公开(公告)号：US20240370521A1

公开(公告)日：2024-11-07

申请号：US18773914

申请日：2024-07-16

Applicant: Huawei Technologies Co., Ltd.

Inventor： Qing Zhou ,  Feng Yu ,  Jian He

IPC: G06F17/15 ,  G06F8/41

Abstract: An operator processing method includes obtaining a real-time shape of any to-be-output first tensor by combining in real time one or more micro-operators in a pre-constructed micro-operator library. Then, a micro-operator included in one combination (for example, a combination with optimal performance because different combinations have different performance) is selected for execution. Micro-operators in the micro-operator library are pre-compiled. Therefore, a compiler is not needed. In addition, shapes of the micro-operators are fixed and different, and are used as a “basis” of “shape space”.

公开(公告)号：US20240154802A1

公开(公告)日：2024-05-09

申请号：US18415995

申请日：2024-01-18

Applicant: Huawei Technologies Co., Ltd.

Inventor： Jian He

IPC: H04L9/08 ,  G06N5/04

CPC classification number: H04L9/088 ,  G06N5/04

Abstract: In accordance with an embodiment, a method includes: obtaining a plurality of execution operators from a first model, wherein the plurality of execution operators comprise a first operator, and the first operator indicates decryption processing logic; and sequentially executing the plurality of execution operators based on a hierarchical relationship between the plurality of execution operators. Sequentially executing the plurality of execution operators includes: executing the first operator comprising decrypting, based on the decryption processing logic, first data corresponding to the first operator to obtain second data, and executing, based on the second data, one or more execution operators arranged after the first

公开(公告)号：US11419171B2

公开(公告)日：2022-08-16

申请号：US17126799

申请日：2020-12-18

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Jianjian Zhu ,  Feng Li ,  Qiubin Song ,  Jian He ,  Jiqing Li

IPC: H04W76/15 ,  H04W76/11 ,  H04L45/24 ,  H04L61/5007

Abstract: This application provides a subflow establishment method. A local-end device obtains an identifier of an operator currently accessed by a network interface card of the local end device. The local-end device obtains a plurality of IP addresses of a peer-end device and identifiers of operators to which the plurality of IP addresses are homed; and establishes, by matching a plurality of operators corresponding to the local-end device and the plurality of operators corresponding to the peer-end device, a subflow between an IP address pair homed to a same operator. In this way, the established subflow does not cross operators as far as possible. This reduces negative impact of cross-operator communication of some subflows on transmission performance of a multipath connection, thereby better leveraging an advantage of the multipath connection in transmission performance in comparison with a single-path connection.

公开(公告)号：US10983731B2

公开(公告)日：2021-04-20

申请号：US16789141

申请日：2020-02-12

Applicant: Huawei Technologies Co., Ltd.

Inventor： Jian He ,  Xiaoke Ni ,  Yimin Yao

IPC: G06F3/06 ,  G06F13/42 ,  G06F13/40 ,  G06F13/38

Abstract: A packet forwarding method, device, and system for use in non-volatile memory express (NVMe) over fabric are described. According to the method, a first packet sent by a control device is received; a second packet is generated according to the first packet; a network port for forwarding the second packet is selected according to an identity of a target non-volatile memory; and the second packet is encapsulated into a packet that meets an NVMe over fabric forwarding requirement, and the encapsulated second packet is sent by using the selected network port.

公开(公告)号：US09329783B2

公开(公告)日：2016-05-03

申请号：US14704735

申请日：2015-05-05

Applicant: Huawei Technologies Co., Ltd.

Inventor： Jian He ,  Guangyu Shi ,  Xiaoke Ni ,  Norbert Egi ,  Xiancai Li ,  Yu Liu ,  Huawei Liu

IPC: G06F13/12 ,  G06F13/38 ,  G06F3/06 ,  G06F13/10 ,  G06F13/40 ,  G06F13/42

CPC classification number: G06F13/4282 ,  G06F3/061 ,  G06F3/0611 ,  G06F3/0655 ,  G06F3/0673 ,  G06F12/10 ,  G06F13/10 ,  G06F13/28 ,  G06F13/4022 ,  G06F13/4221 ,  G06F2212/1016

Abstract: Embodiments of the present invention provide a data processing system and a data processing method. An MMIO address of a data request is obtained, where data stored in a PCIe storage device can be directly obtained from the MMIO address, and according to the MMIO address, a network adapter can directly read the data from the PCIe storage device of the data processing system, and transmit the data to a second data processing system, or directly write the data received from the second data processing system into the PCIe storage device. Therefore, the processing system can implement transmitting data directly from the PCIe storage device to the network adapter without using memory. During data transmission between the two data processing systems, a usage ratio of a resource, such as memory and a CPU, is reduced, and efficiency of data transmission is improved.

公开(公告)号：US10585628B2

公开(公告)日：2020-03-10

申请号：US15961851

申请日：2018-04-24

Applicant: Huawei Technologies Co., Ltd.

Inventor： Jian He ,  Xiaoke Ni ,  Yimin Yao

IPC: G06F15/167 ,  G06F3/06 ,  G06F13/42 ,  G06F13/40 ,  G06F13/38

Abstract: A packet forwarding method, device, and system for use in non-volatile memory express (NVMe) over fabric. According to the method, a first packet sent by a control device is received, and an input/output command or a management command that complies with the NVMe protocol and that is included in the first packet is obtained; a second packet is generated according to the input/output command or the management command, where the input/output command or the management command complies with the NVMe protocol and is included in the first packet, and the second packet is a packet for executing the input/output command or the management command; a network port for forwarding the second packet is selected according to an identity of a target non-volatile memory; and the second packet is encapsulated into a packet that meets an NVMe over fabric forwarding requirement, and the encapsulated second packet is sent by using the selected network port.

公开(公告)号：US20200050551A1

公开(公告)日：2020-02-13

申请号：US16598755

申请日：2019-10-10

Applicant: Huawei Technologies Co., Ltd.

Inventor： Jian He ,  Xiaoke Ni

IPC: G06F12/0875 ,  G06F13/16

Abstract: In a data access method, after an interface card receives a first data write instruction or a first data read instruction, the interface card generates a second data write instruction or a second data read instruction, and writes the second data write instruction or the second data read instruction into a cache. No resource of a processor of a storage device is used. After the interface card writes the second data write instruction or the second data read instruction into the cache, a cache control unit sends the second data write instruction or the second data read instruction to a storage subsystem. No resource of the processor of the storage device is used. Alternatively, the cache control unit may instruct the storage subsystem to execute the second data write instruction or the second data read instruction.

公开(公告)号：US20170168849A1

公开(公告)日：2017-06-15

申请号：US15443452

申请日：2017-02-27

Applicant: Huawei Technologies Co., Ltd.

Inventor： Jian He ,  Xiancai Li ,  Xiaoke Ni ,  Yimin Yao

IPC: G06F9/44 ,  G06F11/27 ,  G11C29/12 ,  G06F11/22

Abstract: A computer device and a memory startup method of a computer are provided, where a basic input/output system initializes only a first part of memory in a memory initialization phase after a computer is powered on and started, so that an operating system can be started, and after the operating system is started, the basic input/output system continues to initialize memory that is not initialized in the computer, so that, in a startup phase, the computer can start the operating system without needing to wait until all memory has been initialized; therefore, a time from being started to entering the operating system is reduced for the computer, and a user can quickly enter the operating system to perform an operation, thereby improving user experience.

公开(公告)号：US09678918B2

公开(公告)日：2017-06-13

申请号：US15280700

申请日：2016-09-29

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Jian He ,  Guangyu Shi ,  Xiaoke Ni ,  Norbert Egi ,  Xiancai Li ,  Yu Liu ,  Huawei Liu

IPC: G06F13/12 ,  G06F13/38 ,  G06F13/42 ,  G06F3/06 ,  G06F13/10 ,  G06F12/10 ,  G06F13/28 ,  G06F13/40

CPC classification number: G06F13/4282 ,  G06F3/061 ,  G06F3/0611 ,  G06F3/0655 ,  G06F3/0673 ,  G06F12/10 ,  G06F13/10 ,  G06F13/28 ,  G06F13/4022 ,  G06F13/4221 ,  G06F2212/1016

Abstract: Embodiments of the present invention provide a data processing system and a data processing method. An MMIO address of a data request is obtained, where data stored in a PCIe storage device can be directly obtained from the MMIO address, and according to the MMIO address, a network adapter can directly read the data from the PCIe storage device of the data processing system, and transmit the data to a second data processing system, or directly write the data received from the second data processing system into the PCIe storage device. Therefore, the processing system can implement transmitting data directly from the PCIe storage device to the network adapter without using memory. During data transmission between the two data processing systems, a usage ratio of a resource, such as memory and a CPU, is reduced, and efficiency of data transmission is improved.

公开(公告)号：US20170017607A1

公开(公告)日：2017-01-19

申请号：US15280700

申请日：2016-09-29

Applicant: HUAWEI TECHNOLOGIES CO.,LTD.

Inventor： Jian He ,  Guangyu Shi ,  Xiaoke Ni ,  Norbert Egi ,  Xiancai Li ,  Yu Liu ,  Huawei Liu

IPC: G06F13/42 ,  G06F12/10 ,  G06F13/28 ,  G06F3/06

CPC classification number: G06F13/4282 ,  G06F3/061 ,  G06F3/0611 ,  G06F3/0655 ,  G06F3/0673 ,  G06F12/10 ,  G06F13/10 ,  G06F13/28 ,  G06F13/4022 ,  G06F13/4221 ,  G06F2212/1016

Abstract: Embodiments of the present invention provide a data processing system and a data processing method. An MMIO address of a data request is obtained, where data stored in a PCIe storage device can be directly obtained from the MMIO address, and according to the MMIO address, a network adapter can directly read the data from the PCIe storage device of the data processing system, and transmit the data to a second data processing system, or directly write the data received from the second data processing system into the PCIe storage device. Therefore, the processing system can implement transmitting data directly from the PCIe storage device to the network adapter without using memory. During data transmission between the two data processing systems, a usage ratio of a resource, such as memory and a CPU, is reduced, and efficiency of data transmission is improved.

Abstract translation: 本发明的实施例提供了一种数据处理系统和数据处理方法。 获得数据请求的MMIO地址，其中可以从MMIO地址直接获取存储在PCIe存储设备中的数据，并且根据MMIO地址，网络适配器可以直接从PCIe存储设备读取数据 处理系统，并将数据发送到第二数据处理系统，或直接将从第二数据处理系统接收的数据写入PCIe存储设备。 因此，处理系统可以在不使用存储器的情况下实现从PCIe存储设备直接发送数据到网络适配器。 在两个数据处理系统之间的数据传输期间，诸如存储器和CPU之类的资源的使用率被降低，并且提高了数据传输的效率。