US 12,242,860 B2
Reset of a multi-node system
Daniel John Pelham Wilkinson, Bristol (GB); and Bjorn Dag Johnsen, Oslo (NO)
Assigned to GRAPHCORE LIMITED, Bristol (GB)
Filed by Graphcore Limited, Bristol (GB)
Filed on Feb. 22, 2023, as Appl. No. 18/172,661.
Claims priority of application No. 2202815 (GB), filed on Mar. 1, 2022.
Prior Publication US 2023/0281018 A1, Sep. 7, 2023
Int. Cl. G06F 9/4401 (2018.01); G06F 9/52 (2006.01)
CPC G06F 9/4418 (2013.01) [G06F 9/522 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A processing system configured to support an application, the processing system comprising:
a plurality of processing nodes, each of which comprises:
a processing unit configured to execute a set of instructions for a part of the application; and
at least one storage configured to store a copy of a number indicating a number of times a global reset of the processing system has been performed; and
reset circuitry configured to cause:
a respective processing node to be reset as part of one of the global resets for the processing system; and
application state for the part of the application for the respective processing node to be restored from a checkpoint,
wherein each of the processing nodes comprises update circuitry configured to, as part of the reset of the respective processing node, update the copy of the number held in the at least one storage of the respective processing node from a first value to a second value,
wherein the processing unit of a second of the processing nodes is configured to, prior to the reset of the second of the processing nodes, execute its sets of instructions to cause the sending of one or more data frames to a first of the processing nodes, each of the one or more data frames comprising the first value of the number,
wherein circuitry of the first of the processing nodes is configured to, following the reset of the first of the processing nodes, and in response to determining a mismatch between the first value of the number in the one or more data frames and the second value of the number held in the at least one storage of the first of the processing nodes, cause data in the one or more data frames to be discarded, and
wherein the processing unit of the first of the processing nodes is configured to, following the reset of the first of the processing nodes, stall at a first barrier synchronization until a further one of the processing units, which is also arranged to participate in the first barrier synchronization, reaches the first barrier synchronization in its application processing.