86 results about "Fat tree" patented technology

A molecular-wire crossbar interconnect for signal routing and communications between a first level and a second level in a molecular-wire crossbar is provided. The molecular wire crossbar comprises a two-dimensional array of a plurality of nanometer-scale switches. Each switch is reconfigurable and self-assembling and comprises a pair of crossed wires which form a junction where one wire crosses another and at least one connector species connecting the pair of crossed wires in the junction. The connector species comprises a bi-stable molecule. Each level comprises at least one group of switches and each group of switches comprises at least one switch, with each group in the first level connected to all other groups in the second level in an all-to-all configuration to provide a scalable, defect-tolerant, fat-tree networking scheme. The primary advantage is ease of fabrication, because an active switch is formed any time two wires cross. This saves tremendously on circuit area (a factor of a few times ten), since no other wires or ancillary devices are needed to operate the switch or store the required configuration. This reduction of the area of a configuration bit and its switch to just the area of two crossing wires is a major advantage in constructing a defect-tolerant interconnect network.

Multipath load-balancing algorithms, which can be used for data center networks (DCNs), are provided. A multipath load-balancing algorithm can be, for example, a distributed multipath load-balancing algorithm or a centralized multipath load-balancing algorithm. Algorithms of the subject invention can be used for, e.g., hierarchical DCNs and/or fat-tree DCNs. Algorithms of the subject invention are effective and scalable and significantly outperform existing solutions.

The invention belongs to a content data network technology, and discloses a data transmission method of a content-centric datacenter network. The method is characterized in that an NDN (Named Data Network) based content-centric routing and forwarding strategy is used; multicast characteristics and expandability of the CCDN (content-centric datacenter network) are guaranteed only by storing a complete PIT (Pending Interest Table) and part of an FIB (forwarding information base), by adopting a Hybrid Content and Location routing strategy and by taking limited storage resources and datacenter network topology of a datacenter switch into consideration; and different from an on-path caching strategy of the NDN, an off-path mechanism that a host provides data caching is adopted by taking large data volume characteristic of the datacenter network and host storage capacity of redundancy into consideration. According to multistage topological characteristics, of the datacenter network, such as Fat-Tree, the CCDN guarantees Distance-aware Content based Forwarding through FIB learning, average path length of data transmission is decreased, and network throughput is improved. By the aid of the adaptive forwarding strategy, the CCDN can still provide accurate and efficient data forwarding in network failure.

A method for building scalable Ethernet switch network and huge Ethernet switch is provided. The switch network is composed of a plurality of tiers of Ethernet switches forming a fat-tree topology. The bottommost tier connects to the end-stations. All switches except those in the top tier are typical IEEE802.1Q switches configured purposefully such that all packets from the end-stations are forwarded upstream to the top tier and are tagged with ingress port information. The top-tier switches possess a novel learning and forwarding capability such that a received packet is forwarded downstream to the destination end-station through one of a plurality of possible paths selected by a load balancing algorithm. Said method can also be applied to build an Ethernet switch of a large number of ports.

To exchange data between adjacent nodes at high speed while using an existing network including a fat tree and a multistage crossbar switch. This invention provides a parallel computer system including: a plurality of nodes each of which includes a processor and a communication unit; a switch for connecting the plurality of nodes with each other; a first network for connecting each of the plurality of nodes and the switch; and a second network for partially connecting the plurality of nodes with each other. Further, the first network is comprised of one of a fat tree and a multistage crossbar network. Further, the second network partially connects predetermined nodes among the plurality of nodes directly with each other.

The invention discloses a data stream forwarding method facing Fat-Tree data center network architecture, and belongs to the technical field of data center networks. The data stream forwarding method facing the Fat-Tree data center network architecture is characterized in that forwarding of a stream in a network can be dynamically controlled and the route forwarding strategy of the stream can be dynamically adjusted according to the size of the stream in the network by combining a data center network and the SDN technology in the data center network regarding Fat-Tree as network topology. The method has the functions of a timing network link information acquisition module, a data stream information acquisition module, a large data stream detecting and processing module, a load balancing control forwarding module and a large data stream forwarding module. According to the data stream forwarding method facing the Fat-Tree data center network architecture, load balancing of a data stream in the data center network can be achieved, the forwarding strategy can be dynamically adjusted according to the different data stream sizes, the link utilization ratio of the data center network can be improved, the forwarding delay of the data stream can be shortened, and the probability of link jamming is reduced.

The invention discloses a data center network topology system based on module expansion. And therefore, a problem that an expansion capability of tree topology of a data center is limited to the number of ports of a switch device and insufficiency of dynamic selectivity of a downlink can be solved. The implementation steps of system are as follows: an intermediate layer is introduced between a core layer and a convergence layer of a fat tree structure; a small network that is formed by connection of switches and servers of each of the layer is recorded as a basic module, so that a plurality of basic modules are constructed; four additional ports of switches of the core layer are utilized to connect all the basic modules to form a large whole network, so that a demand of construction of a data center network can be met and dynamic selection of a down link can be satisfied as well as a high halving bandwidth is provided; and lateral expanding, vertical expanding or mixing expanding is carried out on the network topology system as well as expansion of the network scale is supported flexibly. According to the invention, the system has a high network fault tolerance performance; more server devices can be accommodated; and a demand of application expansion in the future can be met; therefore, the system can be applied to construction of a data center network and high bandwidth data transmission can be provided.

In a first embodiment of the present invention, a non-blocking switch fabric is provided comprising: a first set of intra-domain switches; a second set of intra-domain switches; a set of inter-domain switches located centrally between the first set of intra-domain switches and the second set of intra-domain switches, wherein each of the ports of each of the inter-domain switches is connected to an intra-domain switch from the first or second set of intra-domain switches.

The invention is applied to the field of network communication, and provides a network topology structure based on a fat tree high scalability hypercube. The network topology structure comprises multiple switches and multiple servers. The switches employ a recursion unit hierarchical mode to form a topology network. A lowest recursion unit employs an m-port n- tree fat tree network structure, and the number of servers supported by the lower recursion unit is: g0=2*(m/2)n, wherein m is the number of ports of the switches in a fat tree, and n is the number of layers of the fat tree structure. In a network topology, the k-th port of the i-th switch at a highest layer is interconnected with the k-th server of the i-th subunit for realizing a multipath network topology structure, wherein i and k belong to a set of {1,2, ..., gk-1}. Through fusing a fat tree network topology in hierarchical recursion, the high scalability of the network topology is realized, at the same time, the characteristics of equal bandwidth and multipath performance of a fat tree network are also inherited, the bandwidth throughput is quite large, the fault tolerance is quite good, and the average time delay is quite small.

Methods and apparatus are provided for routing data packets between source and destination switches in a fat tree network. For each packet, a route is selected having three or less routing phases such that the route follows a shortest path across the network between the source and destination switches. The data packet is transmitted from the source switch to the destination switch, via the route, on one of first and second virtual channels unless the route includes a predetermined one of a down-up turn and an up-down turn. If the route includes the predetermined one of a down-up turn and an up-down turn, the data packet is transmitted, via the route, on the first virtual channel up to the switch 1 at which the turn occurs and on the second virtual channel from that switch. This provides full-connectivity in fat tree networks with deadlock free operation.

The invention falls within the technology of multistage interconnection network such as fat-trees, comprising at least one switch located at a stage (s) and configured to send, through an output port from a number (k) of output ports forming an ordered list, at least a data packet containing a destination address identified by a n-tuple with a plurality (n) of components (p_n−1, . . . , p₁, p₀), and sε{0 . . . (n−1)}. The invention has application for both source and distributed routing, as deterministic and as adaptive routing, selecting an output port to be the unique or the default option to forward the packets at the switch which is the output port that has a position in the ordered list of output ports corresponding to the component (p_s) of the destination address at the position given by the stage (s) of the switch.

The invention discloses a data center network load balancing method based on an SDN and employing fat-tree topological structure. The data center network load balancing method comprises: a network link load can be updated in real time and dynamically, a sub-topology is found out between any two hosts in communication from the whole network topology, and a path lowest in link load is found out from the sub-topology. The route is the communication path of the two hosts. According to the data center network load balancing method, by taking the whole topological link load condition into account, the link load and the state information of a relevant switch are used as the Dijkstra algorithm of a search weight to search for the path lowest in link load. The data center network load balancing method has the advantages that the complexity of the algorithm is relative low, less information is stored at the controller end to ensure the network communication quality, reduce the loss rate of data packets and increase the network throughput, and therefore, the load balance of the network is guaranteed.

The invention discloses an LSA sending method, an apparatus and a system, related to the technical field of communication and used for reducing LSA sending times. The method comprises the following steps: a first router generates a first path table according to level position information thereof, the level position information of a second router and the level position information of at least one third router; the first router sends a first link sate updating LSU message to the second router; the second router receives the first link state updating LSU message from the first router; the secondrouter updates a link state database LSDB thereof according to the first LSA; and the second router sends the first LSA to the router corresponding to an identifier in the first path table. The method, the apparatus and the system provided by the embodiment of the invention are applied to sending of the LSA in leaf-spine topology or fat-tree topology.

When numbers of node in network increases to a certain scale, network topology interconnection techniques between exchangers are produced. Synthesized interconnection mode of fat tree topology and full net topology is utilized in the invention, that is to say fat tree interconnection is utilized for each branch and full net interconnection mode is applied among branches. Based on number of branch, two branches fat trees, three branches fat trees and four branches fat trees are named. Comparing with network topology structure of nonblocking fat tree under condition of the same number of exchnger, the invented structure possesses advantages of available large scale of networked scale, lowered communication delay, and higher rate of performances/cost.

The invention discloses a detecting method and device of the error connection in a data center network. The method includes the steps of S1, constructing a data center network topology according to a Fat-Tree network; S2, configuring a centralized controller according to multiple servers and multiple switchers; S3, collecting Fat-Tree network topology information through the communication of the multiple servers and the multiple switchers with respective adjacent equipment; S4, transmitting the Fat-Tree network topology information to the centralized controller to generate whole-network Fat-Tree network topology information, and performing error connection detecting and restoring calculation according to the whole-network Fat-Tree network topology information. The method has the advantages that error connection detecting and restoring calculation are performed by constructing the data center network topology, simple and fast detecting is achieved, artificial judgment is not needed, automation is achieved, and work efficiency is increased.

The invention discloses a fat tree type network-on-chip mapping method based on a differential evolution and predatory search strategy, which comprises the following steps of: (1) initializing the current optimal mapping result and defining a plurality of peripheral solutions which use any one solution as a center to form a limit array in the solution space; (2) setting limit arrays R[0], R[1],..., R[T-1] with the limit total amount of T at the periphery of the current optimal mapping result and defining the current limit variable to be R[i]; (3) searching the periphery of the current limit variable R[i] by adopting a differential evolution method; if a better solution is found out, updating the current optimal mapping result and turning back to the step (2); otherwise, turning to the step (4); (4) updating the current limit variable i which is equal to i plus 1; if i is smaller than T-1, turning back to the step (3); and otherwise, outputting the current optimal mapping result. By using the method, the problem of local optimum is solved through adjustment of limitation, the network energy consumption is greatly reduced, the mapping running time is reduced and the mapping of a low-energy consumption and rapid large-scale IP (Internet Protocol) core in the fat tree type network on a chip can be realized.

The invention discloses a dynamic fault tolerance method and a system based on a fat tree structure. The method comprises the following steps: configuring a fat tree network; leading a switch in the fat tree network to periodically check the states of ports, and sending a link failure message to another switch when the certain port is found to fail; and realizing a dynamic fault tolerance routing according to the link failure information.

Deadlock-free routing of data packets between source and destination switches in a fat tree network is provided. For each packet, a route is selected having three or less routing phases such that the route follows a shortest path across the network between the source and destination switches. The data packet is transmitted from the source switch to the destination switch, via the route, on one of first and second virtual channels unless the route includes a predetermined one of a down-up turn and an up-down turn (see e.g. figures 7 and 8). If the route includes the predetermined turn, the data packet is transmitted, via the route, on the first virtual channel up to the switch at which the turn occurs and on the second virtual channel from that switch. Direct fat tree networks, in which end nodes are connected to switches in a plurality of levels of the fat tree topology, are also provided.

The invention provides a data center network routing method capable of tolerating error connection. The method comprises the following steps that a Fat-Tree data center network topology structure is constructed; a centralized controller is configured, and the centralized controller is communicated with each server or interchanger in the Fat-Tree data center network topology structure; physical network mapping and logical network mapping are performed on the Fat-Tree data center network topology structure, and physical topology information is obtained; the centralized controller collects the list information of edge layer interchangers at which each interchanger can arrive through respective ports, and a corresponding objective edge layer interchanger list is the generated for each interchanger; the centralized controller calculates and installs corresponding routing table entries for arriving at each edge layer interchanger for each interchanger according to the generated objective edge layer interchanger lists. The data center network routing method capable of tolerating the error connection has the double advantages of efficiently utilizing network resources and improving the network throughput.

The invention discloses a blended data center networking and routing method based on visible light communication. The method comprises steps of establishing or keeping wired Fat-Tree structures among multiple racks in a wired data central network system; networking the racks in the Fat-Tree structures into wireless 2D-Torus structures with multiple pod layers; and calculating routing a Path h between any rack pairs according to the pod layers of the 2D-Torus structures and the wired Fat-Tree structures. According to the invention, by coupling networked wireless 2D-Torus structures with the wired Fat-Tree rack structures and using the mixed structure as a whole to work, on the premise that existing devices and layouts of a wired data center are not changed, a cross-rack wireless connection without any control is established, so a wired data center is greatly expanded in low cost and flexibility of the network is improved.