SHANDONG YINGXIN COMPUTER TECHNOLOGIES CO., LTD.

The present application discloses a power supply redundancy control system for a GPU server, comprising a power supply redundancy module, a BMC, a CPLD and a GPU module. The power supply redundancy module comprises a first PSU and a second PSU, and the GPU module comprises several GPUs, the first PSU is connected to the CPLD by means of a first bus. The second PSU is connected to the CPLD by means of a second bus. The BMC is connected to the CPLD by means of a first I2C bus and a second I2C bus, and sends heartbeat information to the CPLD. The CPLD is connected to the BMC by means of a third bus and a fourth bus, and the CPLD is connected to the several GPUs by means of a third I2C bus. In the present application, when the BMC is abnormalous or restarted, the CPLD can control the overall power consumption of the server, and can also ensure that the server will not go down, reducing the loss brought to a user due to the BMC being abnormalous or restarted.

Disclosed are a distributed quantum imaging method, apparatus and system, and a computer-readable storage medium. The distributed quantum imaging system comprises a plurality of laser devices that are placed at different spatial positions, a plurality of spatial light modulators, a detector and an imaging processor, wherein each laser device uniquely corresponds to one spatial light modulator. Each spatial light modulator is used for modulating a light field parameter generated by a corresponding laser device during each measurement process, and projecting a modulated light signal onto an object to be measured; the detector is used for collecting transmitted light obtained after an output light signal of each laser device passes through said object, converting the transmitted light into a corresponding measurement electrical signal and sending the measurement electrical signal to the imaging processor; and the imaging processor is used for performing reconstruction by using a compressed sensing algorithm, a sensing matrix that is constructed on the basis of light field information during a plurality of measurement processes, and the measurement electrical signal, so as to obtain information of said object. By means of the present application, the quantum imaging efficiency and the quantum imaging resolution can be effectively improved.

A Peripheral Component Interconnect express (PCIe) fault auto-repair method is provided. According to the method, when a PCIe link in a system runs, an operation state of the system is monitored by acquiring a Correctable Error (CE) error count and an Uncorrectable Error (UCE) error count in the PCIe link; when the CE error count reaches a corresponding error threshold value, or the UCE error count reaches a corresponding error threshold value, an error device is removed from the system to avoid continuous adverse influence of the error device on operation of the system. Moreover, an SI parameter register of the error device is modified according to pre-stored optimization parameters of all PCIe devices in a server, the SI parameter of the error device is automatically optimized, and the error device is re-accessed to the system after the PCIe fault repair.

A key name generation method includes constructing a member structure library according to a set data structure template and member categories contained in various types of components, wherein the member structure library contains member structure information corresponding to each member category; upon acquiring a reading instruction for data, retrieving, from the member structure library on the basis of a data identifier carried in the reading instruction for the data, target member structure information matching the data identifier; and determining, on the basis of the target member structure information and a structure array identifier contained therein, key description information corresponding to the target member structure information, and taking the key description information as a key name of the data. A key name generation apparatus and a medium are also disclosed herein.

A method and system for sharing a same GPU by a plurality of services, a device and a storage medium are provided. The method includes: in response to receiving a request of creating GPU services, creating the corresponding GPU services according to the request, creating GPU Pods of a corresponding quantity according to the GPU services, and associating the GPU services with the GPU Pods (S 1 ); creating Kubernetes Pods according to a configuration of the GPU Pods, associating the Kubernetes Pods with the GPU Pods (S 2 ); in response to receiving a calculating request, according to the calculating request, determining a specification of a GPU graphic memory or GPU time slice required to be applied for, and comparing with a threshold specified by the GPU services (S 3 ); in response to the specification of the GPU graphic memory or time slice being less than the threshold, reading current residual resource amounts of the GPU Pods and the Kubernetes Pods, and comparing with the specification of the GPU graphic memory or time slice (S 4 ); and in response to the specification of the GPU graphic memory or time slice being less than a sum of the current residual resource amounts of the GPU Pods and the Kubernetes Pods, according to a current resource utilization rate, dispatching the GPU Pods and the Kubernetes Pods for calculation (S 5 ).

If a resource of a PCI device needs to be designated and allocated with a fixed address range and a designated and allocated address is not within a preset address interval range, a resource address of the PCI device is merely allocated according to the designated and allocated address thereof; and when resource address information of a parent device PCI bridge of the PCI device is calculated, the resource address information of the parent device PCI bridge of the PCI device is calculated merely according to another child device, other than this PCI child device, of the parent device PCI bridge, rather than taking the designated and allocated fixed address range of the PCI device into consideration, so as to prevent a length error of a resource allocation address of the parent device PCI bridge resulting in a serious system error.

Disclosed is a multi-processor server node cross-board signal interconnection device. The device includes a middle backboard, a plurality of first signal connectors arranged on a surface of the middle backboard, a plurality of second signal connectors arranged on a back surface of the middle backboard and having the same distribution region as all the first signal connectors, a plurality of signal leading-out terminals arranged on all node single boards and configured to be connected to all the second signal connectors, and a plurality of signal interconnection cables connected between the second signal connectors. In this way, all the second signal connectors transmit signals on the middle backboard by means of the signal interconnection cables. All the second signal connectors and the first signal connectors are arranged in the same region, such that a risk of signal interference is eliminated.

Disclosed are an object recognition method and apparatus, and a device and a medium. The method comprises: by means of a convolutional layer in a convolutional neural network, extracting pre-selected feature information of objects to be recognized, and performing dimension reduction processing on the pre-selected feature information by using a pooling layer, to obtain feature information; processing the feature information by means of a fully connected layer, to obtain processed feature information, extracting a target feature value from the processed feature information, and compiling statistics on the number of objects to be recognized; by means of the fully connected layer and according to the target feature value, determining corresponding target queue information from pre-stored queue information; and by means of the fully connected layer, performing a comparative analysis on the processed feature information and reference feature information of each sample object in a queue corresponding to the target queue information, and outputting a corresponding recognition result. Therefore, the time consumed by a fully connected layer in a convolutional neural network can be shortened, the time delay of a recognition process is reduced, and the recognition efficiency is improved.

The present application discloses a task status updating method including: listening for K8S events by using a K8S event listener to obtain a Pod status change event, and generating a Pod status update event based on the Pod status change event (S 12 ); listening for the Pod status update event by using a Pod status change event listener, determining a Pod status corresponding to the Pod status update event as a Pod status of a corresponding target task in the deep learning platform when the Pod status update event is listened out, and generating a task status update event (S 13 ); and listening for the task status update event by using a task status change event listener, and updating a current status of the target task as the Pod status of the target task when the task status update event is listened out (S 14 ).

The present disclosure discloses an encoding and decoding method, apparatus and device based on an erasure pool, and a computer readable storage medium. The method includes: storing, upon receiving original data that needs to be subjected to data encoding and decoding, the original data into a shared memory; sending an encoding and decoding signal to an encoding and decoding process, so that the encoding and decoding process obtains the original data from the shared memory after receiving the encoding and decoding signal, and calls a hard codec plug-in to perform a hard encoding and decoding operation; controlling, if the hard encoding and decoding operation fails, the encoding and decoding process to call a soft codec plug-in to perform a soft encoding and decoding operation; and obtaining, after the encoding and decoding process completes the encoding and decoding operation, an encoding and decoding result from the shared memory.