公开(公告)号：US20180367443A1

公开(公告)日：2018-12-20

申请号：US16109694

申请日：2018-08-22

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Ziliang SHI ,  Huaxi GU

IPC: H04L12/721 ,  G06F17/50

CPC classification number: H04L45/12 ,  G06F9/50 ,  G06F15/17312 ,  G06F17/5077 ,  H04L45/26

Abstract: A method for routing a packet in an accelerator network is provided, the method includes: receiving, by a first accelerator which is any accelerator in an accelerator network, an acceleration request packet which includes an acceleration sequence and to-be-accelerated data from a CPU, the acceleration sequence is used to indicate a plurality of acceleration types used for accelerating the data and indicate an order of the plurality of acceleration types; determining, by the first accelerator, a target loop path of the acceleration request packet based on an acceleration information table which includes location information and acceleration type information of a plurality of accelerators in the accelerator network and accelerators on the target loop path are capable of performing acceleration processing on the acceleration request packet according to an indication of the acceleration sequence; and routing, by the first accelerator, the acceleration request packet according to the target loop path.

公开(公告)号：US20170300437A1

公开(公告)日：2017-10-19

申请号：US15639274

申请日：2017-06-30

Applicant: Huawei Technologies Co., Ltd.

Inventor： Ziliang SHI ,  Qi WANG ,  Tao LIU

IPC: G06F13/36 ,  G06F13/42

CPC classification number: G06F13/36 ,  G06F13/42 ,  H04L41/00 ,  Y02D50/10

Abstract: A service acceleration method is disclosed. The method includes: querying acceleration type information and information about an idle acceleration ability that are of the multiple acceleration engines, and forming an accelerator resource pool according to the acceleration type information and the information about the idle acceleration ability; after receiving an acceleration application of a first service processing unit in the multiple service processing units, allocating, from the accelerator resource pool according to a preset allocation granularity, a first idle acceleration ability and a connection number that are corresponding to the first acceleration ability and a first acceleration type; and feeding back a result packet that is obtained after acceleration processing to the first service processing unit. The method and the apparatus that are provided in the present invention resolve a prior-art problem of resource waste caused by inappropriate utilization of accelerator resources.