Abstract: The disclosure provides an airflow regulation and control apparatus. A ventilation hole is formed in a windshield in a penetrating manner; a rectangular plate is mounted on the windshield; the rectangular plate is lifted by means of power provided by a lifting wing disposed thereon, and the height of the rectangular plate when moving upwards exceed the height of an upper edge of a main body portion of the windshield; when wind strength is low, the rectangular plate is located at a lower portion under the action of gravity and shields the ventilation hole; when the wind strength is increased, the lifting wing is blown by an airflow to generate upwards lifting force, and drives the rectangular plate to moved upwardly so as to shield an opening in an upper portion of the windshield and block the airflow.

Abstract: An out-of-band updating method and system of an Expander, applied to a controller connected to the expander by means of a communication bus. The method comprises: receiving firmware updating data sent by a user (S101); fragmenting the firmware updating data (S102); and only when determining that the expander is in an idle state, sending the fragmented data until each piece of data is sent to the Expander, such that the expander completely receiving the firmware updating data completes updating by restarting (S103). The data transmission speed is facilitated to be improved, the normal operation of a service is ensured, and the conditions of influencing state monitoring and log loss are avoided.

Abstract: The present invention discloses a filtering method and system of parallel computing results, through simultaneously generating the input value of the first valid position fvp of each fragment, and simultaneously computing to obtain the output result corresponding to the input value of each first valid position fvp with the respective first valid position fvp of each fragment, and according to the output result of the first valid position fvp of the first fragment, the parallel computing results are filtered through the manner of selecting the output results of the second to the S-th fragments in sequence, to finally obtain correct parallel computing results. In the present invention, through adopting the manner of parallel filtering, the original serial filtering computation is changed to parallel computation of S fragments, the computing time is only one S-th of the original time, thereby improving the computing efficiency and satisfying the timing requirements of parallel computation.

Abstract: A load balancing method, apparatus and device for a parallel model training task, and a computer-readable storage medium, includes: acquiring data traffic and a theoretical computational amount of each network layer in a target model; determining a theoretical computing capability of each computing device, and obtaining an initial computational amount corresponding to each computing device according to the theoretical computing capability and the theoretical computational amount; performing a load balancing operation according to the initial computational amount by using multiple device critical layer position division rule, so as to obtain a plurality of initial balancing schemes; compiling statistics on time performance parameters corresponding to the initial balancing schemes, and determining an intermediate balancing scheme from the initial balancing schemes according to the time performance parameters; and adjusting the intermediate balancing scheme according to the data traffic, so as to obtain a final

Abstract: The present invention provides a method and apparatus for data caching. The method comprises: output matrixes are acquired one by one, a plurality of acquired output matrixes are written alternately into two queue sets of a first cache unit according to a sequence in which the output matrixes are acquired, and the output matrixes stored line by line in a first cache unit are written into a second cache unit one by one, according to the sequence in which the output matrixes are written into the second cache unit, valid data of each output matrix of the second cache unit is determined one by one according to preset parameters, and the valid data of each output matrix is written into a third cache unit, and the valid data of the output matrixes stored in the third cache unit are configured to be sequentially written into a memory according to a sequence in which the valid data are written into the third cache unit.

Abstract: A nonvolatile memory scheduling method, system and device, and a computer-readable storage medium. The method includes: obtaining a utilization rate of a current memory space (S101); when the utilization rate is less than a first threshold, determining a first space to be converted of the memory space, converting the first space into a high-speed storage space, and storing data in a conventional storage device into the high-speed storage space (S102); and when the utilization rate is greater than a second threshold, determining a second space to be converted of the high-speed storage space, storing data stored in the second space into the conventional storage device, and converting the second space into the memory space (S103).

Abstract: Provided is a safety control method for an AI server, which is applied in an FPGA. The method includes: obtaining a current electrical current and a current power of a GPU in the AI server according to a preset frequency; determining whether the GPU satisfies a first control privilege transfer requirement; when the GPU satisfies the first control privilege transfer requirement, taking over control privilege of a heat dissipation system from a BMC; and controlling the heat dissipation system according to the current electrical current and the current power of the GPU; wherein the first control privilege transfer requirement includes: the current electrical current of the GPU exceeds a preset electrical current, or a rate of change of the current electrical current of the GPU exceeds a preset rate of change of electrical current, or the current power of the GPU exceeds a first preset power.

Abstract: Disclosed are a method for realizing an nGraph framework supporting an FPGA backend device, and a related apparatus. The method includes: integrating an OpenCL standard API library into an nGraph framework; creating, in the nGraph framework, an FPGA backend device creation module for registering an FPGA rear-end device, initializing an OpenCL environment and acquiring the FPGA backend device; creating, in the nGraph framework, an FPGA buffer space processing module for opening up an FPGA buffer space and for reading and writing an FPGA cache; creating, in the nGraph framework, an OP kernel implementation module for creating an OP kernel and compiling the OP kernel; and creating, in the nGraph framework, an FPGA compiling execution module for registering, scheduling and executing the OP kernel.

Abstract: A method and apparatus for remotely updating firmware in batches, and a computer-readable storage medium. Wherein the method is applied to a DHCP server connected with a Baseboard Management Controllers (BMC) of each server node. The DHCP server assigns IP addresses to each server node in advance, and records a BMC MAC address of each server node at the same time. When detecting that a firmware version to be matched stored in itself is updated, authorized server nodes and corresponding BMC IP addresses are determined on the basis of each BMC MAC address and authorization state. The firmware version of the corresponding authorized server nodes is read according to the BMC IP address of each authorized server, and a plurality of servers to be upgraded is determined by comparing the firmware version of each authorized server node with the firmware version to be matched.

Abstract: A data storage method, apparatus, and device, and a readable storage medium. The method includes: after a random access memory is powered on, obtaining target data to be stored in a fixed storage address of the random access memory; determining a target transmission mode from a bit value change transmission mode and a bit value fixed transmission mode, wherein the target transmission mode is different from a historical transmission mode determined after the random access memory is powered on last time; and transmitting the target data from and to the random access memory according to the target transmission mode. The method can prevent data from being stolen after power-down of the target data, and guarantees the data security.

Abstract: A method and system for identifying garbage data, an electronic device, and a storage medium. The method includes: uploading an object to a distributed object storage system; acquiring a head object identifier in an index storage pool of the distributed object storage system; querying a data storage pool for a target data group corresponding to the head object identifier; marking a tail object corresponding to a tail object identifier in the target data group as a target tail object; and marking tail objects in the data storage pool other than the target tail object as garbage data.

Abstract: The method includes: an FPGA board feeds back the quantity of controllers and the total quantity of DDR memories after receiving a hardware information acquisition request from a host; after a data space application request is received from the host, on the basis of the data space application request, perform data slice processing on data to be calculated, wherein the data space application request carries the dedicated application space capacity of each DDR and the data to be calculated, and the total quantity of slices of the data to be calculated is the same as the total quantity of DDR memories; and transmit each sliced data to a corresponding DDR space, and according to a data storage position of the sliced data in each DDR, read the data from the DDR memory space in parallel by means of the plurality of controllers and calculate same.

Abstract: Disclosed are a distributed block storage service command processing method, apparatus, a device and a medium. The method includes: by means of tgtadm, acquiring a rbd command of a block storage service, parsing the rbd command to obtain a corresponding task parameter, and then transmitting the task parameter to an ISCSI target (S11); by means of tgtd, starting a dedicated thread for processing a task corresponding to the task parameter, and monitoring a real-time processing result of the current task (S12); and returning the real-time processing result to the tgtadm by means of the ISCSI target, whereby the tgtadm displays the real-time processing result on a human-computer interface (S13).

Abstract: Provided are a Central Processing Unit (CPU) resource reservation method, apparatus, and device, and a computer-readable memory medium. The method includes: selecting a target working node according to a received Virtual Machine (VM) startup request; obtaining a total number of virtual cores and a number of allocatable physical cores in the target working node statistically; performing calculation to obtain an available CPU quota according to the total number of virtual cores and the number of allocatable physical cores; and performing CPU resource reservation configuration on the target working node by use of the available CPU quota. According to the CPU resource reservation method, the reservation of CPU resources in a VM system may be implemented more flexibly and efficiently.

Abstract: An abnormal index node (inode) dynamic repair method, system, and apparatus, and a readable memory medium. The method includes: obtaining an overall memory structure of an abnormal inode in a memory, and converting the overall memory structure into an editable file; determining an abnormal field in the editable file according to error reporting information; writing an expected value of a file system to a position of the abnormal field in the editable file to obtain a repaired editable file; and importing the repaired editable file back to the memory.

Abstract: An image processing method is provided, which is applied to a deep learning model. A cache queue is provided in front of each layer of the deep learning model; a plurality of computation tasks are preset for each layer of the deep learning model in advance, and are configured for computing weight parameters and corresponding to-be-processed data in a plurality of channels in each corresponding layer in parallel, and storing a computation result into a cache queue behind the corresponding layer thereof; in addition, as long as the cache queue in front of the layer includes the computation result stored in the previous layer, the layer can obtain the to-be-processed data from the computation result, subsequent computation is performed, and a parallel pipeline computation mode is also formed between the layers.

Abstract: A method, an apparatus, a device, and a medium for creating a thick-provisioned volume. The method includes: creating a thin volume; acquiring a current workload of a cluster; determining a volume filling thread quantity corresponding to the thin volume and a Quality of Service (QOS) policy based on the workload and a preset load threshold; filling the thin volume using a volume filling thread determined according to the volume filling thread quantity and the QOS policy, so as to complete the creation of a thick-provisioned volume.

Abstract: A method and system for identifying garbage data, an electronic device, and a storage medium. The method includes: uploading an object to a distributed object storage system (S101); acquiring a head object identifier in an index storage pool of the distributed object storage system (S102); querying a data storage pool for a target data group corresponding to the head object identifier (S103); marking a tail object corresponding to a tail object identifier in the target data group as a target tail object (S104); and marking tail objects in the data storage pool other than the target tail object as garbage data (S105). The method can identify garbage data in a distributed object storage system.

Abstract: There are provided a device for preventing an entire cabinet from being powered down, a cabinet, and a method for preventing an entire cabinet from being powered down. The device includes a diode and a first protection circuit. An anode of the diode is grounded, a cathode of the diode is connected to an external voltage supplying copper busbar via the first protection circuit, and the diode is connected in parallel with an external node. The first protection circuit is arranged to monitor a current flowing through the diode, and cut off a connection line between the diode and the voltage supplying copper busbar in a case that the current exceeds a preset fusing current threshold.

Abstract: A method for monitoring and managing a rack server is provided. The method includes: editing, in a rack management controller RMC for monitoring and managing the rack server in a centralized manner, a configuration file based on configuration information regarding the numbers and positions of server nodes, node mid-planes, fan control boards, fans and power supplies in a rack; and automatically monitoring and managing, by the RMC, components based on the configuration file by the RMC automatically acquiring accurate numbers and positions of nodes and power supplies corresponding to the rack configuration to select a corresponding cooling-speed regulation curve. With the method, racks with different configurations can be dynamically monitored by modifying the configuration file including the configuration information of the rack and uploading the modified configuration file.