Abstract: A method for all-to-all message exchange between program tasks including N>1 hierarchy levels ln, n=1 to N, in which a first level l1 includes a plurality of group tasks and each higher level l(n>1) includes at least one group of level l(n?1) groups to which that task belongs in respective hierarchy levels of the network topology; sending a message via the interconnection network to a respective destination task whose hierarchical identifier is determined; and using the hierarchical identifier to send the program task and the network topology, such that the resulting exchange pattern for the all-to-all message exchange exploits a hierarchical distance in the network topology in a desired manner.

Abstract: Congestion management for data traffic in a virtual domain identifies a congestion source and sends a message to the source to adjust data traffic rates. The source may be a virtual machine hosted by a physical server with one or more virtual servers incorporated. A congestion manager may identify the source and send the message to the source without affecting other data sources hosted by the physical server or the virtual servers. In some embodiments, information about the congestion source may be encapsulated in a packet payload readable only by the congestion source so only the congestion source receives the instruction to adjust the transmission rate.

Abstract: A virtual network is implemented on a physical network. A virtual network data packet is tunneled through the physical network via encapsulation within a physical network data packet and via transmission of the physical network data packet through the physical network. A network congestion notification capability of the virtual network is preserved and modified during transmission of virtual network data through the physical network and vice-versa. Congestion notification metadata can be copied from a header of a virtual network data packet to a header of a physical network data packet when the virtual network data packet is encapsulated into the physical network data packet. Congestion notification metadata can be copied from a header of a physical network data packet to a header of a virtual network data packet when the virtual network data packet is decapsulated from the physical network data packet.

Abstract: In one embodiment, a system includes a hardware processor and logic integrated with and/or executable by the processor, the logic being adapted to receive a traffic flow having a plurality of packets, classify the traffic flow into a traffic class based on a characteristic of the traffic flow, the traffic class being selected from a plurality of traffic classes, store an identifier of the selected traffic class to one or more of the packets, and transmit the traffic flow according to its destination based on a priority of its selected traffic class. In more embodiments, additional systems, methods, and computer program products for prioritizing traffic flow handling are described.

Abstract: In one embodiment, a system includes a hardware processor and logic integrated with and/or executable by the processor. The logic is configured to classify a traffic flow into a traffic class based on at least one criteria related to the traffic flow and request that a credit manager remap flow credits corresponding to a first traffic class to flow credits corresponding to a second traffic class. In another embodiment, a method for providing credit-based flow control includes classifying a traffic flow into a traffic class based on at least one criteria related to the traffic flow, the traffic class being selected from a plurality of traffic classes. The method also includes storing an identifier indicating the traffic class of packets of the traffic flow according to a virtual local area network (VLAN) identifier in a three bit VLAN tag portion of a header of one or more of the packets.

Abstract: In one embodiment, a computer program product includes a computer readable storage medium having program instructions embodied therewith. The embodied program instructions are readable/executable by a processor to receive, by the processor, a packet via a network fabric, the network fabric having a plurality of interconnected fabric switches. The embodied program instructions are also readable/executable by the processor to determine, by the processor, a path through the network fabric by consulting a source-routing table. Moreover, the embodied program instructions are readable/executable by the processor to store, by the processor, source-routing information to a packet header for the packet, the source-routing information including the path. In addition, the embodied program instructions are readable/executable by the processor to send, by the processor, the packet according to an indication in the source-routing information.

Abstract: A method and system are provided for implementing a hierarchical high radix switch with a time-sliced crossbar. The hierarchical high radix switch includes a plurality of inputs and a plurality of outputs. Each input belongs to one input group; each input group sends consolidated requests to each output, by ORing the requests from the local input ports in that input group. Each output port belongs to one output group; each output port grants one of the requesting input groups using a rotating priority defined by a next-to-serve pointer. Each output group consolidates the output port grants and allows one grant to pass back to an input group. Each input port in an input group evaluates all incoming grants in an oldest packet first manner to form an accept. Each input group consolidates the input port accepts and selects one accept to send to the output port.

Abstract: Congestion management for data traffic in a virtual domain identifies a congestion source and sends a message to the source to adjust data traffic rates. The source may be a virtual machine hosted by a physical server with one or more virtual servers incorporated. A congestion manager may identify the source and send the message to the source without affecting other data sources hosted by the physical server or the virtual servers. In some embodiments, information about the congestion source may be encapsulated in a packet payload readable only by the congestion source so only the congestion source receives the instruction to adjust the transmission rate.

Abstract: A method to investigate congestion in a computer network may include network devices to route data packets throughout the network. The method includes, for example, sending a probe packet to network devices from a source node to gather information about the traffic queues at each network device that is examined by the probe packet. A routing table at each network device that receives the probe packet is based on the gathered information for respective each traffic queue.

Abstract: IEEE 802.1Q and Enhanced Transmission Selection provide only eight different traffic classes that may be used to control bandwidth in a particular physical connection (or link). Instead of relying only on these eight traffic classes to manage bandwidth, the embodiments discussed herein disclose using an Enhanced Transmission Selection scheduler that permits a network device to set the bandwidth for an individual virtual LAN. Allocating bandwidth in a port based on a virtual LAN ID permits a network device to allocate bandwidth to, e.g., millions of unique virtual LANs. Thus, this technique may increase the granular control of the network fabric and its performance.

Abstract: A method for generating a switch fabric topology, comprising constructing a first switch fabric topology, modifying the first switch fabric topology to generate a second switch fabric topology, wherein modifying the first switch fabric topology comprises isolating center stage sets of the first switch fabric topology, and replacing each of the isolated center stage sets with a single × switching element to generate the second switch fabric topology, wherein is an integer representing a radix of the switching element determined in connection with the constructing of the first switch fabric topology.

Abstract: A method for generating a switch fabric topology, comprising constructing a first switch fabric topology, modifying the first switch fabric topology to generate a second switch fabric topology, wherein modifying the first switch fabric topology comprises isolating center stage sets of the first switch fabric topology, and replacing each of the isolated center stage sets with a single × switching element to generate the second switch fabric topology, wherein is an integer representing a radix of the switching element determined in connection with the constructing of the first switch fabric topology.

Abstract: An approach is described for routing data to a plurality of output terminals via a integrated switch router including a crossbar switch having both a crossbar and a plurality of crossbar bypass lines. Whereas the crossbar may connect each input of the crossbar switch to each output of the crossbar switch, each of the plurality of crossbar bypass lines may connect a single input of the crossbar switch to a corresponding single output of the crossbar switch. According to such approach, a replicated copy of a multicast packet may be forwarded to an output terminal via a crossbar bypass line in parallel with other data forwarded via the crossbar, thus increasing integrated switch router bandwidth.

Abstract: In one embodiment, a system includes a network having one or more devices interconnected therebetween, the network including a first device having a first port and logic integrated with and/or executable by a first processor, the logic being adapted to exchange credit exchange (CE) capabilities between the first port and a second port of a second device in the network, receive a credit grant packet from the second port, the credit grant packet indicating credits granted to the first port, and send one or more packets of data from the first port to the second port, the one or more packets corresponding with the granted credits. Other systems and methods for providing credit-based flow control are described according to more embodiments.

Abstract: Disclosed herein is a method for deadlock avoidance in a network comprising partitioning in a module, a set of virtual lanes into deadlock avoidance virtual channels; where each deadlock avoidance virtual channel comprises a plurality of virtual lanes; assigning each incoming packet at the module to one of the deadlock avoidance virtual channels, and within the deadlock avoidance virtual channel to a virtual lane whose relative index corresponds to an absolute index of the virtual lane the packet would have been assigned to if the network was not provisioned with deadlock avoidance virtual channels; and transitioning each packet as it traverses the module from one deadlock avoidance virtual channel to another deadlock avoidance virtual channel when a deadlock avoidance policy so demands it.

Abstract: Flow control of data packets in a network may be enabled to at least one side of a virtual switching interface to provide a lossless environment. In some embodiments, wherever two buffer queues are in communication with at least one buffer queue being connected to a virtual switching interface, flow control may be used to determine if a threshold has been exceeded in one of the buffer queues. When exceeded, the transmission of data packets may cease to one of the buffer queues to prevent packet dropping and loss of data.

Abstract: A method for scheduling a crossbar using distributed request-grant-accept arbitration between input group arbiters and output group arbiters in a switch unit is provided. The switch unit may be a hierarchical high radix switch with a timesliced crossbar that is configured to transfer packets between a plurality of input ports and a plurality of output ports, organized into groups, using wide words. The timesliced crossbar transfers data for a given packet once per supercycle, in a designated timeslice of that supercycle. Multiple buffered packets from one input port to multiple output ports are transferred by utilizing different timeslices of the supercycle.

Abstract: A method for scheduling a crossbar using distributed request-grant-accept arbitration between input group arbiters and output group arbiters in a switch unit is provided. The switch unit may be a hierarchical high radix switch with a timesliced crossbar that is configured to transfer packets between a plurality of input ports and a plurality of output ports, organized into groups, using wide words. The timesliced crossbar transfers data for a given packet once per supercycle, in a designated timeslice of that supercycle. Multiple buffered packets from one input port to multiple output ports are transferred by utilizing different timeslices of the supercycle.

Abstract: A method for generating a switch fabric topology, comprising constructing a first switch fabric topology, modifying the first switch fabric topology to generate a second switch fabric topology, wherein modifying the first switch fabric topology comprises isolating center stage sets of the first switch fabric topology, and replacing each of the isolated center stage sets with a single × switching element to generate the second switch fabric topology, wherein is an integer representing a radix of the switching element determined in connection with the constructing of the first switch fabric topology.

Abstract: A method for generating a switch fabric topology, comprising constructing a first switch fabric topology, modifying the first switch fabric topology to generate a second switch fabric topology, wherein modifying the first switch fabric topology comprises isolating center stage sets of the first switch fabric topology, and replacing each of the isolated center stage sets with a single × switching element to generate the second switch fabric topology, wherein is an integer representing a radix of the switching element determined in connection with the constructing of the first switch fabric topology.