公开(公告)号：US09026766B2

公开(公告)日：2015-05-05

申请号：US14282576

申请日：2014-05-20

Applicant: Huawei Technologies Co., Ltd.

Inventor： Wei Wang ,  Jiang Liu ,  Xishi Qiu ,  Jianguo Wu

IPC: G06F12/00 ,  G06F12/02

CPC classification number: G06F12/023 ,  G06F9/5016

Abstract: Embodiments of the present invention provide a memory management method, a memory management apparatus and a NUMA system. The memory management method includes: determining, according to a memory demand information which includes memory demand information sent by a processor, whether a memory controller meeting the memory demand information exists in a local processing node which the processor; and if exists, determining, in the memory controller meeting the memory demand information, a memory management area meeting the memory demand information, and allocating the memory management area meeting the memory demand information to the processor. Therefore, the memory controller and the memory management area do not need to be determined in a processing node that does not meet the requirements, which can rapidly find a storing area meeting the requirements, and improve the memory allocation efficiency.

Abstract translation: 本发明的实施例提供一种存储器管理方法，存储器管理装置和NUMA系统。 存储器管理方法包括：根据包括处理器发送的存储器需求信息的存储器需求信息，确定满足存储器需求信息的存储器控​​制器是否存在于处理器的本地处理节点中; 如果存在，则在满足存储器需求信息的存储器控​​制器中确定满足存储器需求信息的存储器管理区域，并且将满足存储器需求信息的存储器管理区域分配给处理器。 因此，不需要在不满足要求的处理节点中确定存储器控制器和存储器管理区域，可以快速找到满足要求的存储区域，并提高存储器分配效率。

公开(公告)号：US20140237192A1

公开(公告)日：2014-08-21

申请号：US14263212

申请日：2014-04-28

Applicant: Huawei Technologies Co., Ltd.

Inventor： Yiyang Liu ,  Wei Wang ,  Xishi Qiu

IPC: G06F12/08 ,  G06F12/12

CPC classification number: G06F12/0833 ,  G06F11/3419 ,  G06F11/3447 ,  G06F11/3466 ,  G06F11/3471 ,  G06F12/1009 ,  G06F12/121 ,  G06F2201/86 ,  G06F2201/865 ,  G06F2201/88 ,  G06F2212/62

Abstract: Embodiments of the present invention provide a method and an apparatus for constructing a memory access model, and relate to the field of computers. The method includes: obtaining a page table corresponding to a process referencing a memory block, and clearing a Present bit included in each page table entry stored in the page table; and constructing a memory access model of the memory block according to the number of access times of each page in the memory block and time obtained through timing, where the memory access model at least includes the number of access times and an access frequency of each page in the memory block. The apparatus includes: a first obtaining module, a first monitoring module, a first increasing module, and a second obtaining module. The present invention can reduce the memory consumption and an impact on the system performance, and avoid a system breakdown.

Abstract translation: 本发明的实施例提供了一种用于构建存储器访问模型并涉及计算机领域的方法和装置。 该方法包括：获取与引用存储块的过程相对应的页表，并清除存储在页表中的每个页表项中包括的当前位; 以及根据存储块中的每一页的访问次数和通过定时获得的时间来构建存储器块的存储器访问模型，其中存储器访问模型至少包括访问次数和每页的访问频率 在内存块中。 该装置包括：第一获取模块，第一监视模块，第一增加模块和第二获取模块。 本发明可以减少存储器消耗和对系统性能的影响，并避免系统崩溃。

公开(公告)号：US20140007129A1

公开(公告)日：2014-01-02

申请号：US14017124

申请日：2013-09-03

Applicant: Huawei Technologies Co., Ltd.

Inventor： Yanzhong Ye ,  Wei Wang ,  Hanjun Guo ,  Xishi Qiu

IPC: G06F9/48

CPC classification number: G06F9/4856

Abstract: Embodiments of the present invention disclose a method and an apparatus for resource migration, related to the field of processors, to avoid cross-node memory access and cross-node cache access, save inter-node high-speed bandwidth, and improve system performance. The method of the present invention includes: when a node is removed, obtaining a process in the node; determining a destination node to which the process and the memory corresponding to the process are migrated, according to mapping between the process and the memory corresponding to the process, and affinity of the process; and migrate the process in the node and the memory corresponding to the process to the destination node. The embodiments of the present invention are mainly applied to a resource migration procedure.

Abstract translation: 本发明的实施例公开了与处理器领域相关的用于资源迁移的方法和装置，以避免跨节点存储器访问和跨节点高速缓存访​​问，节省节点间高速带宽并提高系统性能。 本发明的方法包括：当节点被去除时，获得节点中的过程; 根据与进程对应的进程和存储器之间的映射以及进程的亲和性，确定与进程对应的进程和存储器迁移到的目的地节点; 并将与进程对应的节点和存储器中的进程迁移到目标节点。 本发明的实施例主要应用于资源迁移过程。

公开(公告)号：US20130124914A1

公开(公告)日：2013-05-16

申请号：US13732017

申请日：2012-12-31

Applicant: HUAWEI TECHNOLOGIES CO., LTD.

Inventor： Yiyang Liu ,  Wei Wang ,  Haizhen Zheng

IPC: G06F11/14

CPC classification number: G11B5/48 ,  G06F11/1004 ,  G06F11/1666

Abstract: Embodiments of the present invention disclose a method and a device for detecting data reliability, which belong to the field of computer technologies. The method includes: dividing source data into multiple source data blocks; establishing a corresponding duplicate data block for each source data block, where the duplicate data block and the source data block are stored in different physical memory blocks respectively; establishing a corresponding reliability maintenance data structure for the source data block; and registering the reliability maintenance data structure with a first reliability maintenance thread of a central processing unit, the source data block corresponding to the reliability maintenance data structure registered with the first reliability maintenance thread, comparing a reliability check value of the source data block with a reliability check value recorded in the reliability maintenance data structure, and replacing the source data block having the error with the duplicate data block.

Abstract translation: 本发明的实施例公开了一种属于计算机技术领域的用于检测数据可靠性的方法和装置。 该方法包括：将源数据划分为多个源数据块; 为每个源数据块建立相应的重复数据块，其中重复数据块和源数据块分别存储在不同的物理存储器块中; 为源数据块建立相应的可靠性维护数据结构; 以及将所述可靠性维护数据结构与中央处理单元的第一可靠性维护线程对应，所述源数据块与所述第一可靠性维护线程中登记的可靠性维护数据结构相对应，将所述源数据块的可靠性检查值与 记录在可靠性维护数据结构中的可靠性检查值，以及用复制数据块替换具有错误的源数据块。