Abstract: Examples disclosed herein relate to path-synchronous performance monitoring of an interconnection network based on source code attribution. A processing node in the interconnection network has a profiler module to select a network transaction to be monitored, determine a source code attribution associated with the network transaction to be monitored, and issue a network command to execute the network transaction to be monitored. A logger module creates, in a buffer, a node temporal log associated with the network transaction and the network command. A drainer module periodically captures the node temporal log. The processing node has a network interface controller to receive the network command and mark a packet generated for the network command to be temporally tracked and attributed back to the source code attribution at each hop of the interconnection network traversed by the marked packet.

Abstract: Examples disclosed herein relate to partial cache coherence. In some examples disclosed herein, a node connected to a memory fabric may include local cache connected to a local processor and a memory coherency proxy to. The memory coherency proxy may configure a portion of a fabric memory on the memory fabric as a proxy backing memory and expose the proxy backing memory to other nodes in the memory fabric as a fictitious local memory on the node, and may implement partial coherency for memory requests directed to the fictitious local memory. The fictitious local memory may have a memory address region different from a memory address region of a native local memory on the node.

Abstract: Examples disclosed herein relate to path-synchronous performance monitoring of an interconnection network based on source code attribution. A processing node in the interconnection network has a profiler module to select a network transaction to be monitored, determine a source code attribution associated with the network transaction to be monitored, and issue a network command to execute the network transaction to be monitored. A logger module creates, in a buffer, a node temporal log associated with the network transaction and the network command. A drainer module periodically captures the node temporal log. The processing node has a network interface controller to receive the network command and mark a packet generated for the network command to be temporally tracked and attributed back to the source code attribution at each hop of the interconnection network traversed by the marked packet.

Abstract: Examples disclosed herein relate to locally polling the value of a flag to determine whether a resource is free for a thread to use in a system with multiple processing nodes that are incoherent with regards to each other. A flag in a direct attached memory to one of the processing nodes is set to indicate that the resource is not free for the thread to use. A previous tail of a lock list is determined from a list master. The previous tail is located on another one of the processing nodes.

Abstract: Examples disclosed herein relate to partial cache coherence. In some examples disclosed herein, a node connected to a memory fabric may include local cache connected to a local processor and a memory coherency proxy to. The memory coherency proxy may configure a portion of a fabric memory on the memory fabric as a proxy backing memory and expose the proxy backing memory to other nodes in the memory fabric as a fictitious local memory on the node, and may implement partial coherency for memory requests directed to the fictitious local memory. The fictitious local memory may have a memory address region different from a memory address region of a native local memory on the node.

Abstract: Examples disclosed herein relate to locally polling the value of a flag to determine whether a resource is free for a thread to use in a system with multiple processing nodes that are incoherent with regards to each other. A flag in a direct attached memory to one of the processing nodes is set to indicate that the resource is not free for the thread to use. A previous tail of a lock list is determined from a list master. The previous tail is located on another one of the processing nodes.

Abstract: An example method for allocating resources in accordance with aspects of the present disclosure includes collecting proposals from a plurality of modules, the proposals assigning the resources to the plurality of modules and resulting in topology changes in a computer network environment, identifying a set of proposals in the proposals, the set of proposals complying with policies associated with the plurality of modules, instructing the plurality of modules to evaluate the set of proposals, selecting a proposal from the set of proposals, and instructing at least one module associated with the selected proposal to instantiate the selected proposal.

Abstract: A technique includes using circuit switches to selectively couple packet switches of a switch assembly to the port connectors of the assembly.

Abstract: Embodiments herein relate to addition or modification to a forwarding table based on an address. A first packet having a source address and a location value may be received. The source address includes a source of the first packet and the location value indicates at least part of a route along a network to the source address. The forwarding table is not modified or no new entry is added to the forwarding table, if the forwarding table does not include the source address.

Abstract: One example provides a network device including a queue to receive frames from a source, a processor, and a memory communicatively coupled to the processor. The memory stores instructions causing the processor, after execution of the instructions by the processor, to determine whether a flow control threshold of the queue has been exceeded, and in response to determining that the flow control threshold of the queue has been exceeded, generate a message to be sent to the source of the frame that exceeded the flow control threshold. The message includes a pause duration for which the source is to stop transmitting frames.

Abstract: A method may include determining whether a host destination address of a packet received at a particular switch of a switching fabric is associated with a virtual switch address. In response to a determination that the host destination address is associated with the virtual switch address, the method may also include encapsulating the packet with the virtual switch address as a destination fabric address, where the virtual switch address is associated with a virtual switch including a first physical switch and a second physical switch. The method may further include selecting one of the first physical switch and the second physical switch as a routing path. The method may further include routing the packet to the selected physical switch based on the virtual switch address and transmitting the packet from the selected physical switch to a client switch.

Abstract: A configurable Clos network includes leafs and spines and a switch fabric that connects the leafs and the spines. The switch fabric couples each leaf port of each leaf to at least one spine port of each spine.

Abstract: A technique includes using circuit switches to selectively couple packet switches of a switch assembly to the port connectors of the assembly.

Abstract: An example method for allocating resources in accordance with aspects of the present disclosure includes collecting proposals from a plurality of modules, the proposals assigning the resources to the plurality of modules and resulting in topology changes in a computer network environment, identifying a set of proposals in the proposals, the set of proposals complying with policies associated with the plurality of modules, instructing the plurality of modules to evaluate the set of proposals, selecting a proposal from the set of proposals, and instructing at least one module associated with the selected proposal to instantiate the selected proposal.

Abstract: An optical interconnect (200) includes: a reflective body (230) having a first reflective surface (235) and a second reflective surface (240) opposite the first reflective surface (235); a first optical waveguide (205) that directs a first optical signal received from a first communicating device (105) to the first reflective surface (235); a second optical waveguide (210) that directs the first optical signal from the first reflective surface (235) of the reflective body (230) to a second communicating device (110); a third optical waveguide (215) that directs a second optical signal received from the second communicating device (110) to the second reflective surface (240) of the reflective body (230); and a fourth optical waveguide (220) that directs the second optical signal from the second reflective surface (240) of the reflective body (230) to the first communicating device (105).

Abstract: A method for load balancing Ethernet traffic within a fat tree network (315, 455) includes randomly assigning incoming messages (510) into hash classes using a hash function (520); allocating the hash classes among uplinks (550); and transmitting the incoming messages on the uplinks (550) according to the hash class. A network switch (515) for load balancing communication flows in a fat tree network (315, 455) includes downlinks (545) and uplinks (550); the network switch (515) being configured to route communication flows among the downlinks (545) and uplinks (550); a hash module (520) which receives a MAC address from a message (510) and outputs a hash address; and a TCAM lookup module (535) which allocates the hash address into a hash class and allocates the hash class to one of the uplinks (550).

Abstract: One example provides a network device including a queue to receive frames from a source, a processor, and a memory communicatively coupled to the processor. The memory stores instructions causing the processor, after execution of the instructions by the processor, to determine whether a flow control threshold of the queue has been exceeded, and in response to determining that the flow control threshold of the queue has been exceeded, generate a message to be sent to the source of the frame that exceeded the flow control threshold. The message includes a pause duration for which the source is to stop transmitting frames.

Abstract: Multi-tenant network provisioning is disclosed. An example method of multi-tenant network provisioning includes setting at least one rate limiter on output ports of a node in the network on a tenant-by-tenant basis. The method also includes enforcing communication rates over shared edge links based on the rate limiter.

Abstract: A configurable Clos network includes leafs and spines and a switch fabric that connects the leafs and the spines. The switch fabric couples each leaf port of each leaf to at least one spine port of each spine.

Abstract: In a method (400) for routing packets between a plurality of top switches (110a-110n) and a plurality of leaf switches (120a-120n) using a balancing table (204, 208, 210) in a fat tree network (100), a failed link between at least one top switch (110n) and at least one leaf switch (120n) is detected (402). In addition, the balancing table (204, 208, 210) is modified (406) based on the detected failed link, and the packets are routed (408) between the plurality of top switches (110a-110n) and the plurality of leaf switches (120a-120n) in the fat tree network (100) based on the modified balancing table (204, 208, 210).