US 11,888,931 B1
Massively parallel in-network compute
William Brad Matthews, San Jose, CA (US); Puneet Agarwal, Cupertino, CA (US); and Bruce Hui Kwan, Sunnyvale, CA (US)
Assigned to Innovium, Inc., San Jose, CA (US)
Filed by Innovium, Inc., San Jose, CA (US)
Filed on May 11, 2022, as Appl. No. 17/742,354.
Application 17/742,354 is a continuation of application No. 17/200,463, filed on Mar. 12, 2021, granted, now 11,425,195.
This patent is subject to a terminal disclaimer.
Int. Cl. H04L 67/1074 (2022.01); H04L 49/15 (2022.01); G06N 20/00 (2019.01); G06N 3/045 (2023.01); H04L 45/02 (2022.01)
CPC H04L 67/1076 (2013.01) [G06N 3/045 (2023.01); G06N 20/00 (2019.01); H04L 49/15 (2013.01); H04L 45/08 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A method comprising:
receiving, by a network switch implemented at least in part with one or more hardware processors, vector chunks over a plurality of switch ports;
wherein the vector chunks have values generated by a plurality of compute nodes for a common set of vector elements;
wherein the common set of vector elements includes a plurality of subsets of vector elements;
wherein each vector chunk of the vector chunks is received over a respective switch port, of the plurality of switch ports, from a respective compute node in the plurality of compute nodes;
wherein said each vector chunk has a subset of the values for a respective subset in the plurality subsets of vector elements;
reducing, by the network switch, the vector chunks generated by the plurality of compute nodes into a single result chunk;
sending, by the network switch, the single result chunk to each compute node of the plurality of compute nodes.