Abstract: A packet-processing method includes looking up a first match-action table on a network interface card (NIC) for a received packet; in response to finding a matching entry indicating an action, queuing the received packet in a first queue and storing the action data in an instruction memory; and responsive to not finding a matching entry, queuing the received packet in the first queue and a second queue. The method includes selecting a first packet from the first queue for processing, which comprises performing a corresponding action stored in the instruction memory; selecting a second packet from the second queue for processing, which comprises forwarding a portion of the second packet to a processor, which looks up a second match-action table; and receiving, from the processor, a lookup result, thereby allowing a third packet in the first queue corresponding to the second packet to be processed based on the lookup result.

Abstract: The efficient storage of transformation information in a switch is provided. A respective port of the switch can include a memory device capable of storing transformation information. During operation, the switch can apply a selection mechanism to the transformation information learned at the switch for identifying a target port. The switch can then store the information in the memory device of the target port. Upon receiving a packet, the ingress port can apply the selection mechanism to the header information of the packet for determining a location of a first piece of transformation information associated with the packet. The ingress port can obtain the first piece of transformation information by looking up the header information in the location and storing it in a local memory device. The ingress port can then transform the packet based on the first piece of transformation information for determining an egress port for the packet.

Abstract: A packet-processing method includes looking up a first match-action table on a network interface card (NIC) for a received packet; in response to finding a matching entry indicating an action, queuing the received packet in a first queue and storing the action data in an instruction memory; and responsive to not finding a matching entry, queuing the received packet in the first queue and a second queue. The method includes selecting a first packet from the first queue for processing, which comprises performing a corresponding action stored in the instruction memory; selecting a second packet from the second queue for processing, which comprises forwarding a portion of the second packet to a processor, which looks up a second match-action table; and receiving, from the processor, a lookup result, thereby allowing a third packet in the first queue corresponding to the second packet to be processed based on the lookup result.

Abstract: The efficient storage of transformation information in a switch is provided. A respective port of the switch can include a memory device capable of storing transformation information. During operation, the switch can apply a selection mechanism to the transformation information learned at the switch for identifying a target port. The switch can then store the information in the memory device of the target port. Upon receiving a packet, the ingress port can apply the selection mechanism to the header information of the packet for determining a location of a first piece of transformation information associated with the packet. The ingress port can obtain the first piece of transformation information by looking up the header information in the location and storing it in a local memory device. The ingress port can then transform the packet based on the first piece of transformation information for determining an egress port for the packet.

Abstract: The efficient storage of transformation information in a switch is provided. A respective port of the switch can include a memory device capable of storing transformation information. During operation, the switch can apply a selection mechanism to the transformation information learned at the switch for identifying a target port. The switch can then store the information in the memory device of the target port. Upon receiving a packet, the ingress port can apply the selection mechanism to the header information of the packet for determining a location of a first piece of transformation information associated with the packet. The ingress port can obtain the first piece of transformation information by looking up the header information in the location and storing it in a local memory device. The ingress port can then transform the packet based on the first piece of transformation information for determining an egress port for the packet.

Abstract: A packet-processing method includes looking up a first match-action table on a network interface card (NIC) for a received packet; in response to finding a matching entry indicating an action, queuing the received packet in a first queue and storing the action data in an instruction memory; and responsive to not finding a matching entry, queuing the received packet in the first queue and a second queue. The method includes selecting a first packet from the first queue for processing, which comprises performing a corresponding action stored in the instruction memory; selecting a second packet from the second queue for processing, which comprises forwarding a portion of the second packet to a processor, which looks up a second match-action table; and receiving, from the processor, a lookup result, thereby allowing a third packet in the first queue corresponding to the second packet to be processed based on the lookup result.

Abstract: Systems and methods are described for providing per traffic class routing of data within a network. A network switch has the capability to classify traffic data based on High Performance Computing (HPC) related characteristics. Traffic classes are defined based on aspects of HPC, such as routing, ordering, redirection, quiesce, HPC protocol configuration, and telemetry. A switch can receive packets at an ingress port of a switch fabric, and determine traffic classifications for the packets. The traffic classification is selected from a group of defined traffic classes. Then, the switch can generate a fabric specific flag for the at least one packet that indicates the determined traffic classification, where the fabric specific flag is used for routing packets based on their assigned traffic classification. Examples of traffic classes include: low latency class; dedicated access class; bulk data class; best efforts class; and scavenger class.

Abstract: A switch architecture for a data-driven intelligent networking system is provided. The system can accommodate dynamic traffic with fast, effective congestion control. The system can maintain state information of individual packet flows, which can be set up or released dynamically based on injected data. Each flow can be provided with a flow-specific input queue upon arriving at a switch. Packets of a respective flow are acknowledged after reaching the egress point of the network, and the acknowledgement packets are sent back to the ingress point of the flow along the same data path. As a result, each switch can obtain state information of each flow and perform flow control on a per-flow basis.

Abstract: A switch architecture for a data-driven intelligent networking system is provided. The system can accommodate dynamic traffic with fast, effective congestion control. The system can maintain state information of individual packet flows, which can be set up or released dynamically based on injected data. Each flow can be provided with a flow-specific input queue upon arriving at a switch. Packets of a respective flow are acknowledged after reaching the egress point of the network, and the acknowledgement packets are sent back to the ingress point of the flow along the same data path. As a result, each switch can obtain state information of each flow and perform flow control on a per-flow basis.

Abstract: Systems and methods are described for providing per traffic class routing of data within a network. A network switch has the capability to classify traffic data based on High Performance Computing (HPC) related characteristics. Traffic classes are defined based on aspects of HPC, such as routing, ordering, redirection, quiesce, HPC protocol configuration, and telemetry. A switch can receive packets at an ingress port of a switch fabric, and determine traffic classifications for the packets. The traffic classification is selected from a group of defined traffic classes. Then, the switch can generate a fabric specific flag for the at least one packet that indicates the determined traffic classification, where the fabric specific flag is used for routing packets based on their assigned traffic classification. Examples of traffic classes include: low latency class; dedicated access class; bulk data class; best efforts class; and scavenger class.

Abstract: A switch architecture for a data-driven intelligent networking system is provided. The system can accommodate dynamic traffic with fast, effective congestion control. The system can maintain state information of individual packet flows, which can be set up or released dynamically based on injected data. Each flow can be provided with a flow-specific input queue upon arriving at a switch. Packets of a respective flow are acknowledged after reaching the egress point of the network, and the acknowledgement packets are sent back to the ingress point of the flow along the same data path. As a result, each switch can obtain state information of each flow and perform flow control on a per-flow basis.