In-network compute operation spreading
In one embodiment, a network switch device includes a network interface to receive vectors from endpoint devices, and an aggregation and reduction accelerator to perform elementwise and vector splitting operations with the vectors as input yielding at least two intermediate vector results, wherein the network interface is to send the at least two intermediate vector results to different corresponding network switches in different switch aggregation trees, receive at least two final vector results of an aggregation and reduction process from the different switch aggregation trees, and combine the at least two final vector results to yield a combined final vector result, wherein the network interface is to send the combined final vector result to the endpoint devices.
1 . A network switch device, comprising:
a network interface to receive vectors from endpoint devices as part of an aggregation and reduction process; and
an aggregation and reduction accelerator to:
perform elementwise and vector splitting operations with the received vectors as input yielding at least two intermediate vector results, including a first intermediate vector result and a second intermediate vector result, wherein the network interface is to send the at least two intermediate vector results to different corresponding network switches in different switch aggregation trees including sending the first intermediate vector result to a first network switch in a first switch aggregation tree and the second intermediate vector result to a second network switch in a second switch aggregation tree;
receive at least two final vector results of the aggregation and reduction process from the different switch aggregation trees including receiving a first final vector result from the first network switch and a second final vector result from the second network switch; and
combine the at least two final vector results to yield a combined final vector result, wherein the network interface is to send the combined final vector result to the endpoint devices.
2 . The device according to claim 1 , wherein the aggregation and reduction accelerator is to:
split the received vectors yielding at least two groups of split vectors; and
perform elementwise operations on the split vectors according to the at least two groups yielding the at least two intermediate vector results for the at least two groups, respectively.
3 . The device according to claim 1 , wherein the aggregation and reduction accelerator is to:
perform elementwise operations on the received vectors yielding a given vector result; and
split the given vector result yielding the at least two intermediate vector results.
4 . The device according to claim 1 , wherein the aggregation and reduction accelerator is to perform elementwise operations with the received vectors as input, the elementwise operations being selected from any one or more of the following: summation; point wise minimum; point wise maximum; bitwise OR, bitwise NOR, bitwise XOR, maximum value with index, minimum value with index, and a logical operation.
5 . The device according to claim 1 , wherein the aggregation and reduction accelerator is to perform elementwise summation with the received vectors as input, the received vectors having a floating-point value data type.
6 . The device according to claim 1 , wherein:
the network interface is to receive the vectors from the endpoint devices in aggregation and reduction protocol packets; and
the aggregation and reduction accelerator is to perform vector splitting according to packet boundaries.
7 . The device according to claim 1 , wherein the aggregation and reduction accelerator is to selectively perform elementwise operations with vector splitting for first received vectors and selectively perform elementwise operations without performing vector splitting for second received vectors.
8 . The device according to claim 7 , wherein the aggregation and reduction accelerator is to determine to selectively perform elementwise operations with vector splitting for first received vectors based on:
the first received vectors having a floating-point value data type; and
the elementwise operations including summation.
9 . The device according to claim 8 , wherein the aggregation and reduction accelerator is to determine to selectively perform elementwise operations without vector splitting for second received vectors based on either:
the second received vectors not having a floating-point value data type; or
the elementwise operations not including summation.
10 . The device according to claim 1 , wherein the aggregation and reduction accelerator, responsively to an administrator decision, is to selectively perform elementwise operations with vector splitting for first received vectors and selectively perform elementwise operations without performing vector splitting for second received vectors.
11 . The device according to claim 1 , wherein the aggregation and reduction accelerator, responsively to an aggregation manager decision based on a topology of the network, is to selectively perform elementwise operations with vector splitting for first received vectors and selectively perform elementwise operations without performing vector splitting for second received vectors.
12 . A networking method, comprising:
receiving vectors from endpoint devices as part of an aggregation and reduction process;
performing elementwise and vector splitting operations with the received vectors as input yielding at least two intermediate vector results, including a first intermediate vector result and a second intermediate vector result;
sending the at least two intermediate vector results to different corresponding network switches in different switch aggregation trees including sending the first intermediate vector result to a first network switch in a first switch aggregation tree and the second intermediate vector result to a second network switch in a second switch aggregation tree;
receiving at least two final vector results of the aggregation and reduction process from the different switch aggregation trees including receiving a first final vector result from the first network switch and a second final vector result from the second network switch;
combining the at least two final vector results to yield a combined final vector result; and
sending the combined final vector result to the endpoint devices.
13 . The method according to claim 12 , further comprising:
splitting the received vectors yielding at least two groups of split vectors; and
performing elementwise operations on the split vectors according to the at least two groups yielding the at least two intermediate vector results for the at least two groups, respectively.
14 . The method according to claim 12 , further comprising:
performing elementwise operations on the received vectors yielding a given vector result; and
splitting the given vector result yielding the at least two intermediate vector results.
15 . The method according to claim 12 , further comprising performing elementwise operations with the received vectors as input, the elementwise operations being selected from any one or more of the following: summation; point wise minimum; point wise maximum; bitwise OR, bitwise NOR, bitwise XOR, maximum value with index, minimum value with index, and a logical operation.
16 . The method according to claim 12 , further comprising performing elementwise summation with the received vectors as input, the received vectors having a floating-point value data type.
17 . The method according to claim 12 , wherein:
the receiving vectors from endpoint devices includes receiving the vectors from the endpoint devices in aggregation and reduction protocol packets; and
the performing includes performing vector splitting according to packet boundaries.
18 . The method according to claim 12 , further comprising:
selectively performing elementwise operations with vector splitting for first received vectors; and
selectively performing elementwise operations without performing vector splitting for second received vectors.
19 . The method according to claim 18 , further comprising determining to selectively perform elementwise operations with vector splitting for first received vectors based on:
the first received vectors having a floating-point value data type; and
the elementwise operations including summation.
20 . The method according to claim 19 , further comprising determining to selectively perform elementwise operations without vector splitting for second received vectors based on either:
the second received vectors not having a floating-point value data type; or
the elementwise operations not including summation.
21 . The method according to claim 12 , further comprising, responsively to an administrator decision:
selectively performing elementwise operations with vector splitting for first received vectors; and
selectively performing elementwise operations without performing vector splitting for second received vectors.
22 . The method according to claim 12 , further comprising, responsively to an aggregation manager decision based on a topology of the network:
selectively performing elementwise operations with vector splitting for first received vectors; and
selectively performing elementwise operations without performing vector splitting for second received vectors.
23 . A system, comprising a plurality of network switches including different switch aggregation trees of the network switches having a leaf layer of the network switches, wherein:
each of the network switches in the leaf layer is to:
receive vectors from respective endpoint devices as part of an aggregation and reduction process;
perform elementwise and vector splitting operations with the received vectors as input yielding intermediate vector results; and
send different ones of the intermediate vector results to the network switches in the different switch aggregation trees;
the network switches in the different switch aggregation trees are to perform at least part of the aggregation and reduction process yielding different final vector results and provide the different final vector results to the network switches in the leaf layer; and
each of the network switches in the leaf layer is to:
receive the different final vector results from the different switch aggregation trees;
combine the different final vector results to yield a combined final vector result; and
send the combined final vector result to the respective endpoint devices.