10 results about "Distribution tree" patented technology

In a multirank memory system in which the clock distribution trees of each rank are permitted to drift over a wide range (e.g., low power memory systems), the fine-interleaving of commands between ranks is facilitated through the use of techniques that cause each addressed rank to properly sample commands intended for that rank, notwithstanding the drift. The ability to perform such “microthreading” provides for substantially enhanced memory capacity without sacrificing the performance of single rank systems. This disclosure provides methods, memory controllers, memory devices and system designs adapted to these ends.

In a multirank memory system in which the clock distribution trees of each rank are permitted to drift over a wide range (e.g., low power memory systems), the fine-interleaving of commands between ranks is facilitated through the use of techniques that cause each addressed rank to properly sample commands intended for that rank, notwithstanding the drift. The ability to perform such “microthreading” provides for substantially enhanced memory capacity without sacrificing the performance of single rank systems. This disclosure provides methods, memory controllers, memory devices and system designs adapted to these ends.

A multicast communication method is applied to a software-defined wide area network SD-WAN. In the multicast communication method, a bit index explicit replication (Bit Index Explicit Replication, BIER) capability is deployed on site devices in the SD-WAN, a BIER header in a multicast packet indicates a site device that a multicast packet needs to pass through, and no multicast distribution tree need to be constructed. In addition, the multicast packet is forwarded through an SD-WAN tunnel between site devices, and an intermediate node does not need to sense a multicast service, to reduce a quantity of multicast table entries on a network device and accelerate convergence of the multicast table entries, so that normal running of the multicast service is ensured.

Embodiments of the present application provide a data processing method, a server and an electronic whiteboard. The server receives registration requests sent by multiple electronic whiteboards in sequence, and establishes a data distribution tree according to the registration requests sent by the electronic whiteboards. The root node of the data distribution tree is the server, and each electronic whiteboard is a node of the data distribution tree. When the server receives data sent by at least one of the multiple electronic whiteboards, the data is preprocessed, and the preprocessed target data is sent to each child node of the server according to the data distribution tree. Each node in the data distribution tree that has a child node forwards the received target data to each child node corresponding to the node after receiving the target data. Embodiments of the present application can reduce the load of the server, effectively prevent the target data sent by the server from being lost, and ensure that the content displayed on each electronic whiteboard is consistent.

The application provides a video routing optimization method and device based on a graph neural network, comprising: constructing a network snapshot model; sorting users according to requirements to obtain a user sequence and initializing an inflow node set; using a graph neural network to encode nodes of a non-directional weighted graph to obtain context-aware embedding representation; according to the user sequence and the context-aware embedding representation, sequentially planning paths for each access user, selecting a next-hop node according to path context aggregation and attention decision mechanism based on a currently planned partial path, completing current path planning and merging the current path to a shared distribution structure when the next-hop node belongs to the initialized inflow node set, and updating the initialized inflow node set; generating a shared distribution tree according to paths of the shared distribution structure, and marking traffic of an upstream link shared by multiple destination nodes as a downstream demand maximum value. In this way, the efficiency of path multiplexing is high, and global transmission and energy consumption cost are low.

A method, apparatus, device, and storage medium for calculating nondeterministic multicast distribution paths are disclosed. The method includes: determining the optimal source node corresponding to each destination node, and grouping destination nodes corresponding to the same optimal source node into the same node group, resulting in multiple node groups; for each node group, generating a multicast distribution tree with the optimal source node associated with the node group as the root, all destination nodes within the node group as leaf nodes, and satisfying preset constraints; merging the generated multicast distribution trees to obtain the multicast distribution path. This application decomposes the complex multi-source multicast problem into multiple relatively simple single-source multicast sub-problems, and then merges the solutions to the single-source multicast sub-problems (i.e., the multicast distribution trees) to construct a nondeterministic multicast distribution path. This achieves automatic and efficient determination of multicast distribution paths under the condition that multiple source nodes exist and simultaneously satisfy preset constraints.

The invention provides a video routing optimization method and device based on a graph neural network. The method comprises the following steps: constructing a network snapshot model; sorting the users according to requirements to obtain a user sequence, and initializing an inflow node set; coding nodes of the undirected weighted graph by adopting a graph neural network to obtain context-aware embedding representation; according to the user sequence and the context-aware embedded representation, performing sequential path planning on each access user, selecting a next-hop node according to path context aggregation and an attention decision-making mechanism based on a currently planned part of paths, and when the next-hop node belongs to an initialized inflow node set, executing the next-hop node selection to the next-hop node when the next-hop node belongs to the initialized inflow node set; current path planning is completed, a current path is combined to the shared distribution structure, and the initialized inflow node set is updated; and generating a shared distribution tree according to the path of the shared distribution structure, and marking the flow of an upstream link shared by the plurality of destination nodes as a downstream demand maximum value. Therefore, the path multiplexing efficiency is high, and the global transmission and energy consumption cost is low.

The invention relates to a method for optimizing the broadcast of multicast content within a communication network implementing a plurality of network slices, the method comprising:-receiving (11), from a terminal, a signaling message representing a request to attach the terminal to a multicast group associated with the broadcast of a multicast stream, the message comprising a multicast address identifying the multicast group, and a network slice identifier representing that the terminal preferentially is used for routing candidate network slices of the multicast stream; -obtaining (12) an identifier of a target network slice for routing the multicast flow based on the candidate network slice identifier and data representative of the current state of the communication network; -connecting (13) the terminal to a distribution tree deployed in the target network slice for distributing the multicast stream.

The invention relates to the technical field of spiking neural networks, and provides a route generation method and device based on a mixed mode. The method comprises the following steps: receiving a routing request of a current pulse event, analyzing an application layer time constraint, and distributing a total time window from a source node to each destination node as a time sub-budget corresponding to each hop in a network forwarding path; generating a routing path meeting a time constraint for the routing request of the current pulse event; predicting a destination node set of a routing request of a next pulse event, and constructing a pre-computation multicast distribution tree based on the predicted destination node set and a time constraint requirement; and when a routing request of a next pulse event is received, if the routing request of the next pulse event is matched with the predicted destination node set, using the pre-computation multicast distribution tree as a routing path so as to save the time overhead of real-time routing computation. The method is used for solving the problem that the real-time performance and the calculation efficiency of event-driven communication are difficult to balance.