US 12,430,179 B2
Distribution of a workload among nodes of a system with a NUMA architecture
Aditya Chatterjee, Bangalore (IN)
Assigned to Advanced Micro Devices, Inc., Santa Clara, CA (US)
Filed by Advanced Micro Devices, Inc., Santa Clara, CA (US)
Filed on Jul. 12, 2022, as Appl. No. 17/863,247.
Prior Publication US 2024/0020173 A1, Jan. 18, 2024
Int. Cl. G06F 9/50 (2006.01); G06F 9/445 (2018.01)
CPC G06F 9/505 (2013.01) [G06F 9/44505 (2013.01); G06F 9/5072 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for distribution of a workload among nodes of a non-uniform memory access (NUMA) architecture, comprising:
receiving the workload and data batches, the data batches to be processed by the workload;
obtaining a determined distribution based on a graph, wherein each node of the graph has an associated batch size, disjoint set, and execution time, by performing operations including: selecting an initial node of the graph as a current node for a current path; traversing to a new compatible node of the graph for which the total time for the path is less than a current time for the path; and repeating the traversing until a sum of batch sizes of each node in the current path is equal to a number of the data batches;
assigning workload processes to the nodes according to the determined distribution, wherein the determined distribution is selected out of a set of distributions so that execution time of the workload, when executed according to the distribution, is minimal; and
executing the workload according to the determined distribution.