公开(公告)号：US20230114242A1

公开(公告)日：2023-04-13

申请号：US18046151

申请日：2022-10-12

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Weibin Lin ,  Tao Li ,  Haocheng Liu ,  Lixia Xu ,  Sheng Li

IPC: G06F12/0815

Abstract: A data processing method and apparatus, and a heterogeneous system, pertaining to the field of computer technologies are provided. The heterogeneous system includes a processor connected to an accelerator. A secondary memory is connected to the accelerator. The processor is configured to write to-be-processed data into the secondary memory and trigger the accelerator to access and process the to-be-processed data stored in the secondary memory according to a processing instruction. The accelerator is configured to write a processing result of the to-be-processed data into the secondary memory and to trigger the processor to read the processing result. Processing efficiency is enhanced by reducing the number of times of interaction between the processor and the accelerator and simplifying the procedure for data processing.

公开(公告)号：US20240296067A1

公开(公告)日：2024-09-05

申请号：US18648093

申请日：2024-04-26

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Shanggang Xie ,  Wenkai Ling ,  Weibin Lin ,  Shengyu Shen ,  Xiaoyu Chen

IPC: G06F9/48 ,  G06F9/50

CPC classification number: G06F9/4881 ,  G06F9/5016 ,  G06F9/505 ,  G06F9/5094 ,  G06F2209/485 ,  G06F2209/503

Abstract: A computing device includes a scheduling apparatus and at least two processing unit sets, each processing unit set includes at least one processing unit, and the scheduling apparatus is communicatively connected to processing units in the at least two processing unit sets. The data processing method in this application includes: The scheduling apparatus receives a data processing request, where the data processing request includes a function identifier, and the function identifier indicates a function that needs to be called to process the data processing request; and further, the scheduling apparatus determines a first processing unit set from the at least two processing unit sets of the computing device based on the function identifier, and determines a target processing unit in the first processing unit set, and the determined target processing unit processes the data processing request, to obtain a data processing result.

公开(公告)号：US20210397559A1

公开(公告)日：2021-12-23

申请号：US17464093

申请日：2021-09-01

Applicant: Huawei Technologies Co., Ltd.

Inventor： Tianxiang Chen ,  Weibin Lin ,  Xinyu Hou

IPC: G06F12/0875 ,  G06F13/38 ,  G06F3/06 ,  H04L12/861

Abstract: A data processing method comprising: After receiving an ith Peripheral Component Interconnect Express (PCIe) packet, a network interface card stores a jth instruction segment in a jth storage unit that is in a first storage area. When all n instruction segments of a first send queue entry (SQE) are stored in the first storage area, the network interface card obtains the first SQE, an identifier of a queue pair (QP) to which the first SQE belongs, and a location identifier of the first SQE in the QP according to the instructions in n storage units in the first storage area; the network interface card performs data processing based on the identifier of the QP to which the first SQE belongs and the location identifier of the first SQE in the QP.

公开(公告)号：US20240296327A1

公开(公告)日：2024-09-05

申请号：US18646489

申请日：2024-04-25

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Shengyu Shen ,  Zeshan Chang ,  Jingyan Wang ,  Guangcan Mi ,  Xin Liu ,  Qihang Duan ,  Wenkai Ling ,  Weibin Lin

IPC: G06N3/08

CPC classification number: G06N3/08

Abstract: This application provides a model training system and method. The system includes a first group, where the first group includes an MEMS and S×C processors, S is a quantity of nodes in the first group, C is a quantity of processors in one node, and both S and C are positive integers, the MEMS, configured to construct an optical transmission channel between any two of the S nodes, and the S×C processors, configured to jointly train a model. In one iteration of joint model training, the S×C processors are configured to run model training in respective processors, to obtain respective corresponding data. At least two of the S×C processors transmit target data through the optical transmission channel. A processor that receives the target data may be configured to adjust a parameter for model training in the processor based on the target data.

公开(公告)号：US11782869B2

公开(公告)日：2023-10-10

申请号：US17843825

申请日：2022-06-17

Applicant: Huawei Technologies Co., Ltd.

Inventor： Shengyu Shen ,  Yumin Wu ,  Weibin Lin ,  Jiangle Huang ,  Wenkai Ling

IPC: G06F15/173 ,  G06F3/06 ,  G06F13/28 ,  H04L67/1097

CPC classification number: G06F15/17331 ,  G06F3/0631 ,  G06F13/28 ,  H04L67/1097

Abstract: A data transmission method, performed by a receive end, includes: receiving notification information that is about to-be-transmitted data and that is sent by a transmit end, where the notification information includes a size of the to-be-transmitted data of the transmit end; generating a plurality of RDMA read requests based on the size of the to-be-transmitted data; and sending the plurality of RDMA read requests to the transmit end, where a bandwidth occupied by data read by the plurality of sent RDMA read requests is less than or equal to an ingress bandwidth of the receive end.

公开(公告)号：US20230153264A1

公开(公告)日：2023-05-18

申请号：US18150647

申请日：2023-01-05

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Weibin Lin ,  Yaoguo Hu ,  Shengyu Shen

IPC: G06F15/173 ,  H04L67/1097

CPC classification number: G06F15/17331 ,  H04L67/1097

Abstract: A data transmission method is provided. The method includes: a network interface card of a source device obtains a first notification message and a second notification message, wherein the first notification message indicates that a first to-be-processed remote direct memory access (RDMA) request exists in a first queue of the source device, the first queue stores a request of a first service application in the source device, the second notification message indicates that a second to-be-processed RDMA request exists in a second queue of the source device, and the second queue stores a request of a second service application in the source device; and the network interface card determines a processing sequence of the first queue and the second queue based on service levels, and sends the first to-be-processed RDMA request and the second to-be-processed RDMA request to a destination device according to the processing sequence.

公开(公告)号：US12259843B2

公开(公告)日：2025-03-25

申请号：US17561019

申请日：2021-12-23

Applicant: Huawei Technologies Co., Ltd.

Inventor： Xinyu Hou ,  Qun Jia ,  Weibin Lin

IPC: G06F13/42 ,  G06F13/40 ,  G06F15/17 ,  G06F15/173 ,  H04L49/109

Abstract: An artificial intelligence (AI) switch chip includes a first AI interface, a first network interface, and a controller. The first AI interface is used by the AI switch chip to couple to a first AI chip in a first server. The first network interface is used by the AI switch chip to couple to a second server. The controller receives, through the first AI interface, data from the first AI chip, and then sends the data to the second server through the first network interface. By using the AI switch chip, when a server needs to send data in an AI chip to another server, an AI interface may be used to directly receive the data from the AI chip, and then the data is sent to the other server through one or more network interfaces coupled to the controller.

公开(公告)号：US20240348686A1

公开(公告)日：2024-10-17

申请号：US18677139

申请日：2024-05-29

Applicant: Huawei Technologies Co., Ltd.

Inventor： Weibin Lin ,  Wenkai Ling ,  Shengyu Shen ,  Shanggang Xie ,  Qiuping Pan ,  Xiaoyu Chen

IPC: H04L67/1097 ,  H04L49/90

CPC classification number: H04L67/1097 ,  H04L49/90

Abstract: A remote data access method is applied to a first device that is coupled to and communicates with a second device. The first device and the second device each are configured with a network interface card (NIC) having a remote direct memory access (RDMA) function. The method is performed by a NIC of the first device, and the method includes: generating a first packet including a type identifier and a function identifier; and sending the first packet to the second device. The type identifier in the first packet indicates that the first packet is a remote memory function call (RMFC)-type packet, and the RMFC-type packet indicates the second device to perform a preset operation on local data of the second device. The function identifier in the first packet indicates an operation type of the preset operation.

公开(公告)号：US11620227B2

公开(公告)日：2023-04-04

申请号：US17464093

申请日：2021-09-01

Applicant: Huawei Technologies Co., Ltd.

Inventor： Tianxiang Chen ,  Weibin Lin ,  Xinyu Hou

IPC: G06F12/00 ,  G06F12/0875 ,  G06F3/06 ,  G06F13/38 ,  H04L49/90

Abstract: A data processing method comprising: After receiving an ith Peripheral Component Interconnect Express (PCIe) packet, a network interface card stores a jth instruction segment in a jth storage unit that is in a first storage area. When all n instruction segments of a first send queue entry (SQE) are stored in the first storage area, the network interface card obtains the first SQE, an identifier of a queue pair (QP) to which the first SQE belongs, and a location identifier of the first SQE in the QP according to the instructions in n storage units in the first storage area; the network interface card performs data processing based on the identifier of the QP to which the first SQE belongs and the location identifier of the first SQE in the QP.