a wire winding PIN part protruding out of the side surface of the shell, the lead connector is correspondingly welded with the wire winding PIN part, the other end of each PIN is provided with a patch PIN part, the patch PIN part is led out from the lower end of the shell and corresponds to the wire winding PIN part on the side surface of the shell, and in the first direction, the patch PIN part is welded with the wire winding PIN part in a welding manner; the wire winding PIN parts and the patch PIN parts are arranged in a staggered mode, and the first direction is the extension direction of the upper end and the lower end of the shell. The reliability and the comprehensive performance of the network transformer are improved.

According to the updating method of the FC switch monitoring lookup table based on the LFU, the storage of the monitoring port lookup table is realized by adopting the idea of Robin Hood Hashing, and according to the strategy, a key with a longer detection distance hijacks the position of a key with a shorter detection distance when a Hash conflict occurs during insertion, so that the query performance is optimized, and the updating method is suitable for a high-performance network scene; secondly, modular design is adopted, message IDs existing in a monitoring port lookup table are stored in a grouped mode according to the access frequency through a dictionary and a double-linked list structure, and global sorting is avoided; thirdly, when the message ID stored in the monitoring port query table is full and updated, a double-layer elimination strategy is adopted, the message ID which is least recently used can be efficiently positioned, all frequencies do not need to be traversed, rapid updating is achieved, the method is suitable for a high-concurrency or large-scale FC network communication scene, and therefore the judgment efficiency of the target motion state is improved.

A computationally-efficient system for encoding a message object implements instructions including determining a token of the message object. The token identifies a structure of the message object. The instructions include obtaining a dictionary definition based on the token. The dictionary definition describes the structure of the message. The message includes multiple entries. Each of the entries is characterized by a data type and a field identifier. The dictionary definition defines an order of the entries. The instructions include, according to an order specified by the dictionary definition, selecting each entry of a set of entries in the message in sequence and writing the entry to a byte queue exclusive of the data type and the field identifier. The instructions include initiating transmission of the byte queue over a communications network.

The utility model relates to the technical field of network security cabinets, in particular to a high-performance network security cabinet convenient for wire arrangement. Comprising a cabinet bottom plate and a wire arranging and arranging plate, the wire arranging and arranging plate is installed on the two sides of a cabinet side plate, a wire arranging plate base is arranged at the bottom of the wire arranging and arranging plate, a wire arranging plate middleware is inserted into the side, away from installation supporting feet, of the wire arranging plate base, and a wire arranging plate cover plate is inserted into the side, away from the wire arranging plate base, of the wire arranging plate middleware. Limiting sliding grooves are formed in the two ends of the wiring board base, the two ends of the wiring board middle piece and the two ends of the wiring board cover plate, and inserting fixing blocks are inserted into the limiting sliding grooves. When the cables need to be overhauled, the row of cables corresponding to the cables is found out, the plug-in fixing blocks are taken down from the cable arranging and arranging plate, the cables can be overhauled only by dismantling a row of arranging plate middleware or arranging plate cover plate, and the dismantled area is fixed again by using the plug-in fixing blocks after the overhaul is completed. And the efficiency and the effect of cable arrangement and maintenance are improved.

The utility model relates to the technical field of network security, and discloses a high-performance network security machine device convenient for wire arrangement, which comprises a protection box, a box door is fixedly assembled on the outer wall of the protection box, a mounting rack is arranged on the inner wall of the protection box, the top of the mounting rack is movably connected with a network security device, and the top of the network security device is fixedly connected with the box door. And the inner wall of the protection box is rotationally connected with a rotating shaft. A lifting column is pulled, the lifting column drives a tooth block to move towards the interior of a rotating disc, when the tooth block completely enters the interior of the rotating disc, a tooth groove is separated from the tooth block at the moment, a rotating shaft can rotate, the rotating disc is rotated at the moment, the rotating disc drives the rotating shaft to rotate, the rotating shaft drives a wire winding shaft to rotate, and the wire winding shaft winds an accessory line to the outer wall; and after winding is completed, the lifting column is pressed to reset the structure, so that the clamping relation between the tooth block and the tooth groove is restored, the rotating shaft cannot rotate, and the problem that the circuit is complex is solved.

The invention discloses a database domain name access method and device, equipment and a storage medium. The method comprises the steps that service component topology change information of a target database in a remote direct memory access environment is obtained, and the remote direct memory access environment is deployed on a containerized application management platform; based on the business component topology change information, updating a first service binding relationship corresponding to the containerized application management platform, and determining a second service binding relationship corresponding to the containerized application management platform; and performing domain name access on the target database based on the second service binding relationship. The method can realize efficient access to the database domain name, can adapt to dynamically changing service end points in a general network environment, improves the flexibility of database domain name access, improves the stability of database connection, and reduces the operation and maintenance cost, thereby ensuring efficient and stable operation of a high-performance network, and improving the reliability of the network. And the reliability and user experience of the database system are improved.

This invention discloses a method and apparatus for accurate disturbance prediction and compensation in networked control systems. It includes treating the impact of time delay on the networked control system and external disturbances together as a lumped disturbance, and establishing an equivalent input disturbance state-space model for the networked control system. This method and apparatus target networked control systems with external disturbances, time delays, and packet loss. Through an EID estimator, it effectively reduces the impact of time delay on the performance of the networked control system. A disturbance predictor based on multiple LSTM neural networks is designed to achieve accurate multi-step disturbance prediction, improving the disturbance suppression capability of the networked control system. Furthermore, a networked predictive controller is employed to compensate for the impact of packet loss on the performance of the networked control system in real time, even with continuous packet loss, using multi-step control input prediction values. This improves the control accuracy of the networked control system and meets the requirements of high-performance networked control systems.

The invention relates to an RISC-V server network processing system and method based on DPU (Data Processing Unit) unloading. The system comprises a host comprising an RISC-V main CPU (Central Processing Unit) and a main memory, and a data processing unit DPU connected with the host through an I / O (Input / Output) bus, and completely transferring the processing function of the network data plane from the host to a data processing unit (DPU). According to the invention, the bottleneck faced by the RISC-V server during high-performance network processing in the prior art is overcome, the RISC-V main CPU can be liberated from heavy network data processing tasks, and meanwhile, the internal I / O bottleneck of a chip system (SoC) is bypassed.

The invention discloses a disturbance prediction accurate compensation method and device for a networked control system, and the method comprises the steps: taking the influence of time delay on the networked control system and the external interference of the networked control system as lumped disturbance, and building an equivalent input interference state space model of the networked control system. According to the method and the device, for the networked control system with external interference, time delay and data packet loss, the influence of the time delay on the performance of the networked control system can be effectively reduced through the EID estimator. And a disturbance predictor based on a plurality of LSTM neural networks is designed, so that multi-step accurate prediction of disturbance can be realized, and the disturbance suppression capability of the networked control system is improved. And through the networked prediction controller, the influence of packet loss on the performance of the networked control system can be compensated in real time by using a multi-step control input prediction value under the condition that the data packets are continuously lost, so that the control precision of the networked control system is improved, and the requirements of the high-performance networked control system are met.

The utility model discloses a high-performance network core switch, which comprises a switch, a groove is arranged on the switch, guide grooves are oppositely arranged on the groove, and the high-performance network core switch further comprises a clamping shell which is in sliding connection with the guide grooves through oppositely arranged guide parts; the switch is provided with a clamping groove in the sliding stroke of the clamping shell, and the clamping shell is provided with a protruding part matched with the clamping groove. When the clamping shell slides on the guide grooves to the end of the stroke, the clamping shell is clamped with the clamping grooves through the protruding parts. According to the high-performance network core switch provided by the utility model, when the heat dissipation holes have much dust, the clamping shell slides towards the outer side of the switch, so that the protruding part of the clamping shell is separated from the clamping groove until the switch is completely separated from the clamping shell through sliding, and at the moment, the clamping shell can be taken away to carry out subsequent cleaning on the heat dissipation holes in the clamping shell; according to the device, the heat dissipation holes can be independently detached in a sliding clamping mode, operation is convenient, the heat dissipation holes can be conveniently cleaned, and the cleaning efficiency is improved.

The present invention discloses a high-performance network security cabinet that is convenient for wire management, and relates to the technical field of network security cabinets. In order to solve the problem that the connecting wires between the various components of the high-performance network security cabinets currently on the market are scattered inside the cabinet, and the connecting wires are prone to knotting, which is not only inconvenient for line maintenance, but also affects the aesthetics of the interior of the cabinet. The upper end of the outer side of the network security cabinet body is fixedly connected with a control device, the side of the lower end of the network security cabinet body is fixedly connected with a fire hydrant placement box, the upper end of the dry powder fire hydrant is fixedly connected with a control solenoid valve, the lower end of the network security cabinet body is located on the inner side of the support base and is fixedly connected with a universal wheel, one end of the mounting plate is fixedly connected to the inner side of the movable cabinet door, and the inner side of the cooling fan at the upper end of the network security cabinet body is fixedly connected with a temperature sensor and a smoke sensor.

This invention discloses a high-performance network device with multi-protocol fusion, comprising an application layer, a network fusion layer, a data link layer, and a physical layer connected in sequence. The application layer is configured to perform high-speed data interaction with the network fusion layer via the AXI-Stream protocol. The network fusion layer is configured to identify and parse the input data packets, converting data frames from Interlaken, Aurora, Ethernet, and FC protocols into a unified AXI-Stream format and providing error detection flags to the upper layer. The data link layer is configured to generate, encapsulate, and encode frame structures according to the frame format requirements of different network protocols based on the standardized data stream output by the network fusion layer. The physical layer includes a high-speed SerDes module for implementing serial-to-parallel conversion. The high-performance network device with multi-protocol fusion provided by this invention significantly improves system parallelism, transmission bandwidth, and reliability, while reducing CPU processing burden and transmission latency by implementing multi-protocol fusion and a unified interface at the hardware level.

The application discloses a high-performance network acceptance system and method of a smart data center, and relates to the field of data center network testing.The system adopts a layered architecture integrating hardware and software, the hardware layer includes an acceptance control node, a test terminal node and a data acquisition probe; the core software is integrated in the acceptance control node and includes a multi-protocol adaptation index system module, a real-time analysis module and a result output module connected in sequence.The method is executed through five steps of preprocessing, protocol and index configuration, layered automatic testing, real-time analysis and closed-loop evaluation, covers three dimensions of parameter surface, sample surface and set communication, fuses active testing and passive monitoring dual-mode verification, and is compatible with IB and RoCE dual protocols and multi-vendor hardware.The application can automatically complete topology discovery, index verification, performance testing and abnormal positioning, accurately simulates smart business traffic, outputs a standardized acceptance report and optimization suggestions, and significantly improves acceptance efficiency and accuracy.

The invention discloses a differential entropy guided optical diffraction neural network layout optimization design method. The method comprises the following steps: step 1, defining an ODNN expression power index; 2, designing a Gaussian process regression model; 3, constructing an initial data set; and step 4, optimizing the flow guided by differential entropy. According to the method, the performance index of the ODNN expression ability is constructed through the differential entropy, the performance index is fused into the Bayesian optimization framework, and meanwhile, the dual requirements of performance improvement and scale simplification are balanced by adopting a dual-target combination acquisition strategy. According to the method, in a Fast-MNIST data set classification and reconstruction task, the accuracy Acc and the reconstruction quality SSIM are improved by 6.3% and 5.4% respectively, the number of high-performance layouts obtained in the optimization process is doubled, and the output layout scheme is richer; in the obtained high-performance network layout, the average number of used neurons is reduced by 34%, and the network layout is remarkably simplified.

The application discloses an end-network cooperative reliable multicast method and system, and relates to the technical field of artificial intelligence distributed computing and high-performance network communication. The application specifically comprises the following steps: a sending end analyzes a target expert set of a mixed expert model gate output, constructs a layered multicast bitmap containing a network node bitmap and a local expert bitmap, and sends the layered multicast bitmap encapsulated in a message extension header; a switching device analyzes the network node bitmap in the message extension header, performs parallel bitwise AND operation with a locally preset port bitmap mask, determines a target port set to be forwarded, and performs message replication and path pruning. The application aims to solve the problem of highly dynamic target set distribution in the mixed expert model scene through source end bitmap coding and switching hardware pruning mechanism, guarantee high reliable transmission, and significantly reduce system overhead and long tail delay.

The present application relates to a method for scheduling multiple variable paths and variable bandwidth based on a high-performance network. The method prioritizes the transmission of more urgent requests based on their priorities, and transmits more user requests within a transmission cycle. The method uses as many node-disjoint paths as possible to transmit data on the network, avoiding congestion caused by excessive load on a certain link, making more effective use of the entire network, improving overall bandwidth utilization, and increasing data transmission reliability. The method also takes into account the delay of path switching, making bandwidth scheduling more practical.

The present application discloses a standard compliance test and verification system for high-performance network audio streams, including a test detection system, and a variety of sound source distribution systems, audio jump systems, multi-form high-performance network audio stream input systems, multi-form high-performance network audio stream processing systems, multi-form high-performance network audio stream output systems, and multi-brand network switching systems that are respectively connected thereto. According to the embodiments of the present application, problems such as non-interoperability and non-openness between various high-performance network audio stream transmission protocols in the current professional audio industry field are solved, and the test and verification of standard compliance indicators such as electrical performance, network transmission, and control performance are supported for multi-brand, multi-type, and multi-form high-performance network audio stream products, improving the test efficiency.

The invention discloses a high-performance network acceptance system and method for an intelligent computing data center, and relates to the field of data center network testing. The system adopts a layered architecture integrating a hardware layer and software, wherein the hardware layer comprises an acceptance control node, a test terminal node and a data acquisition probe; the core software is integrated on an acceptance control node and comprises a multi-protocol adaptation index system module, a real-time analysis module and a result output module which are in signal connection in sequence. According to the method, five-step execution of preprocessing, protocol and index configuration, layered automatic testing, real-time analysis and closed-loop evaluation is carried out, three dimensions of a parameter surface, a sample surface and integrated communication are covered, active testing and passive monitoring dual-mode verification are fused, and IB and RoCE dual protocols and multi-manufacturer hardware are compatible. According to the invention, topology discovery, index verification, performance test and abnormity positioning can be automatically completed, intelligent calculation service flow is accurately simulated, a standardized acceptance report and optimization suggestion are output, and the acceptance efficiency and accuracy are significantly improved.