Abstract: A network element includes processing circuitry and multiple ports. The ports connect using links to a communication network. The processing circuitry is configured to receive packets via the ports and forward the received packets to respective destination addresses via the ports. The destination addresses are organized in address groups, each address group including multiple destination addresses of nodes connected to a common network element in the communication network. The processing circuitry is further configured to, in response to identifying that a given port connects to a faulty link, determine one or more address groups that became unreachable via the given port due to the faulty link, generate a notification reporting one or more of the determined address groups that are unreachable via any port other than the given port, and transmit the notification to one or more other network elements, via one or more ports other than the given port.

Abstract: An apparatus includes an interface and a processor. The interface communicates with a network including network elements interconnected in a Cartesian topology. The processor defines first and second groups of turns, each turn includes a hop from a previous network element to a current network element and a hop from the current network element to a next network element. Based on the turns, the processor specifies rules that when applied to packets traversing respective network elements, guarantee that no deadlock conditions occur in the network. The rules for a given network element include (i) forwarding rules to reach a given target without traversing the turns of the second group, and (ii) Virtual Lane (VL) modification rules for reassigning packets, which traverse turns of the first group and which are assigned to a first VL, to a different second VL. The processor configures the given network element with the rules.

Abstract: A network device includes multiple ports, multiple buffer slices, a controller, and buffer control circuitry. The multiple ports are configured to communicate packets over a network. The multiple buffer slices are linked respectively to the multiple ports. The controller is configured to allocate a group of two or more of the buffer slices to a selected port among the ports. The buffer control circuitry is configured to buffer the packets, communicated via the selected port, in the group of the buffer slices, using zero-copy buffering.

Abstract: A router includes routing circuitry and a plurality of ports. The routing circuitry is configured to receive from a first subnetwork, via one of the ports, a packet destined to be delivered to a target node located in a second subnetwork, to select a mapping, from among two or more mappings, depending on a topological relation between the first subnetwork and the second subnetwork, to map a Layer-3 address of the packet into a Layer-2 address using the selected mapping, and to forward the packet via another one of the ports to the Layer-2 address.

Abstract: An apparatus includes an interface and a processor. The interface communicates with a network including network elements interconnected in a Cartesian topology. The processor defines first and second groups of turns, each turn includes a hop from a previous network element to a current network element and a hop from the current network element to a next network element. Based on the turns, the processor specifies rules that when applied to packets traversing respective network elements, guarantee that no deadlock conditions occur in the network. The rules for a given network element include (i) forwarding rules to reach a given target without traversing the turns of the second group, and (ii) Virtual Lane (VL) modification rules for reassigning packets, which traverse turns of the first group and which are assigned to a first VL, to a different second VL. The processor configures the given network element with the rules.

Abstract: A router includes routing circuitry and a plurality of ports. The routing circuitry is configured to receive from a first subnetwork, via one of the ports, a packet destined to be delivered to a target node located in a second subnetwork, to select a mapping, from among two or more mappings, depending on a topological relation between the first subnetwork and the second subnetwork, to map a Layer-3 address of the packet into a Layer-2 address using the selected mapping, and to forward the packet via another one of the ports to the Layer-2 address.

Abstract: A computing system including network elements arranged in at least one group. A plurality of the network elements are designated as spines and another plurality are designated as leaves, the spines and leaves are interconnected in a bipartite topology, and at least some of the spines and leaves are configured to: receive in a first leaf, from a source node, packets destined to a destination node via a second leaf, forward the packets via a first link to a first spine and to the second leaf via a second link, in response to detecting that the second link has failed, apply a detour path from the first leaf to the second leaf, including a detour link in a spine-to-leaf direction and another detour link a leaf-to-spine direction, and forward subsequent packets, which are received in the first leaf and are destined to the second leaf, via the detour path.

Abstract: A network element includes a plurality of ports, multiple computational modules, configurable forwarding circuitry and a central block. The ports include child ports coupled to child network elements or network nodes and parent ports coupled to parent network elements. The computational modules collectively perform a data reduction operation of a data reduction protocol. The forwarding circuitry interconnects among ports and computational modules. The central block receives a request indicative of child ports, a parent port, and computational modules required for performing reduction operations on data received via the child ports, for producing reduced data destined to the parent port, to derive from the request a topology that interconnects among the child ports, parent port and computational modules for performing the data reduction operations and to forward the reduced data for transmission to the selected parent port, and to configure the forwarding circuitry to apply the topology.

Abstract: A method in which a plurality of process are configured to hold a block of data destined for other processes, with data repacking circuitry including receiving circuitry configured to receive at least one block of data from a source process of the plurality of processes, the repacking circuitry configured to repack received data in accordance with at least one destination process of the plurality of processes, and sending circuitry configured to send the repacked data to the at least one destination process of the plurality of processes, receiving a set of data for all-to-all data exchange, the set of data being configured as a matrix, the matrix being distributed among the plurality of processes, and transposing the data by each of the plurality of processes sending matrix data from the process to the repacking circuitry, and the repacking circuitry receiving, repacking, and sending the resulting matrix data to destination processes.

Abstract: A network element includes multiple ports and forwarding circuitry. The ports are configured to serve as network interfaces for exchanging packets with a communication network. The forwarding circuitry is configured to receive a multicast packet that is to be forwarded via a plurality of the ports over a plurality of paths through the communication network to a plurality of destinations, to identify a path having a highest latency among the multiple paths over which the multicast packet is to be forwarded, to forward the multicast packet to one or more of the paths other than the identified path, using a normal scheduling process having a first forwarding latency, and to forward the multicast packet to at least the identified path, using an accelerated scheduling process having a second forwarding latency, smaller than the first forwarding latency.

Abstract: A method in which a plurality of process are configured to hold a block of data destined for other processes, with data repacking circuitry including receiving circuitry configured to receive at least one block of data from a source process of the plurality of processes, the repacking circuitry configured to repack received data in accordance with at least one destination process of the plurality of processes, and sending circuitry configured to send the repacked data to the at least one destination process of the plurality of processes, receiving a set of data for all-to-all data exchange, the set of data being configured as a matrix, the matrix being distributed among the plurality of processes, and transposing the data by each of the plurality of processes sending matrix data from the process to the repacking circuitry, and the repacking circuitry receiving, repacking, and sending the resulting matrix data to destination processes.

Abstract: A network element includes at least one headroom buffer, and flow-control circuitry. The headroom buffer is configured for receiving and storing packets from a peer network element having at least two data sources, each headroom buffer serving multiple packets. The flow-control circuitry is configured to quantify a congestion severity measure, and, in response to detecting a congestion in the headroom buffer, to send to the peer network element pause-request signaling that instructs the peer network element to stop transmitting packets that (i) are associated with the congested headroom buffer and (ii) have priorities that are selected based on the congestion severity measure.

Abstract: A network element includes a plurality of ports, multiple computational modules, configurable forwarding circuitry and a central block. The ports include child ports coupled to child network elements or network nodes and parent ports coupled to parent network elements. The computational modules collectively perform a data reduction operation of a data reduction protocol. The forwarding circuitry interconnects among ports and computational modules. The central block receives a request indicative of child ports, a parent port, and computational modules required for performing reduction operations on data received via the child ports, for producing reduced data destined to the parent port, to derive from the request a topology that interconnects among the child ports, parent port and computational modules for performing the data reduction operations and to forward the reduced data for transmission to the selected parent port, and to configure the forwarding circuitry to apply the topology.

Abstract: An apparatus includes one or more ports for connecting to a communication network, processing circuitry and a message aggregation circuit (MAC). The processing circuitry is configured to communicate messages over the communication network via the one or more ports. The MAC is configured to receive messages, which originate in one or more source processes and are destined to one or more destination processes, to aggregate two or more of the messages that share a common destination into an aggregated message, and to send the aggregated message using the processing circuitry over the communication network.

Abstract: A network element includes one or more ports for communicating over a network, a processor and packet processing hardware. The packet processing hardware is configured to transfer packets to and from the ports, and further includes data-transfer circuitry for data transfer with the processor. The processor and the data-transfer circuitry are configured to transfer between one another (i) one or more communication packets for transferal between the ports and the processor and (ii) one or more databases for transferal between the packet processing hardware and the processor, by (i) translating, by the processor, the transferal of both the communication packets and the databases into work elements, and posting the work elements on one or more work queues in a memory of the processor, and (ii) using the data-transfer circuitry, executing the work elements so as to transfer both the communication packets and the databases.

Abstract: A network element includes one or more ports for communicating over a network, a processor and packet processing hardware. The packet processing hardware is configured to transfer packets to and from the ports, and further includes data-transfer circuitry for data transfer with the processor. The processor and the data-transfer circuitry are configured to transfer between one another (i) one or more communication packets for transferal between the ports and the processor and (ii) one or more databases for transferal between the packet processing hardware and the processor, by (i) translating, by the processor, the transferal of both the communication packets and the databases into work elements, and posting the work elements on one or more work queues in a memory of the processor, and (ii) using the data-transfer circuitry, executing the work elements so as to transfer both the communication packets and the databases.

Abstract: Communication apparatus includes multiple ports configured to serve as ingress and egress ports, such that the ingress ports receive packets from a packet data network for forwarding to respective egress ports. The ports include an egress port configured for connection to a network interface controller (NIC) serving multiple physical computing units, which have different, respective destination addresses and are connected to the NIC by different, respective communication channels. Control and queuing logic is configured to queue the packets that are received from the packet data network for forwarding to the multiple physical computing units in different, respective queues according to the destination addresses, and to arbitrate among the queues so as to convey the packets from the queues via the same egress port to the NIC, for distribution to the multiple physical computing units over the respective communication channels.

Abstract: A network element including buffer address control circuitry for reading a given entry from a queue in a memory of a device external to the network element, the queue having at least a first entry and a last entry, the given entry including a destination address in the memory, output circuitry for writing data included in a packet received from external to the network element to the destination address in the memory in accordance with the given entry, and next entry assignment circuitry for assigning a next entry by: when the given entry is other than the last entry in the first queue, assigning the next entry to be an entry in the first queue after the given entry, and when the given entry is the last entry in the first queue, assigning the next entry to be the first entry in the first queue. Related apparatus and methods are also described.

Abstract: A method in which a plurality of process are configured to hold a block of data destined for other processes, with data repacking circuitry including receiving circuitry configured to receive at least one block of data from a source process of the plurality of processes, the repacking circuitry configured to repack received data in accordance with at least one destination process of the plurality of processes, and sending circuitry configured to send the repacked data to the at least one destination process of the plurality of processes, receiving a set of data for all-to-all data exchange, the set of data being configured as a matrix, the matrix being distributed among the plurality of processes, and transposing the data by each of the plurality of processes sending matrix data from the process to the repacking circuitry, and the repacking circuitry receiving, repacking, and sending the resulting matrix data to destination processes.

Abstract: A method for communication includes receiving in a computer system a request from a peer computer system. Upon finding that the computer system is currently not ready to process the request, a Negative Acknowledgement (NAK) message is sent from the computer system to the peer computer system, at a sending time that is derived from a time at which the computer system is ready to process the request.