MEMORY MANAGEMENT FOR MACHINE LEARNING TRAINING ON GPU

    公开(公告)号:US20190244096A1

    公开(公告)日:2019-08-08

    申请号:US15948706

    申请日:2018-04-09

    IPC分类号: G06N3/08 G06F3/06 G06T1/20

    摘要: A system and method for memory management. In one embodiment, the method includes generating a dependency structure comprising one or more task identifiers and one or more data object identifiers. The dependency structure includes a list of one or more dependencies for a first data object identifier of the one or more data object identifiers, a first dependency of the list identifying a first task for which a data object identified by the first data object identifier is an input. The method further includes counting the number of dependencies for the first data object identifier, decrementing the count by one when the first task completes execution, and, when the count reaches zero, deallocating the first data object.

    STORAGE SYSTEM AND CONTROL METHOD THEREOF
    3.
    发明申请

    公开(公告)号:US20190235792A1

    公开(公告)日:2019-08-01

    申请号:US16377372

    申请日:2019-04-08

    申请人: Hitachi, Ltd.

    IPC分类号: G06F3/06

    摘要: A capacity control module which manages a physical storage area provided by each of storage devices in a cluster by dividing the physical storage area into a physical storage area having a predetermined size, in a plurality of storage nodes configuring a storage system, and, a storage control module which receives an I/O request from a higher-level device are provided, two storage control modules, which are arranged in different storage nodes and configure a redundancy configuration, are managed as a storage control module pair, the capacity control module preferentially allocates each of the physical storage areas in the vicinity of an arrangement destination of each of the storage control modules configuring the storage control module pair, to the storage control module pair, and storage data is read from or is written on the physical storage area, according to a command applied from the storage control module.

    CONTROLLER, OPERATING METHOD THEREOF AND MEMORY SYSTEM

    公开(公告)号:US20190220225A1

    公开(公告)日:2019-07-18

    申请号:US16059220

    申请日:2018-08-09

    申请人: SK hynix Inc.

    IPC分类号: G06F3/06 G06F9/50

    摘要: A controller includes a core suitable for transferring a plurality of requests; a buffer including a plurality of clusters; a buffer manager suitable for assigning the plurality of requests respectively into the plurality of clusters, and for storing storage information of the buffer and cluster information regarding each of the plurality of clusters, into which the assigned requests are respectively assigned; and a descriptor suitable for updating a descriptor report such that the cluster information regarding each of the plurality of clusters and the assigned requests correspond to each other, respectively.

    COMPRESSION TECHNIQUES FOR DISTRIBUTED DATA
    6.
    发明申请

    公开(公告)号:US20190196907A1

    公开(公告)日:2019-06-27

    申请号:US16293540

    申请日:2019-03-05

    申请人: Intel Corporation

    IPC分类号: G06F11/10 H03M13/15 G06F3/06

    摘要: In one example, uncompressed data is compressed and divided into chunks. Each chunk of the compressed data stream is combined with state information to enable each chunk to be independently decompressed. Each of the compressed chunks is then stored on a different storage device along with its associated state information. A compute operation can then be offloaded to the device or node where each chunk is stored. Each chunk can be independently decompressed for execution of the offloaded operation without transferring all chunks to a central location for decompression and performance of the operation.

    DATA BACKUP SYSTEM AND METHOD THEREOF
    8.
    发明申请

    公开(公告)号:US20190187910A1

    公开(公告)日:2019-06-20

    申请号:US16282449

    申请日:2019-02-22

    IPC分类号: G06F3/06 G06F12/02

    摘要: A data backup method for backing up target data, through a driver module, from a first storage device to a second storage device is disclosed. The first storage device includes a first storage unit that stores the target data, and a first control unit that accesses the first storage unit based on a first logical-to-physical mapping table. The second storage device includes a second storage unit and a second control unit that accesses the second storage unit based on a second logical-to-physical mapping table. The method includes steps of: reading the target data from the first storage unit without accessing the first logical-to-physical mapping table and transmitting the target data to the driver module; transmitting the target data to the second control unit; and writing the target data to the second storage unit without accessing the second logical-to-physical mapping table.