A computer implemented method of determining non-intersecting primary and secondary routes between source and destination nodes in a communications network. The communication network represented by a graph data structure of nodes and edges, the edges having a weight corresponding to a resource involved in traversing the edged. The method including: defining a population set of primary routes, wherein each primary route in the population set identifies a secondary route based on a greedy route-finding algorithm applied to the graph having edges in the primary route excluded from the graph, such that the primary and secondary routes are non-intersecting routes through the graph; applying a genetic algorithm to the population set to iteratively select and crossover one or more pairs of primary routes in the population set; and selecting, from the population set, a primary route and corresponding secondary route having a lowest aggregate weight of edges.

Embodiments of this application disclose a collective communication method, apparatus, and system. The method includes: A first network device receives a first packet; the first network device receives at least one second packet; and the first network device sends a third packet based on the first packet and the at least one second packet. When no connection is established between the first network device and a terminal device, the first network device may aggregate and distribute collective communication packets by using a connection between the first terminal device and another terminal device.

Embodiments of this application disclose a collective communication method, apparatus, and system. The method includes: A first network device receives a first packet; the first network device receives at least one second packet; and the first network device sends a third packet based on the first packet and the at least one second packet. When no connection is established between the first network device and a terminal device, the first network device may aggregate and distribute collective communication packets by using a connection between the first terminal device and another terminal device.

A data link error feedback signaling system includes a transmitting network device and a receiving network device. The receiving network device may be operable to receive a network data unit from the transmitting network device over a data link, detect an error in the network data unit, and provide data link integrity information based on the error to the transmitting network device. The receiving network device may provide the data link integrity information by marking the data link flawed in a routing protocol, transmitting the data link integrity information via an informational protocol, and so on. The transmitting network device may respond to the data link integrity information, such as by marking the data link less preferred, marking the data link down, transmitting an alarm regarding the data link to a network operator, omitting taking an action upon determining that errors are below an error threshold, and so on.

A system, and corresponding method, is described for finding the optimal or the best set of routes from a master to each of its connected slaves, for all the masters and slaves using an interconnect, such as a network-on-chip (NoC). Some embodiments of the invention apply to a class of interconnects that utilize a two-dimensional mesh topology, wherein a set of switches are arranged on a two-dimensional grid. Masters (initiators or sources) inject data packets or traffic into the interconnect. Slaves (targets or destinations) service the data packets or traffic traveling through the interconnect. The interconnect includes switches and links that are part of a path. Additionally, one or more optimal routes, which is defined by the system, move the traffic in a way that avoids deadlock scenarios.

A system, and corresponding method, is described for finding the optimal or the best set of routes from a master to each of its connected slaves, for all the masters and slaves using an interconnect, such as a network-on-chip (NoC). Some embodiments of the invention apply to a class of interconnects that utilize a two-dimensional mesh topology, wherein a set of switches are arranged on a two-dimensional grid. Masters (initiators or sources) inject data packets or traffic into the interconnect. Slaves (targets or destinations) service the data packets or traffic traveling through the interconnect. The interconnect includes switches and links that are part of a path. Additionally, one or more optimal routes, which is defined by the system, move the traffic in a way that avoids deadlock scenarios.

Networks comprising multiple non-overlapping communication topologies are presented. The networks can include a fabric of interconnected network nodes capable of providing multiple communication paths among edge devices. A topology manager constructs communication topologies according to restriction criteria based on required security levels (e.g., top secret, secret, unclassified, etc.). Established topologies do not have overlapping networking infrastructure to within the bounds of the restriction criteria as allowed by the security levels.

Networks comprising multiple non-overlapping communication topologies are presented. The networks can include a fabric of interconnected network nodes capable of providing multiple communication paths among edge devices. A topology manager constructs communication topologies according to restriction criteria based on required security levels (e.g., top secret, secret, unclassified, etc.). Established topologies do not have overlapping networking infrastructure to within the bounds of the restriction criteria as allowed by the security levels.

A system, and corresponding method, is described for finding the optimal or the best set of routes from a master to each of its connected slaves, for all the masters and slaves using a Network-on-Chip (NoC). More precisely, some embodiments of the invention apply to a class of NoCs that utilize a two-dimensional mesh topology, wherein a set of switches are arranged on a two-dimensional grid. Masters (initiators or sources) inject data packets or traffic into the NoC. Slaves (targets or destinations) service the data packets or traffic traveling through the NoC. The NoC includes switches and links. Additionally, the optimal routes defined by the system includes moving the traffic in a way that avoids deadlock scenarios.

A method and apparatus for path computation includes generating a path computation element communication protocol (PCEP) message, whereby the PCEP message comprises a characteristic associated with a level of protection of a Protection Label Switch path (LSP) with a Working LSP; and transmitting the PCEP message for path computation.