In High Frequency Trading, Every Nanosecond Counts!
In the financial world of high frequency trading, high-performance computers are duking it out in real time to score on instantaneous profit opportunities that appear and vanish in the blink of an eye. According to Bob Laliberte, Senior Analyst at ESG Global, "It’s critical for HFT organizations to eliminate potential network latencies from the start, by employing advanced, next-generation network platforms that leverage intent-based logic, ultra-low latency (ULL), high availability, and ease of management." Networking giants like Cisco are very aware of this need, and that knowledge fueled their 2020 acquisition of Exablaze, a company specializing in the design and manufacture of devices that can deliver ultra-low latency network performance. "Clearly, this technology infusion will enable a next-generation platform and enhance Cisco’s solid domain expertise in the financial sector", said Laliberte.
Delivering near "instantaneous" trading will require a next generation physical network designed to deliver highly predictable end-to-end bandwidth featuring ultra-low latency. "Zero-hop" network designs such as AcceleRoute can achieve this through a bufferless architecture that delivers ultra-low latency approaching that of direct links while eliminating congestion in the network core.
·
1 min
The Metaverse will change everything!
It will change how we interact. How we consume information. How we have fun. What devices we use. And underlying all of that is networking. And that will change too. "IT infrastructure that powers the Internet will need major upgrades to bring the Metaverse from theory to practice, according to Raja Koduri, SVP and GM of Accelerated Computing Systems and Graphics at Intel. “(The metaverse will) need several orders of magnitude more powerful computing capability, accessible at much lower latencies across a multitude of device form factors. To enable these capabilities at scale, the entire plumbing of the internet will need major upgrades.”
Metaverse quality of experience will be intimately tied to networking performance, and latency will be key. Delivering an "instantaneous" Metaverse experience will require a next generation physical network that is designed to deliver highly predictable end-to-end bandwidth with ultra-low latency. "Zero-hop" network designs such as AcceleRoute can achieve this through a bufferless architecture that delivers latency approaching that of direct links while eliminating congestion in the network core.
For more information visit the AcceleRoute webpage at www.InventionShare.com
·
2 min
Taking SD-WAN QoS to the Next Level
Backhauling traffic to a corporate data center has been a defacto approach for ensuring full security treatment across all users and applications. But QoS took a hit when services started moving to the cloud. Michael Cooney, Senior Editor at Network World describes how SD-WANs have provided a solution for that in his article entitled "Fannie Mae’s journey to SD-WAN means less reliance on MPLS and VPNs." In that article, Ken Reddick, Director of Network Engineering at Fannie Mae says “What we are moving to is a cloud-edge environment where user traffic is now sent directly where it needs to go without hitting the data center, and what that has brought us is a four-fold increase in network performance and cut latency by 50%.”
SD-WANs are providing a valuable new approach for delivering optimal connectivity between end-users and cloud services, however QoS is still ultimately determined by the underlying physical networks. To take QoS to the next level will require a next generation physical network that is designed to deliver highly predictable end-to-end bandwidth and ultra-low latency. Network designs such as AcceleRoute achieve this through a bufferless architecture that eliminates congestion in the network core. Low latency bandwidth can be dynamically scaled up or down in real-time based on traffic load. Networks such as AcceleRoute provide an ideal underlay network option for SD-WANs by delivering consistently superior service levels regardless of traffic and geography
For more information about AcceleRoute, contact:
Lesley Gent
Director Client Relations, InventionShare™
lgent@InventionShare.com
(613) 225-7236, Ext 131
Or visit our website at www.InventionShare.com
Show More +
Multimedia
Social
Biography
Maged's current research is directed towards an ultra-high-capacity contiguous switching system, employing fast optical switches, which may serve as a large-scale data center and/or a network of global coverage. A network based on a contiguous switching system scales to interconnect hundreds of thousands of access routers through globally distributed bufferless (preferably optical) connectors with a throughput of the order of an exabit per second (million terabits per second), which is several orders of magnitude higher than the throughput of the current global Internet. A bufferless connector may be configured as a fast (low latency) optical switch, a fast optical spectral-temporal connector, or an electronic switch.
Other areas of interest include adaptive encoding and transcoding of panoramic video-signal streams, self-configuring narrow-beam flexible wireless networks, and optimal algorithms for analysis of massive social-network data.
Areas of Expertise
Switch/Router Design
Coherent Global Networks
Large-Scale Data Centers
Telecommuncations
Electrical Engineering
Network Architecture Design
Education
McMaster University
Ph.D.
Electrical Engineering
1972
Affiliations
Member of the Association of Professional Engineers of Ontario, Canada
Institute of Electrical and Electronics Engineers (IEEE)
A large-scale contiguous network comprises access nodes arranged into access groups and distributors arranged into constellations of collocated distributors. The distributors may comprise switches, rotators, or a mixture of switches and rotators. Each access group connects to each distributor of a respective set of distributors selected so that each pair of access groups connects once to a respective distributor. At least one access group comprises a global controller. Each access node has a dual multichannel link to each constellation of a respective set of constellations, the link carrying a set of dual channels connecting through a spectral demultiplexer and a spectral multiplexer to a subset of distributors. Each access node is equipped with a respective access controller having a memory device storing identifiers of dual paths to all other access nodes and the global controller, each path traversing only one distributor.
A large-scale switching system deployed as a global network or a large-scale data center includes a large number of access nodes (edge nodes) interconnected through optical or electronic rotators. The rotators are logically arranged in a matrix and each access node has a channel to each rotator in a respective row and a channel from each rotator of a respective column of the matrix. A dual timing circuit coupled to a diagonal rotator pair exchanges timing data with edge nodes connecting to the diagonal rotator pair to facilitate temporal alignment of data received at input ports of each rotator. Each access node has a path to each other access node traversing only one of the rotators. The rotators may be arranged into constellations of collocated rotators to facilitate connectivity of access nodes to rotators using wavelength-division-multiplexed links.
Spectral-temporal connector for full-mesh networking
US10003865
2018-06-19
A spectral-temporal connector interconnects a large number of nodes in a full-mesh structure. Each node connects to the spectral-temporal connector through a dual link. Signals occupying multiple spectral bands carried by a link from a node are de-multiplexed into separate spectral bands individually directed to different connector modules. Each connector module has a set temporal rotators and a set of spectral multiplexers. A temporal rotator cyclically distributes segments of each signal at each inlet of the rotator to each outlet of the rotator. Each spectral multiplexer combines signals occupying different spectral bands at outlets of the set of temporal rotators onto a respective output link. Several arrangements for time-aligning all the nodes to the connector modules are disclosed.
Large-scale data center based on a contiguous network
US10027510
2018-07-17
A large-scale data center is based on orthogonal connectivity of a distributed single-stage connector, having a large number of disjoint primary switches, to a large number of access nodes and a large number of secondary switches. The secondary switches are coupled to a large number of servers forming a server farm. The orthogonal-connectivity scheme yields a contiguous network providing single-hop paths from each access node to each other access node, from each secondary switch to each other secondary switch, from each access node to each secondary switch, and from each secondary switch to each access node.
A network comprises ingress nodes, egress nodes, primary switches, and secondary switches where any pair of an ingress node and an egress node connects to orthogonal sets of primary switches and each secondary switch connects a respective set of primary switches to an orthogonal set of primary switches. Thus, each ingress node has a primary path and numerous compound paths to each egress node. The primary path traverses a respective primary switch, and each compound path traverses a first primary switch, a secondary switch, and a second primary switch. The disclosed connectivity pattern enables network scalability to accommodate hundreds of thousands of ingress-egress node pairs while permitting a significant proportion of incoming data to be routed through the primary switches avoiding the secondary switches.
A large-scale switching system configured as a global network or a large-scale data center employs switches arranged in a matrix having multiple rows and multiple columns. The switching system supports a large number of access nodes (edge nodes). Each access node has a channel to each switch in a respective row and a channel from each switch of a respective column. Thus, an access node connects to input ports of a set of switches and output ports of a different set of switches. Each access node has a path to each other access node traversing only one of the switches. Controllers of switches of each diagonal pair of switches are integrated or mutually coupled to provide a return control path for each access node. The switches may be arranged into constellations of collocated switches to facilitate edge-node access to switches using wavelength-division-multiplexed links. The switches are preferably fast optical switches.
Distributed routing control in a vast communication network
US9762479
2017-09-11
Multiple network controllers are interconnected in a full mesh structure, e.g., through a cyclical cross connector, to form a distributed control system for a network of a large number of nodes. A network controller acquires characterizing information of links emanating from a respective set of nodes, communicates the information to each other network controller, and determines a route set from each node of the respective set of nodes to each other node of the network. The network controller may determine, for each link included in the route set, identifiers of specific route sets which traverse the link. Accordingly, a state-change of any link in the network can be expeditiously communicated to network controllers to take corrective actions where necessary. A network controller may rank routes of a route set according to some criterion to facilitate selection of a favourable available route for a connection.
Distributed determination of routes in a vast communication network
US10021025
2016-05-12
A method and a system for distributed computation of a routing table for a vast communication network are disclosed. The network nodes are arranged into multiple groups with each group associated with a respective network controller. A network controller of a group acquires characterizing information of links emanating from local nodes of the group, communicates the information to each other network controller, reciprocally receives characterizing information from other network controllers, and determines a generic route set from each local node to each other node of the network. The network controllers collectively determine an inverse routing table identifying all routes traversing each individual link in the entire network and exchange node or link state-transition information for updating individual route sets affected by any state transition. Thus, the processing effort of routes generation and tracking network-elements states is distributed among multiple coordinated network controllers.
An optical spectral-temporal connector, having multiple connector modules, interconnects a large number of nodes in a full-mesh structure. A wavelength-division-multiplexed link from each node is de-multiplexed into wavelength channels individually directed to different connector modules. Each connector module has a set of star couplers, each star coupler connecting to wavelength channels from a respective set of nodes through spectral translators. Each spectral translator cyclically shifts a spectral band of a wavelength channel so that, at any instant of time, spectral bands of signals at inlets of any star coupler are disjoint. A spectral router connects outlets of the set of star couplers to a respective set of nodes. A spectral-translation controller prompts each spectral translator to shift to a new spectral band. Several arrangements for time-aligning all the nodes to the connector modules are disclosed.
Spectral temporal connector for full-mesh networking
US9647792
2017-05-09
A spectral-temporal connector interconnects a large number of nodes in a full-mesh structure. Each node connects to the spectral-temporal connector through a dual link. Signals occupying multiple spectral bands carried by a link from a node are de-multiplexed into separate spectral bands individually directed to different connector modules. Each connector module has a set temporal rotators and a set of spectral multiplexers. A temporal rotator cyclically distributes segments of each signal at each inlet of the rotator to each outlet of the rotator. Each spectral multiplexer combines signals occupying different spectral bands at outlets of the set of temporal rotators onto a respective output link. Several arrangements for time-aligning all the nodes to the connector modules are disclosed.
Multiple petabit-per-second switching system based on orthogonal formations of sets of access nodes
US9860132
2018-01-02
A switching system having a number of nodes interfacing with external network elements and interconnected through independent switches is disclosed. The network may serve as a large-scale data center or a geographically distributed network. The nodes are arranged into a number of formations. Within each formation, the nodes are divided into a number of disjoint sets of nodes. The nodes of each set of nodes are interconnected through a respective switch and are selected so that each set of nodes of any formation is orthogonal to each set of nodes of each other formation. With such a structure, each node has a set of routes to each other node, each route of which traversing at most two switches. The switching system may grow in both capacity and coverage without disturbing an already installed configuration. A switching system of an access capacity of multiple petabits/sec and large coverage is thus realizable.
Multiple petabit-per-second switching system employing latent switches
US9219697
2015-12-22
Access switches of moderate dimensions are interconnected through central switches of large dimensions to form a large-scale switching system. The central switches are configured as latent switches which scale easily to large dimensions. Each access switch has asymmetric connections to the ingress sides and egress sides of the central switches so that paths from an originating access switch to a destination access switch through the central switches are subject to staggered switching delays permitting an access controller of any access switch to select an available path of minimum switching delay for a given flow. Using access switches of 128 dual ports each and central switches of 4096 dual ports each, a switching system of 524288 dual ports is realized. At a port capacity of 10 Gigabits/second, the access capacity exceeds five petabits per second and the bulk of traffic experiences a switching delay below two microseconds.
Petabits-per-second packet switch employing cyclically interconnected switch units
US9634960
2017-04-25
A packet switching system of an access capacity scalable to multiple petabits per second is disclosed. A multiplicity of switch units, each of a relatively small dimension, is organized into orthogonal sets of switch units and the switch units of each set are cyclically interconnected through a respective dual rotator. Each switch unit has a contention-free switching mechanism and is coupled to a same number of dual rotators. Each switch unit has a switch-unit controller configured to route data received from external data sources to external data sinks coupled to any other switch unit by communicating with at most one switch-unit controller of an intermediate switch unit.
Network with a fast-switching optical core providing widely varying flow-rate allocations
US9565487
2017-02-07
Multiple switch planes, each having meshed bufferless switch units, connect source nodes to sink nodes to form a communications network. Each directed pair of source and sink nodes has a first-order path traversing a single switch unit in a corresponding switch plane and multiple second-order paths each traversing two switch units in one of the remaining switch planes. To reduce processing effort and minimize requisite switching hardware, connectivity patterns of source nodes and sink nodes to the switch planes are selected so that each pair of source node and sink node connects only once to a common switch unit. Widely-varying flow rates may be allocated from each source node to the sink nodes. To handle frequent changes of flow-rate allocations, in order to follow variations of traffic distribution, a high-throughput scheduling system employing coordinated multiple scheduler units is provided in each switch plane.
A packet switch that scales gracefully from a capacity of a fraction of a terabit per second to thousands of terabits per second is disclosed. The packet switch comprises edge nodes interconnected by independent switch units. The switch units are arranged in a matrix having multiple rows and multiple columns and may comprise instantaneous or latent space switches. Each edge node has a channel to a switch unit in each column and a channel from each switch unit in a selected column. A simple path traversing only one of the switch units may be established from each edge node to each other edge node. Where needed, a compound path comprising at most two simple paths may be established for any edge-node pair. In a preferred configuration, the switch units connect at input to orthogonal sets of edge nodes. A distributed control system expedites connection-request processing.
Switching system employing independent data switches connecting orthogonal sets of nodes
US9054979
2015-06-09
A switching system formed of a number of nodes interfacing with external network elements and interconnected through a number of independent switches is disclosed. The switches are arranged into a set of primary switches, a set of secondary switches, and a set of tertiary switches and each node connects to a respective primary switch, a respective secondary switch, and a respective tertiary switch. The connection pattern of nodes to switches is selected so that any set of nodes connecting to any primary switch, any set of nodes connecting to any secondary switch, and any set of nodes connecting to any tertiary switch are mutually orthogonal. A distributed control system sets a path from any node to any other node traversing at most two switches. The switching system may serve as a large-scale data-switching center or a geographically distributed network.
Packet-switching node with inner flow equalization
US9148370
2015-09-29
Independent switches arranged into multiple switch planes interconnect nodes coupled to data sources and sinks to form a switching node which scales gracefully from a capacity of a fraction of a terabit per second to hundreds of terabits per second. The switches of each switch plane are arranged in a matrix. Each node connects to an inlet of a selected switch in each column and an outlet of a selected switch in each row in each switch plane. A route set for each directed node pair includes simple paths, each traversing one switch, and compound paths, each traversing two switches. The connectivity of nodes to switches ensures that each switch may be selected to handle data flow of any directed node pair and that all simple paths leading to any node traverse switches which receive data from mutually orthogonal sets of nodes. This feature equalizes flow rates through the switches.
Time-coherent global network employing spectral routers
US9596524
2017-03-14
A network of global coverage, scalable to an access capacity of hundreds of petabits per second, is configured as independent bufferless switches with spectral routers connecting edge nodes to the switches. The switches are logically arranged in at least one matrix, the spectral routers are logically arranged into a matrix of upstream spectral routers and a matrix of downstream spectral routers. Each edge node has a link to an upstream spectral router in each column of the matrix of upstream spectral routers and a link from a downstream spectral router in each row of the matrix of downstream spectral routers. Preferably, all sets of edge nodes connecting to the upstream spectral routers are selected to be mutually orthogonal. Each switch is coupled to a respective switch controller and a respective time indicator. Each switch controller entrains time indicators of a set of subtending edge nodes to enable coherent switching.
High-capacity data switch employing contention-free switch modules
US8774176
2014-07-08
A scalable router-switch that grows from a capacity of a few gigabits per second to hundreds of terabits per second is disclosed. In one embodiment, the router-switch comprises a plurality of switch units arranged in a plurality of combinations. Within each combination, each switch unit cyclically connects to each other switch unit to form a contention-free temporal mesh. Each switch unit belongs to a number of combinations and any two combinations have at most one switch unit in common. The router-switch further includes a distributed-control system which comprises an outer controller associated with each of the switch units and an inner controller associated with each combination. The structural simplicity significantly simplifies the operation and control of the router-switch.
Latent space switch using a single transposing rotator
US8971340
2015-03-03
A single transposing rotator successively connects a set of access ports to a set of memory devices and the set of memory devices to the set of access ports. A set of inlet selectors connecting to rotator inlets and a set of outlet selectors connecting to rotator outlets are coordinated to concurrently connect the access ports to the memory devices through the rotator, and concurrently connect the memory devices to the access ports. Each memory device connects to an inlet selector and a corresponding peer outlet selector. Multiple temporal multiplexers submit upstream control messages from the access ports to a multi-port master controller. Multiple temporal demultiplexers distribute downstream control messages sent from the master controller to the access ports. Alternatively, the multi-port master controller may connect to selected inlet selectors and corresponding peer outlet selectors for successively receiving upstream control messages and sending downstream control messages.
Single-rotator latent space switch with an embedded controller
US9252909
2016-02-02
A single rotator successively connects a set of access ports to a set of memory devices and a multi-port controller and connects the set of memory devices and the multi-port controller to the set of access ports. The rotator has a set of inlets and a set of outlets and cyclically connects each inlet to each outlet during a rotation cycle. A set of inlet selectors connecting to the inlets of the rotator and a set of outlet selectors connecting to the outlets of the rotator are coordinated to concurrently connect the access ports to the memory devices and to the master controller through the rotator, and concurrently connect the memory devices and the master controller to the access ports. Each memory device connects to an inlet selector and a corresponding transposed outlet selector.
Single-rotator latent space switch with an external controller
US9154255
2015-10-06
A latent space switch based on a single rotator and an array of memory devices is disclosed. The switch interfaces with external nodes through a set of access ports. The rotator has a set of inlets and a set of outlets with each inlet connecting to each outlet during a time frame organized into time slots. During each time slot, an inlet alternately connects to an access port and a memory device while a transposed outlet of the inlet alternately connects to the same memory device and another access port. Multiple temporal multiplexers submit upstream control messages from the access ports to a multi-port master controller. Multiple temporal demultiplexers distribute downstream control messages sent from the master controller to the access ports.
Network with a fast-switching optical core providing widely varying flow-rate allocations
US8774200
2015-07-08
Multiple switch planes, each having meshed bufferless switch units, connect source nodes to sink nodes to form a communications network. Each directed pair of source and sink nodes has a first-order path traversing a single switch unit in a corresponding switch plane and multiple second-order paths each traversing two switch units in one of the remaining switch planes. To reduce processing effort and minimize requisite switching hardware, connectivity patterns of source nodes and sink nodes to the switch planes are selected so that each pair of source node and sink node connects only once to a common switch unit. Widely-varying flow rates may be allocated from each source node to the sink nodes. To handle frequent changes of flow-rate allocations, in order to follow variations of traffic distribution, a high-throughput scheduling system employing coordinated multiple scheduler units is provided in each switch plane.
A switching node comprises edge nodes interconnected by independent switch units. The switch units are arranged in at least one switch plane and the switch units of each switch plane are arranged in a matrix having several rows and several columns. Each edge node has a channel to a switch unit in each column in each switch plane and a channel from each switch unit in a selected column in each switch plane. Simple paths, each traversing only one switch unit in a switch plane, may be established for any directed edge-node pair. Additionally, several non-intersecting compound paths, each comprising at most two simple paths, may be established for any edge-node pair. A significant proportion of traffic may be routed through simple paths. The switching node employs distributed control scheme and scales gracefully from a capacity of a fraction of a terabit per second to thousands of terabits per second.
A packet switch that scales gracefully from a capacity of a fraction of a terabit per second to thousands of terabits per second is disclosed. The packet switch comprises edge nodes interconnected by independent switch units. The switch units are arranged in a matrix having multiple rows and multiple columns and may comprise instantaneous or latent space switches. Each edge node has a channel to a switch unit in each column and a channel from each switch unit in a selected column. A simple path traversing only one of the switch units may be established from each edge node to each other edge node. Where needed, a compound path comprising at most two simple paths may be established for any edge-node pair. In a preferred configuration, the switch units connect at input to orthogonal sets of edge nodes. A distributed control system expedites connection-request processing.
Cascaded contention-free switch modules with interleaved consolidation units
US8576839
2013-11-05
A scalable router-switch comprises a plurality of switch units each having consolidation means for data disassembling and reassembling. The switch units are arranged into switch modules and the switch units of each switch module are interconnected through a dual rotator to form a contention-free temporal mesh.
Switch elements, each receiving data from external sources and transmitting data to external sinks, are interconnected through a single rotator to form a switching node. The single rotator has a number of inlets equal to the number of switch elements and a number of outlets equal to the number of switch elements. A first set of channels connects the switch elements to inlets of the rotator and a second set of channels connects the outlets of the rotator to the switch elements. The connectivity pattern of the second set of channels is a transposition of the connectivity pattern of the first set of channels in order to preserve sequential data order of switched data. A controller communicatively coupled to the switch elements exchanges timing data with external nodes of a time-coherent network and schedules data transfer among the switch elements.
A network of global coverage, scalable to hundreds of petabits per second, comprises bufferless switch units each of dimension n×n, n>1, arranged in a matrix of ν columns and ν rows, ν>1, interconnecting a maximum of ν×n edge nodes. Each edge node has ν upstream channels to ν switch units in ν different columns and ν downstream channels from ν switch units in ν different rows. All upstream channels to a switch unit are time-locked to the switch unit, thus enabling coherent switching at the switch unit. The switch units are preferably fast-switching optical nodes. Alternatively, the switch units may comprise fast-switching optical nodes each of dimension m×m, arranged in a first μ×μ matrix, and latent space switches each of dimension n×n, n>1, arranged in a second ν×ν matrix, ν>1, where μ×m=ν×n. An edge node time locks to each optical node and each latent space switch to which it connects.
A switching node comprises edge nodes interconnected by independent switch units. The switch units are arranged in at least one switch plane and the switch units of each switch plane are arranged in a matrix having several rows and several columns. Each edge node has a channel to a switch unit in each column in each switch plane and a channel from each switch unit in a selected column in each switch plane. Simple paths, each traversing only one switch unit in a switch plane, may be established for any directed edge-node pair. Additionally, several non-intersecting compound paths, each comprising at most two simple paths, may be established for any edge-node pair. A significant proportion of traffic may be routed through simple paths. The switching node employs distributed control scheme and scales gracefully from a capacity of a fraction of a terabit per second to thousands of terabits per second.
A packet switch that scales gracefully from a capacity of a fraction of a terabit per second to thousands of terabits per second has edge nodes interconnected by independent switch units. The switch units are arranged in a matrix having multiple rows and multiple columns. A switch unit is implemented as an instantaneous space switch or as a latent space switch. Each edge node has a channel to a switch unit in each column and a channel from each switch unit in a selected column. A simple path traversing only one of the switch units may be established from each edge node to each other edge node. Where needed, a compound path concatenating at most two simple paths may be established for any edge-node pair. In a preferred configuration, the switch units connect at input to orthogonal sets of edge nodes. A distributed control system expedites connection-request processing.
A high capacity network comprises a plurality of edge nodes with asymmetrical connections to a plurality of switch planes, each switch plane comprising fully meshed fast-switching optical switch units. Upstream wavelength channels from each source edge node connect to different switch planes in a manner which ensures that upstream wavelength channels from any two edge nodes connect to a common switch unit in at most a predefined number, preferably one, of switch planes. Thus, switch units in different switch planes connect to upstream channels from orthogonal subsets of source edge nodes. In contrast, downstream wavelength channels from a switch unit in each switch plane connect to one set of sink edge nodes. In an alternate arrangement, the upstream and downstream asymmetry may be reversed.
A scalable router-switch comprises a plurality of switch units each having consolidation means for data disassembling and reassembling. The switch units are arranged into switch modules and the switch units of each switch module are interconnected through a dual rotator to form a contention-free temporal mesh.
High-capacity data switch employing contention-free switch modules
US8223759
2012-07-17
A scalable router-switch that grows from a capacity of a few gigabits per second to hundreds of terabits per second is disclosed. In one embodiment, the router-switch comprises a plurality of switch units arranged in a plurality of combinations. Within each combination, each switch unit cyclically connects to each other switch unit to form a contention-free temporal mesh. Each switch unit belongs to a number of combinations and any two combinations have at most one switch unit in common. The router-switch further includes a distributed-control system which comprises an outer controller associated with each of the switch units and an inner controller associated with each combination. The structural simplicity significantly simplifies the operation and control of the router-switch.
A network comprises ingress nodes, egress nodes, primary switches, and secondary switches where any pair of an ingress node and an egress node connects to orthogonal sets of primary switches and each secondary switch connects a respective set of primary switches to an orthogonal set of primary switches. Thus, each ingress node has a primary path and numerous compound paths to each egress node. The primary path traverses a respective primary switch, and each compound path traverses a first primary switch, a secondary switch, and a second primary switch. The disclosed connectivity pattern enables network scalability to accommodate hundreds of thousands of ingress-egress node pairs while permitting a significant proportion of incoming data to be routed through the primary switches avoiding the secondary switches.
A network of global coverage, scalable to an access capacity of hundreds of petabits per second, is configured as independent space switches with spectral routers connecting edge nodes to the space switches. The space switches are preferably configured as fast optical switches. Each edge node has a link to an upstream spectral router and a link from a downstream spectral router. Each switch is coupled to a respective switch controller and a respective time indicator. Each switch controller entrains edge controllers of a set of subtending edge nodes to enable coherent switching.
Multiple petabits-per-second switching system employing latent switches
CA2915680
2018-09-18
Access switches of moderate dimensions are interconnected through central switches of large dimensions to form a large-scale switching system. The central switches are configured as latent switches which scale easily to large dimensions. Each access switch has asymmetric connections to the ingress sides and egress sides of the central switches so that paths from an originating access switch to a destination access switch through the central switches are subject to staggered switching delays permitting an access controller of any access switch to select an available path of minimum switching delay for a given flow. Using access switches of 128 dual ports each and central switches of 4096 dual ports each, a switching system of 524288 dual ports is realized. At a port capacity of 10 Gigabits/second, the access capacity exceeds five petabits per second and the bulk of traffic experiences a switching delay below two microseconds.
A large-scale switching system configured as a global network or a large-scale data center employs switches arranged in a matrix having multiple rows and multiple columns. The switching system supports a large number of access nodes (edge nodes). Each access node has a channel to each switch in a respective row and a channel from each switch of a respective column. Thus, an access node connects to input ports of a set of switches and output ports of a different set of switches. Each access node has a path to each other access node traversing only one of the switches. Controllers of switches of each diagonal pair of switches are integrated or mutually coupled to provide a return control path for each access node. The switches may be arranged into constellations of collocated switches to facilitate edge-node access to switches using wavelength-division-multiplexed links. The switches are preferably fast optical switches.
Distributed determination of routes in a vast communication network
CA2911730
2017-05-09
A method and a system for distributed computation of a routing table for a vast communication network are disclosed. The network nodes are arranged into multiple groups with each group associated with a respective network controller. A network controller of a group acquires characterizing information of links emanating from local nodes of the group, communicates the information to each other network controller, reciprocally receives characterizing information from other network controllers, and determines a generic route set from each local node to each other node of the network. The network controllers collectively determine an inverse routing table identifying all routes traversing each individual link in the entire network and exchange node or link state-transition information for updating individual route sets affected by any state transition. Thus, the processing effort of routes generation and tracking network-elements states is distributed among multiple coordinated network controllers.
Distributed routing control in a vast communication network
CA2911622
2017-05-09
Multiple network controllers are interconnected in a full mesh structure, e.g., through a cyclical cross connector, to form a distributed control system for a network of a large number of nodes. A network controller acquires characterizing information of links emanating from a respective set of nodes, communicates the information to each other network controller, and determines a route set from each node of the respective set of nodes to each other node of the network. The network controller may determine, for each link included in the route set, identifiers of specific route sets which traverse the link. Accordingly, a state-change of any link in the network can be expeditiously communicated to network controllers to take corrective actions where necessary. A network controller may rank routes of a route set according to some criterion to facilitate selection of a favourable available route for a connection.
An optical spectral-temporal connector, having multiple connector modules, interconnects a large number of nodes in a full-mesh structure. A wavelength- division-multiplexed link from each node is de-multiplexed into wavelength channels individually directed to different connector modules. Each connector module has a set of star couplers, each star coupler connecting to wavelength channels from a respective set of nodes through spectral translators. Each spectral translator cyclically shifts a spectral band of a wavelength channel so that, at any instant of time, spectral bands of signals at inlets of any star coupler are disjoint. A spectral router connects outlets of the set of star couplers to a respective set of nodes. A spectral-translation controller prompts each spectral translator to shift to a new spectral band. Several arrangements for time-aligning all the nodes to the connector modules are disclosed.
Spectral-temporal connector for full-mesh networking
CA2894730
2019-01-08
A spectral-temporal connector interconnects a large number of nodes in a full-mesh structure. Each node connects to the spectral-temporal connector through a dual link. Signals occupying multiple spectral bands carried by a link from a node are de-multiplexed into separate spectral bands individually directed to different connector modules. Each connector module has a set temporal rotators and a set of spectral multiplexers. A temporal rotator cyclically distributes segments of each signal at each inlet of the rotator to each outlet of the rotator. Each spectral multiplexer combines signals occupying different spectral bands at outlets of the set of temporal rotators onto a respective output link. Several arrangements for time-aligning all the nodes to the connector modules are disclosed.
