Abstract: In one embodiment, a method comprises identifying a fat tree network topology comprising top-of-fabric (ToF) switching devices, an intermediate layer of intermediate switching devices connected to each of the ToF switching devices, and a layer of leaf network devices; and causing a first leaf network device to initiate establishment of first and second redundant multicast trees for multicasting of data packets, including: causing first and second ToF switching devices to operate as roots of the first and second multicast trees according to first and second attribute types, respectively, causing the first leaf network device to select first and second of the intermediate switching devices as first and second flooding relays belonging to the first and second attribute types, respectively, and causing the first and second flooding relays to limit propagation of registration messages generated by the first leaf network device to the first and second ToF switching devices, respectively.

Abstract: In one embodiment, a method comprises causing, by an apparatus, establishment of first and second multicast trees within one or more underlay switching fabrics of one or more fat tree topologies, the first and second multicast trees comprising first and second multicast roots for multicast transmission to leaf network devices in the respective first and second multicast trees; causing, by the apparatus, establishment of an overlay tunnel between the first and second multicast roots, the overlay tunnel independent and distinct from the first and second multicast trees; causing the first multicast root to multicast transmit, via the first multicast tree, a data packet having been transmitted to the first multicast root; and causing the first multicast root to unicast transmit the data packet to the second multicast root via the overlay tunnel, for multicast transmission of the data packet by the second multicast root via the second multicast tree.

Abstract: In one embodiment, a supervisory device for a software defined networking (SDN) fabric predicts characteristics of a new traffic flow to be admitted to the fabric, based on a set of initial packets of the flow. The supervisory device predicts an impact of admitting the flow to the SDN fabric, using a heatmap-based saturation model for the SDN fabric. The supervisory device admits the flow to the SDN fabric, based on the predicted impact. The supervisory device uses reinforcement learning to adjust one or more call admission control (CAC) parameters of the SDN fabric, based on captured telemetry data regarding the admitted flow.

Abstract: In one embodiment, a method comprises identifying within a network topology, by an apparatus, a plurality of network devices; and establishing by the apparatus, a multiple tree topology comprising a first multicast tree and a second multicast tree, the first and second multicast trees operable as redundant trees for multicast traffic in the network topology, the establishing including: allocating a first of the network devices as a corresponding root of the first multicast tree, allocating a first group of intermediate devices from the network devices as first forwarding devices in the first multicast tree, allocating a second group of intermediate devices as belonging to first leaf devices in the first multicast tree, and allocating terminal devices of the network devices as belonging to the first leaf devices, and allocating a second of the network devices as the corresponding root of the second multicast tree, allocating the second group of intermediate devices as second forwarding devices in the second mu

Abstract: In one embodiment, a method includes receiving current data, the current data including time series data representing a plurality of time instances. The method includes storing at least a recent portion of the current data in a buffer. The method includes reducing the dimensionality of the current data to generate dimensionality-reduced data. The method includes generating a reconstruction error based on the dimensionality-reduced data and a plurality of neural network metrics. At least one of a size of the recent portion of the current data stored in the buffer or an amount of the reducing the dimensionality of the current data is based on the reconstruction error.

Abstract: In one embodiment, a supervisory device for a software defined networking (SDN) fabric obtains telemetry data regarding congestion levels on a plurality of links in the SDN fabric. The supervisory device predicts seasonal congestion on a particular one of the plurality of links by using the telemetry data as input to a machine learning-based model. The supervisory device identifies a period of time associated with the predicted seasonal congestion on the particular link. The supervisory device initiates, in advance of the identified period of time, re-computation of equal-cost multi-path (ECMP) weights associated with the plurality of links that prevent occurrence of the predicted seasonal congestion on the particular link during the identified period of time.

Abstract: In one illustrative example, a network node connected in a network fabric may identify that it is established as part of a multicast distribution tree for forwarding multicast traffic from a source node to one or more host receiver devices of a multicast group. In response, the network node may propagate in the network fabric a message for advertising the network node as a candidate local source node at which to join the multicast group. The message for advertising may include data such as a reachability metric. The propagation of the message may be part of a flooding of such messages in the network fabric. The network node serving as the candidate local source node may thereafter “locally” join a host receiver device in the multicast group at the network node so that the device may receive the multicast traffic from the source node via the network node.

Abstract: In one embodiment, a supervisory service receives a registration message broadcast by a first vehicle and captured by a RSU in the network of RSUs. The supervisory service registers the first vehicle by validating a signature of the registration message without registering a media access control (MAC) address of the first vehicle and without causing to send a registration response to the first vehicle. The supervisory service receives a message broadcast by a second vehicle addressed to the first vehicle and captured by at least one RSU in the network of RSUs. The supervisory service selects one or more RSUs in the network of RSUs to re-broadcast the message. The supervisory service controls the one or more RSUs to re-broadcast the message.

Abstract: In one embodiment, a particular device along a path in a deterministic network receives a first packet sent from a source towards a destination via the path. The particular device sends the first packet to a next hop device along the path, according to a deterministic schedule associated with the first packet. The particular device determines, after sending the first packet, an action to be performed on the first packet. The particular device then sends a second packet to the next hop device indicative of the determined action. The second packet causes another device along the path to perform the action on the first packet.

Abstract: A first address resolution request may be received by a first access switch from a first device and the address resolution request may be resolved by the first access switch with a central database of a network. Then a second address resolution request may be sent to a sensor by the first access switch in response to resolving the first address resolution request. An address resolution response may then be sent by the sensor to the first device in response to the sensor determining that the first device is a bad endpoint. A session may then be established between the sensor and the first device in response to the sensor sending the address resolution response. The first device may then be prompted by the sensor via the established session to resolve issues that lead the sensor to determine that the first device is a bad endpoint.

Abstract: A wireless device can achieve higher predictability for its transmissions by inserting a placeholder frame in a transmission queue before RoCE data has been received. In addition, a contention countdown associated with the placeholder frame can start before the RoCE data is ready for transmission. Once the RoCE data is available, the device can insert the data into the payload of the placeholder frame, thereby reducing the wait time before the RoCE data can be transmitted wirelessly. Additionally, the device can improve reliability by transmitting RoCE data using multiple subcarrier RUs in a channel. The data blocks and the duplicative data can be transmitted in parallel using the subcarrier RUs. If a subset of the subcarrier RUs are blocked because of narrowband interference, the receiving device can nonetheless recover the data blocks and reconstruct the RoCe packet from the data transported on the RUs that did not have interference.

Abstract: In one embodiment, a first actuator in a network of sensors and actuators executes a walker agent configured to adjust an actuation setting of the first actuator. The actuation setting controls an area of coverage of the first actuator when actuated. The executing agent on the first actuator receives one or more sensor measurements from one or more of the sensors that are in communication range of the first actuator. The executing agent also controls, based on the received one or more sensor measurements, the area of coverage of the first actuator by adjusting its actuation setting, in an attempt to optimize coverage of the sensors in the network by the areas of coverage of the actuators. The first actuator unloads the executing walker agent after adjusting the actuation setting of the first actuator and propagates the agent to another one of the actuators in the network for execution.

Abstract: In one embodiment, a parent device in an unaligned wireless network may determine a superframe comprising a header timeslot followed by a plurality of sub timeslots. The parent device may transmit, to a plurality of child devices in the unaligned wireless network, a beacon during the header timeslot, wherein the beacon comprises i) synchronization information used by the plurality of child devices to synchronize to the header timeslot and ii) reservation information that indicates one or more reserved sub timeslots of the plurality of sub timeslots. The parent device may receive, from a particular child device of the plurality of child devices, a message during a particular sub timeslot of the plurality of sub timeslots that is different than the one or more reserved sub timeslots.

Abstract: In one embodiment, a parent network device, operating according to a first Trickle operation using a first selected minimum contention interval, responds to detecting a loss of attached child network devices by starting a second Trickle operation using a second selected minimum contention interval. The second Trickle operation includes maintaining the second selected minimum contention interval for subsequent iterations of the second Trickle operation. The parent network device initiates an accelerated transmission rate of the advertisement message that is faster than the first and second Trickle operations (using a third selected minimum contention interval less than the first minimum contention interval) in response to receiving a message from one of the lost child network devices, and resumes the first Trickle operation upon recovery of all the lost child network devices.

Abstract: In one embodiment, a supervisory device in a network forms a virtual access point (VAP) for a node in the network. A set of access points (APs) in the network are mapped to the VAP as part of a VAP mapping and the node treats the APs in the VAP mapping as a single AP for purposes of communicating with the network. The supervisory device receives measurements from the APs in the VAP mapping regarding communications associated with the node. The supervisory device identifies a movement of the node based on the received measurements from the APs in the VAP mapping. The supervisory device adjusts the set of APs in the VAP mapping based on the identified movement of the node.

Abstract: In one embodiment, a device in a network receives a query walker agent configured to query information from a distributed set of devices in the network based on a query. The device executes the query walker agent to identify the query. The device updates state information of the executing query walker agent using local information from the device and based on the query. The device unloads the executing query walker agent after updating the state information. The device propagates the query walker agent with the updated state information to one or more of the distributed set of devices in the network, when the updated state information does not fully answer the query.

Abstract: In one embodiment, a cloud-based service instructs one or more networking devices in a local area network (LAN) to form a virtual network overlay in the LAN that redirects traffic associated with a particular node in the LAN to the service. The service receives multicast or broadcast traffic sent by the particular node in the LAN and redirected to the service via the virtual network overlay. The service identifies a group of nodes in the network that are to receive the traffic sent by the particular node, based in part by profiling the traffic associated with the particular node. The service sends the traffic sent by the particular node to at least one networking device in the LAN with an indication of the identified group of nodes in the network that are to receive the traffic sent by the particular node. The at least one networking device forwards the traffic sent by the particular node to the nodes in the identified group.

Abstract: In one embodiment, a server instructs one or more networking devices in a local area network (LAN) to form a virtual network overlay in the LAN that redirects traffic associated with a particular node in the LAN to the server. The server receives the redirected traffic associated with the particular node. The server trains a machine learning-based behavioral model for the particular node based on the redirected traffic. The server controls whether a particular redirected traffic flow associated with the node in the LAN is sent to a destination of the traffic flow using the trained behavioral model.

Abstract: The present disclosure provides a proactive method of prefix disaggregation in a network fabric when one or more communication failures are detected. In one aspect, a method includes determining, by a first node of a network fabric, a corresponding prefix disaggregation policy for at least one second node of the network fabric, the corresponding prefix disaggregation policy identifying one or more network prefixes that are inaccessible via the first node when at least one communication failure is detected in association with the first node; sending the corresponding prefix disaggregation policy to the second node; and causing the second node to implement the prefix disaggregation policy upon detecting the at least one communication failure.

Abstract: In one embodiment, a method comprises: identifying, by a root network device of a directed acyclic graph (DAG) in a low power and lossy network, a child network device in the DAG, including identifying a first rank associated with the child network device; allocating, by the root network device, an allocated rank for the child network device, the allocated rank different from the first rank; and outputting, by the root network device, a message to the child network device specifying the allocated rank, the message causing the child network device to implement the allocated rank in the DAG, including causing the child network device to generate and output a Destination Oriented Directed Acyclic Graph (DODAG) information object (DIO) message specifying the child network device is using the allocated rank.