Abstract: Examples described herein relate to a switch comprising circuitry configured to for packet communications associated with a collective operation to train machine learning (ML) models: utilize a reliable transport protocol for communications from at least one worker node of the collective operation to a switch, wherein the utilize a reliable transport protocol for communications from at least one worker node of the collective operation to the switch comprises store packet receipt state for per-packet communications from the at least one worker node of the collective operation to the switch and utilize a non-reliable transport protocol by the switch to a device that is to perform aggregation of results, wherein the reliable transport protocol comprises a different protocol than that of the non-reliable transport protocol.

Abstract: Examples described herein relate to a network interface device. The network interface device can include circuitry that is to: receive a first packet comprising a first packet header and a first packet payload; receive multiple subsequent packets comprising multiple packet headers for respective multiple subsequent packets; update at least one of the multiple packet headers; and construct egress packets. In some examples, the egress packets include respective one of the multiple packet headers and the first packet payload.

Abstract: A device can include interfaces configured to receive data packets from compute nodes. The device can include circuitry provide data to the compute nodes to synchronize reception of data packets received from the compute nodes. The reception can be synchronized to provide data of the data packets to each memory slot of a memory in an order.

Abstract: Methods and apparatus for employing selective compression for addressing congestion control for Artificial Intelligence (AI) workloads. Multiple interconnected compute nodes are used for performing an AI workload in a distributed environment, such as training an AI model. Periodically, such as following an epoch for processing batches of training data in parallel, the compute nodes exchange Tensor data (e.g., local model gradients) with one another, which may lead to network/fabric congestion. Compute nodes and/or switches in the distributed environment are configured to detect current or projected network/fabric congestion and to selectively apply variable rate compression to packets containing the Tensor data to alleviate/avoid the congestion. Tensor data may be selectively applied at source compute nodes by computing a network pause time and comparing that time to a compression compute time.

Abstract: Examples described herein relate to a switch circuitry that includes circuitry to cause transmission, to multiple destinations, of copies of a packet received from a sender network interface device and circuitry to indicate acknowledgement of packet receipt, from the multiple destinations, to the sender network interface device based on receipt of acknowledgements of packet receipt from the multiple destinations. In some examples, the circuitry is to indicate acknowledgement of packet receipt, from the multiple destinations, to the sender network interface device with a packet index value.

Abstract: Examples described herein relate to a switch comprising: circuitry, when operational, to receive a packet comprising a header and a payload and in conjunction with performance of computation on the packet payload, forward the packet header, but not the payload, to a destination endpoint. In some examples, the destination endpoint of the packet is to perform management of reliable transport. In some examples, the circuitry includes programmable data plane circuitry comprising ingress pipeline or egress pipeline and one or more match action units (MAUs) to perform processing of the payload, wherein the programmable data plane circuitry is to perform computation on the packet payload.

Abstract: In one embodiment, a device in a network receives one or more data units. The device calculates a hash value based on the one or more data units and using a hash function. Ranges of hash values generated by the hash function are assigned to different devices along the path such that any given hash value generated by the hash function is assigned to a predefined number of the devices along the path. The device determines whether the calculated hash value is within the range of hash values assigned to the device. The device stores data derived from the one or more data units, when the device determines that the calculated hash value is within the range of hash values assigned to the device.

Abstract: A method for profiling network traffic. The method includes capturing, from the network traffic using a packet capturing device, a plurality of packets, identifying a first portion of the plurality of packets as a first flow based at least on a common Internet Protocol (IP) address assigned to each packet of the first flow by a network address translation (NAT) device, extracting, by a hardware processor separate from the NAT device and based on an NAT profile of the NAT device, a first data item from the first flow, wherein the first data item is inserted into the first flow by the NAT device for identifying a first host device coupled to the NAT device, and determining, by the hardware processor based on the first data item, that the first flow is generated by the first host device.

Abstract: A method for profiling network traffic. The method includes capturing, from the network traffic using a packet capturing device, a plurality of packets, identifying a first portion of the plurality of packets as a first flow based at least on a common Internet Protocol (IP) address assigned to each packet of the first flow by a network address translation (NAT) device, extracting, by a hardware processor separate from the NAT device and based on an NAT profile of the NAT device, a first data item from the first flow, wherein the first data item is inserted into the first flow by the NAT device for identifying a first host device coupled to the NAT device, and determining, by the hardware processor based on the first data item, that the first flow is generated by the first host device.

Abstract: In one embodiment, a device in a network receives one or more data units. The device calculates a hash value based on the one or more data units and using a hash function. Ranges of hash values generated by the hash function are assigned to different devices along the path such that any given hash value generated by the hash function is assigned to a predefined number of the devices along the path. The device determines whether the calculated hash value is within the range of hash values assigned to the device. The device stores data derived from the one or more data units, when the device determines that the calculated hash value is within the range of hash values assigned to to the device.

Abstract: A method for profiling network traffic. The method includes capturing, from the network traffic using a packet capturing device, a plurality of packets, identifying a first portion of the plurality of packets as a first flow based at least on a common Internet Protocol (IP) address assigned to each packet of the first flow by a network address translation (NAT) device, extracting, by a hardware processor separate from the NAT device and based on an NAT profile of the NAT device, a first data item from the first flow, wherein the first data item is inserted into the first flow by the NAT device for identifying a first host device coupled to the NAT device, and determining, by the hardware processor based on the first data item, that the first flow is generated by the first host device.

Abstract: A method for profiling network traffic. The method includes capturing, from the network traffic using a packet capturing device, a plurality of packets, identifying a first portion of the plurality of packets as a first flow based at least on a common Internet Protocol (IP) address assigned to each packet of the first flow by a network address translation (NAT) device, extracting, by a hardware processor separate from the NAT device and based on an NAT profile of the NAT device, a first data item from the first flow, wherein the first data item is inserted into the first flow by the NAT device for identifying a first host device coupled to the NAT device, and determining, by the hardware processor based on the first data item, that the first flow is generated by the first host device.

Abstract: A method for profiling network traffic. The method includes capturing, from the network traffic using a packet capturing device, a plurality of packets, identifying a first portion of the plurality of packets as a first flow based at least on a common Internet Protocol (IP) address assigned to each packet of the first flow by a network address translation (NAT) device, extracting, by a hardware processor separate from the NAT device and based on an NAT profile of the NAT device, a first data item from the first flow, wherein the first data item is inserted into the first flow by the NAT device for identifying a first host device coupled to the NAT device, and determining, by the hardware processor based on the first data item, that the first flow is generated by the first host device.

Abstract: A method for profiling network traffic. The method includes capturing, from the network traffic using a packet capturing device, a plurality of packets, identifying a first portion of the plurality of packets as a first flow based at least on a common Internet Protocol (IP) address assigned to each packet of the first flow by a network address translation (NAT) device, extracting, by a hardware processor separate from the NAT device and based on an NAT profile of the NAT device, a first data item from the first flow, wherein the first data item is inserted into the first flow by the NAT device for identifying a first host device coupled to the NAT device, and determining, by the hardware processor based on the first data item, that the first flow is generated by the first host device.

Abstract: A method for profiling network traffic. The method includes capturing, from the network traffic using a packet capturing device, a plurality of packets, identifying a first portion of the plurality of packets as a first flow based at least on a common Internet Protocol (IP) address assigned to each packet of the first flow by a network address translation (NAT) device, extracting, by a hardware processor separate from the NAT device and based on an NAT profile of the NAT device, a first data item from the first flow, wherein the first data item is inserted into the first flow by the NAT device for identifying a first host device coupled to the NAT device, and determining, by the hardware processor based on the first data item, that the first flow is generated by the first host device.