Selective congestion notification by a network interface device
Examples described herein relate to a switch, when operational, that is configured to receive in a packet an indicator of number of remaining bytes in a flow and to selectively send a congestion message based on a fullness level of a buffer and indication of remainder of the flow. In some examples, the indicator is received in an Internet Protocol version 4 consistent Options header field or Internet Protocol version 6 consistent Flow label field. In some examples, the congestion message comprises one or more of: an Explicit Congestion Control Notification (ECN), priority-based flow control (PFC), and/or in-band telemetry (INT). In some examples, to selectively send a congestion message to a transmitter based on a fullness level of a buffer that stored the packet and the number of remaining bytes in flow, the switch is to determine whether the buffer is large enough to store the remaining bytes in the flow.
1. An apparatus comprising:
a switch comprising:
circuitry, when operational, configured to receive, in a packet, an indicator of traffic volume to be sent, by a sender, in a flow to the switch and
to selectively send a congestion message based on a fullness level of a buffer in the switch and the indicator of traffic volume to be sent, by the sender, in the flow.
2. The apparatus of claim 1 , wherein the indicator is received in an Internet Protocol version 4 consistent Options header field or Internet Protocol version 6 consistent Flow label field.
3. The apparatus of claim 1 , wherein the congestion message comprises one or more of: an Explicit Congestion Control Notification (ECN), priority-based flow control (PFC), and/or in-band telemetry (INT).
4. The apparatus of claim 1 , wherein to selectively send the congestion message based on the fullness level of the buffer that stored the received packet and the indicator of traffic volume to be sent, the switch is to determine whether the buffer is large enough to store remaining traffic volume in the flow.
5. The apparatus of claim 1 , wherein based on an indication to separate latency-sensitive flows into one or more queues, the switch is to allocate packets of the latency-sensitive flows into the one or more queues associated with latency-sensitive flows.
6. The apparatus of claim 5 , wherein the packets of the latency-sensitive flows are associated with a last phase of an AllReduce operation.
7. The apparatus of claim 5 , wherein the packets of the latency-sensitive flows comprise a mouse flow.
8. The apparatus of claim 1 , wherein the switch comprises one or more of: an infrastructure processing unit (IPU), data processing unit (DPU), smartNIC, forwarding element, switch, router, network interface controller, or network-attached appliance.
9. The apparatus of claim 1 , comprising a server, wherein the server is to configure the switch to selectively send a congestion message to a transmitter based on the fullness level of the buffer that stored the packet and the indicator of traffic volume to be sent.
10. The apparatus of claim 1 , comprising a datacenter, wherein
the datacenter comprises a sender network interface device, the switch, and a destination network interface device and
the sender network interface device is to send the packet with the indicator via the switch to the destination network interface device.
11. The apparatus of claim 10 , wherein the sender network interface device is to reset a flow's connection state information based on volume of transmission relative to a level.
12. The apparatus of claim 1 , wherein the indicator of traffic volume to be sent comprises an amount of packets to be sent or amount of message size to be sent.
13. At least one non-transitory computer-readable medium comprising instructions stored thereon, that if executed by one or more processors, cause the one or more processors to:
configure a network interface device to selectively send a congestion message to a transmitter network interface device based on a fullness level of a buffer and traffic volume to be sent in a flow, wherein the traffic volume to be sent in the flow is based on an indication received in a packet from the transmitter network interface device.
14. The computer-readable medium of claim 13 , wherein the traffic volume to be sent in the flow are received in an Internet Protocol version 4 consistent Options header field or Internet Protocol version 6 consistent Flow label field of the packet.
15. The computer-readable medium of claim 13 , wherein the congestion message comprises one or more of: an Explicit Congestion Control Notification (ECN), priority-based flow control (PFC), and/or in-band telemetry (INT).
16. The computer-readable medium of claim 13 , wherein to selectively send the congestion message to the transmitter network interface device based on the fullness level of the buffer and the traffic volume to be sent in the flow, the network interface device is to determine whether the buffer is large enough to store the traffic volume to be sent in the flow.
17. A method comprising:
determining whether to send a congestion message to a transmitter network interface device based on a fullness level of a buffer meeting or exceeding a congestion level and
refraining from sending a congestion message based on the fullness level of the buffer meeting or exceeding the congestion level based on a traffic volume to be sent in a flow that are to be stored in a buffer and/or latency sensitivity of a current operative phase associated with transmission of the flow, wherein the traffic volume to be sent in the flow is based on an indication received in a packet from the transmitter network interface device.
18. The method of claim 17 , wherein the traffic volume to be sent in the flow that are to be stored in the buffer and/or latency sensitivity of the current operative phase are received in an Internet Protocol version 4 consistent Options header field or Internet Protocol version 6 consistent Flow label field of the packet.
19. The method of claim 17 , wherein the congestion message comprises one or more of: an Explicit Congestion Control Notification (ECN), priority-based flow control (PFC), and/or in-band telemetry (INT).
20. The method of claim 17 , wherein the refraining from sending the congestion message based on the fullness level of the buffer meeting or exceeding the congestion level based on traffic volume to be sent in the flow that is to be stored in the buffer and/or latency sensitivity of the current operative phase is based on whether the buffer is large enough to store the traffic volume to be sent in the flow.
21. The method of claim 17 , wherein the refraining from sending the congestion message based on the fullness level of the buffer meeting or exceeding the congestion level based on traffic volume to be sent in the flow that is to be stored in the buffer and/or latency sensitivity of a current operative phase.
22. The method of claim 17 , comprising:
sending a congestion message to the transmitter network interface device based on one or more of: the buffer not being large enough to store the traffic volume to be sent in the flow.