Fault recovery method and related system
Through the controller to obtain the failure information reported on the exchange node, determine the execution node and update the memory address communication path, solve the problem of low network failure recovery efficiency of the bus exchange, and achieve efficient failure recovery of non -stop driving.
CN119377000APending Publication Date: 2025-01-28HUAWEI TECH CO LTD
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202310939855.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-27
- Publication Date
- 2025-01-28
AI Technical Summary
Technical Problem
With the growth of chip interconnection scale, the possibility and scope of impact of bus switching network failures are increasing, resulting in the efficient repair of bus switching network failures becoming the focus of the industry.
Method used
Provide a fault recovery method to obtain the fault information reported by the exchange node through the controller, determine the execution node used for fault recovery, and update the memory address communication path affected by the fault to achieve non -stop failure recovery.
Benefits of technology
With the case of not blocking the calculation node, the failure recovery of the bus exchange network exchanging networks is performed efficiently, avoiding the waste of computing power, and improving the failure recovery efficiency.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN119377000A_ABST
Abstract
The invention provides a fault recovery method, which is applied to a controller, the controller is used for performing fault recovery when a bus switching network in a computing cluster fails, the bus switching network comprises at least one switching node, and the bus switching network is used for realizing memory access among computing nodes in the computing cluster. The method comprises the steps that a controller obtains fault information reported by at least one switching node, the fault information indicates a fault mode and a fault position, then an execution node used for fault recovery is determined from the at least one switching node according to the fault information, and a memory address communication path influenced by the fault is determined according to the fault information, the execution node is then instructed to update the memory address communication path affected by the failure. According to the method, the memory address communication path influenced by the fault is updated, so that fault recovery is realized without shutdown, shielding of a single or a plurality of computing nodes is not needed, and waste of computing power is avoided.
Need to check novelty before this filing date? Find Prior Art