Abstract: Technologies for improving throughput in a network include a node switch. The node switch is to obtain expected performance data indicative of an expected data transfer performance of the node switch. The node switch is also to obtain measured performance data indicative of a measured data transfer performance of the node switch, compare the measured performance data to the expected performance data to determine whether the measured data transfer performance satisfies the expected data transfer performance, determine, as a function of whether the measured data transfer performance satisfies the expected data transfer performance, whether to force a unit of data through a non-minimal path to a destination, and send, in response to a determination to force the unit of data to be sent through a non-minimal path, the unit of data to an output port of the node switch associated with the non-minimal path. Other embodiments are also described.

Abstract: Technologies for densely packaging network components for large scale indirect topologies include group of switches. The group of switches includes a stack of node switches that includes a first set of ports and a stack of global switches that includes a second set of ports. The stack of node switches are oriented orthogonally to the stack of global switches. Additionally, the first set of ports are oriented towards the second set of ports and the node switches are connected to the global switches through the first and second sets of ports. Other embodiments are also described and claimed.

Abstract: Technologies for link-bandwidth-aware routing are disclosed. In order to avoid congestion while still allowing link bandwidth to be decreased in order to save power, a network switch may select a port to send a packet over based on the present link bandwidth of the data links connected to the various output ports of the network switch. The network switch preferentially sends the packet over the minimal output port, or, if the minimal output port is congested, over one of the ports with the highest available link bandwidth. If the link bandwidth of the data link connected to the selected output port is not high enough, the network switch will automatically dynamically increase the link bandwidth of the data link as necessary.

Abstract: Technologies for densely packaging network components for large scale indirect topologies include group of switches. The group of switches includes a stack of node switches that includes a first set of ports and a stack of global switches that includes a second set of ports. The stack of node switches are oriented orthogonally to the stack of global switches. Additionally, the first set of ports are oriented towards the second set of ports and the node switches are connected to the global switches through the first and second sets of ports. Other embodiments are also described and claimed.

Abstract: Congestion management techniques for communication networks are described. In an example embodiment, an apparatus may comprise circuitry, a communication component for execution by the circuitry to receive a send request identifying a message to be received from an initiator device via a packet transfer process and transmit an acceptance to grant the send request, and a scheduling component for execution by the circuitry to determine whether to defer the packet transfer process and in response to a determination to defer the packet transfer process, select a value of a delay parameter to be included in the acceptance. Other embodiments are described and claimed.

Abstract: Systems, apparatuses and methods may provide for a plurality of node-level agents, wherein each node-level agent aggregates network statistics information from a plurality of probes associated with a communications interface. Additionally, one or more job-level agents may be communicatively coupled to the plurality of node-level agents, wherein each job-level agent aggregates network statistics information from two or more of the node-level agents. Moreover, a system-level agent may be communicatively coupled to the job-level agent(s). The system-level agent may generate a power model based on aggregated network statistics information from the job-level agent(s) and propagate the power model to the node-level agents via the job-level agent(s).

Abstract: Technologies for a system of communicatively coupled network switches in a hierarchical interconnect network topology include two or more groups that each include two or more first and second level switches in which each of the first level switches are communicatively coupled to each of the plurality of second level switches to form a complete bipartite graph. Additionally, each of the groups is interconnected to each of the other groups via a corresponding global link connecting a second level switch of one group to a corresponding second level switch of another group. Further, each of the first level switches are communicatively coupled to one or more computing nodes. Other embodiments are described herein.

Abstract: Technologies for adaptive routing based on network traffic pattern characterization include a network switch configured to receive a network packet via one of a plurality of input ports and identify a set of the plurality of output ports associated with a path usable to forward the received network packet to a destination computing device along. The network switch is further configured to adjust a total congestion value for each of the set of output ports based on a type of the path to which each of the set of output ports corresponds and a value of a minimal path counter to which each of the set of output ports corresponds and enqueue the received network packet into an output buffer queue of one of the set of output ports based on the total congestion value. Other embodiments are described herein.

Abstract: Technologies for adaptive routing using throughput estimation that includes a network switch. The network switch is configured to determine an adjusted average saturation count for each output buffer queue as a function of a present value of a saturation counter of a corresponding output buffer queue and a weighted average saturation count and a running average saturation count for each of the plurality of output buffer queues as a function of the corresponding captured present value and the adjusted average saturation count. The network switch is further configured to determine a congestion rate value for each output buffer queue and a total congestion value as a function of the congestion rate values and a standard occupancy congestion corresponding to a respective one of the plurality of output buffer queues. Other embodiments are described herein.

Abstract: Technologies for adaptive routing based on aggregated congestion information include a network switch that includes a plurality of output ports. The network switch is configured to determine a maximum local occupancy count for each output port based on a maximum local occupancy count of output buffer queues of each output port, a local congestion value based on the maximum local occupancy count, and a remote congestion value for a corresponding remote input buffer queue of a remote computing device communicatively coupled to a corresponding output port. The network switch is further configured to determine, for each output port, a total congestion value as a function of the local congestion value and the remote congestion value and enqueue the network packet into one of the output buffer queues of one of the output ports based on the total congestion values of the output ports. Other embodiments are described herein.

Abstract: Technologies for improving throughput in a network include a node switch. The node switch is to obtain expected performance data indicative of an expected data transfer performance of the node switch. The node switch is also to obtain measured performance data indicative of a measured data transfer performance of the node switch, compare the measured performance data to the expected performance data to determine whether the measured data transfer performance satisfies the expected data transfer performance, determine, as a function of whether the measured data transfer performance satisfies the expected data transfer performance, whether to force a unit of data through a non-minimal path to a destination, and send, in response to a determination to force the unit of data to be sent through a non-minimal path, the unit of data to an output port of the node switch associated with the non-minimal path. Other embodiments are also described.

Abstract: Technologies for densely packaging network components for large scale indirect topologies include group of switches. The group of switches includes a stack of node switches that includes a first set of ports and a stack of global switches that includes a second set of ports. The stack of node switches are oriented orthogonally to the stack of global switches. Additionally, the first set of ports are oriented towards the second set of ports and the node switches are connected to the global switches through the first and second sets of ports. Other embodiments are also described and claimed.

Abstract: Technologies for dynamic bandwidth management of interconnect fabric include a compute device configured to calculate a predicted fabric bandwidth demand which is expected to be used by the interconnect fabric in a next epoch and subsequent to a present epoch. The compute device is additionally configured to determine whether any global links and/or local links of the interconnect fabric can be disabled during the next epoch as a function of the calculated predicted fabric bandwidth demand and a number of redundant paths associated with the links of the interconnect fabric. The compute device is further configured to disable one or more of the global links and/or the local links that can be disabled, the one or more local links of the plurality of local links that can be disabled. Other embodiments are described herein.

Abstract: Technologies for efficiently managing link faults between switches include a fabric monitor. The fabric monitor is to generate routing rules indicative of an ordering of a plurality of global switches connected to a plurality of node switches in a group, monitor a status of links between the global switches and the node switches to determine whether one or more downlinks have failed in the group, adjust, in response to a determination that one or more downlinks have failed in the group, the ordering of the global switches in the routing rules, and send the adjusted routing rules to the group. Other embodiments are also described and claimed.

Abstract: Technologies for link-bandwidth-aware routing are disclosed. In order to avoid congestion while still allowing link bandwidth to be decreased in order to save power, a network switch may select a port to send a packet over based on the present link bandwidth of the data links connected to the various output ports of the network switch. The network switch preferentially sends the packet over the minimal output port, or, if the minimal output port is congested, over one of the ports with the highest available link bandwidth. If the link bandwidth of the data link connected to the selected output port is not high enough, the network switch will automatically dynamically increase the link bandwidth of the data link as necessary.

Abstract: Technologies for a system of communicatively coupled network switches in a hierarchical interconnect network topology include two or more groups that each include two or more first and second level switches in which each of the first level switches are communicatively coupled to each of the plurality of second level switches to form a complete bipartite graph. Additionally, each of the groups is interconnected to each of the other groups via a corresponding global link connecting a second level switch of one group to a corresponding second level switch of another group. Further, each of the first level switches are communicatively coupled to one or more computing nodes. Other embodiments are described herein.

Abstract: Technologies for adaptive routing using throughput estimation that includes a network switch. The network switch is configured to determine an adjusted average saturation count for each output buffer queue as a function of a present value of a saturation counter of a corresponding output buffer queue and a weighted average saturation count and a running average saturation count for each of the plurality of output buffer queues as a function of the corresponding captured present value and the adjusted average saturation count. The network switch is further configured to determine a congestion rate value for each output buffer queue and a total congestion value as a function of the congestion rate values and a standard occupancy congestion corresponding to a respective one of the plurality of output buffer queues. Other embodiments are described herein.

Abstract: Technologies for adaptive routing based on network traffic pattern characterization include a network switch configured to receive a network packet via one of a plurality of input ports and identify a set of the plurality of output ports associated with a path usable to forward the received network packet to a destination computing device along. The network switch is further configured to adjust a total congestion value for each of the set of output ports based on a type of the path to which each of the set of output ports corresponds and a value of a minimal path counter to which each of the set of output ports corresponds and enqueue the received network packet into an output buffer queue of one of the set of output ports based on the total congestion value. Other embodiments are described herein.

Abstract: Technologies for adaptive routing based on aggregated congestion information include a network switch that includes a plurality of output ports. The network switch is configured to determine a maximum local occupancy count for each output port based on a maximum local occupancy count of output buffer queues of each output port, a local congestion value based on the maximum local occupancy count, and a remote congestion value for a corresponding remote input buffer queue of a remote computing device communicatively coupled to a corresponding output port. The network switch is further configured to determine, for each output port, a total congestion value as a function of the local congestion value and the remote congestion value and enqueue the network packet into one of the output buffer queues of one of the output ports based on the total congestion values of the output ports. Other embodiments are described herein.

Abstract: Systems, apparatuses and methods may provide for a plurality of node-level agents, wherein each node-level agent aggregates network statistics information from a plurality of probes associated with a communications interface. Additionally, one or more job-level agents may be communicatively coupled to the plurality of node-level agents, wherein each job-level agent aggregates network statistics information from two or more of the node-level agents. Moreover, a system-level agent may be communicatively coupled to the job-level agent(s). The system-level agent may generate a power model based on aggregated network statistics information from the job-level agent(s) and propagate the power model to the node-level agents via the job-level agent(s).