Abstract: Embodiments provide mechanisms to facilitate compute operations for deep neural networks. One embodiment comprises a graphics processing unit comprising one or more multiprocessors, at least one of the one or more multiprocessors including a register file to store a plurality of different types of operands and a plurality of processing cores. The plurality of processing cores includes a first set of processing cores of a first type and a second set of processing cores of a second type. The first set of processing cores are associated with a first memory channel and the second set of processing cores are associated with a second memory channel.

Abstract: An automated networking device replacement system includes a chassis housing in which management module and a fabric services subsystem are provided. The fabric services subsystem identifies that a first networking device has been provided in the chassis, and determines that the first networking device is a replacement for a second networking device. In response, the fabric services subsystem automatically identifies at least one fabric-services-based configuration that was performed on the second networking device, and automatically performs the at least one fabric-services-based configuration on the first networking device. The fabric services subsystem then transmits a management-module-based configuration instruction to the management module that causes the management module to automatically perform at least one management-module-based configuration on the first networking device that was performed on the second networking device.

Abstract: A device may receive first topology information from a first network device of a network, and may receive second topology information from a second network device of the network. The device may assign a first BGP-LS identifier to the first network device, and may associate the first topology information with the first BGP-LS identifier. The device may assign a second BGP-LS identifier to the second network device, and may associate the second topology information with the second BGP-LS identifier. The device may store the first topology information, as a first route, based on the first BGP-LS identifier, and may store the second topology information, as a second route, based on the second BGP-LS identifier. The device may select the first route or the second route as a primary route, and may utilize the primary route to control routing of traffic through the network.

Abstract: In a network fabric, such as a network having CLOS topology, non-volatile memory express (NVMe®) endpoints may be connected to one of centralized discovery controller (CDC) distributed services placed among multiple leaf switches, which may have resource constraints. Connection scale, delay, and jitter may occur if the CDC distributed services are not placed on leaf switches close to NVMe endpoints served by the CDC distributed services. System and method embodiments are disclosed for placement of CDC services on a switching network fabric close to endpoints that are served by the CDC services. The placement of CDC services may be implemented via push registration, pull registration, and/or manual DDC adding/registration such that the CDC services may be placed on desired leaf switches close to endpoints.

Abstract: System and method embodiments for placement of Centralized Discovery Controller (CDC) services on a switching network fabric are disclosed. Nodes on the switching network fabric are categorized into leader nodes, e.g., spine switches, and follower nodes, e.g., leaf switches. The CDC services are divided into centralized CDC services and distributed CDC services, which has direct connections to endpoints for seamless horizontal scaling of endpoints. The centralized CDC services are placed on a leader node and the distributed CDC services are placed among the follower nodes. Operations of the CDC services are monitored. In case of a node failure, CDC services deployed on the failed node are moved or transferred to one or more other nodes that have the same role as the failed node. Implementation of the CDC service placement embodiments enables seamless horizontal scaling for endpoints and provides capability for more visibility and monitoring of endpoint.

Abstract: Techniques are disclosed for disseminating network service-specific mapping information across administrative domains. In one example, a network device receives an indication of a route target and one or more underlay tunnels configured to support a service route. The service route is configured to transport network traffic associated with a first network service of a plurality of network services. The network device defines, based on the indication, a first transport class of a plurality of transport classes. The network device receives a service route for the first network service and stores a correspondence between the service route and the first transport class. The network device receives network traffic associated with the first network service and forwards, based on the correspondence, the network traffic along the underlay tunnels specified by the first transport class.

Abstract: Systems and methods provide automated, lightweight, and centralized implementations for distributing and securely managing the lifecycle of containerized third-party applications on a cluster of network switches that typically have relatively limited computing resources. The deployed containerized applications may be seamlessly updated and expanded separately from the host network operating system without the complexities involved in a registry infrastructure, such as Kubernetes, Docker, or similar container orchestration platform, without compromising security, and without interfering with the common switching functions and switching protocols of switches in the cluster.

Abstract: Systems and methods use zoning configuration to construct border gateway protocol (BGP) route maps. The constructed route maps may then be used to selectively advertise routes between BGP peer devices without having to advertise all available routes to all entities or subsystems in a network. Advantageously, this significantly reduces potential security risks to the network and improves scalability of services in non-volatile memory express over Fabric and other systems.

Abstract: Techniques are disclosed for disseminating network service-specific mapping information across administrative domains. In one example, a network device receives an indication of a route target and one or more underlay tunnels configured to support a service route. The service route is configured to transport network traffic associated with a first network service of a plurality of network services. The network device defines, based on the indication, a first transport class of a plurality of transport classes. The network device receives a service route for the first network service and stores a correspondence between the service route and the first transport class. The network device receives network traffic associated with the first network service and forwards, based on the correspondence, the network traffic along the underlay tunnels specified by the first transport class.

Abstract: Border gateway protocol (BGP) configuration systems and methods automatically establish a BGP peering relationship between a BGP peer device and a centralized discovery controller (CDC) that allow the CDC to advertise routes to the BGP peer device. Various embodiments accomplish this by exchanging BGP information between the CDC, the BGP peer device, and a BGP auto-configuration service, which configures BGP parameters that establish the BGP peering relationship.

Abstract: A switching fabric role assignment system includes a plurality of switch devices coupled together in a switching fabric. A first switch device included in the plurality of switch devices receives an endhost device identification communication from an endhost device when the endhost device is connected to the first switch device. The first switch device uses the endhost device identification communication to identify an endhost device type of the endhost device and determines, based on the endhost device type, a first switch device role for the first switch device. The first switch device then transmits a first switch device role communication that identifies the first switch device role to a second switch device included in the plurality of switch devices and connected to the first switch device. The second switch device may then determine, based on the first switch device role, a second switch device role for the second switch device.

Abstract: An expander I/O module discovery/management system includes a secondary system chassis housing an expander I/O module coupled to a server device. The server device identifies the secondary system chassis and an expander I/O module port utilized by that server device, and then generates and transmits an expander I/O module reporting communication identifying the secondary system chassis and the expander I/O module port. A primary system chassis houses a switching I/O module coupled to the expander I/O module. The switching I/O module receives the expander I/O module reporting communication and determines that the secondary system chassis identified in the expander I/O module reporting communication is different than the primary system chassis. In response, the switching I/O module assigns a virtual slot to the expander I/O module, and assigns a virtual port associated with the virtual slot to the expander I/O module port identified in the expander I/O module reporting communication.

Abstract: A multi-fabric VLAN configuration system includes a first fabric with server devices that are configured to communicate using VLANs, a primary I/O module coupled to the server devices, and a first fabric management system coupled to the server devices and the primary I/O module. The first fabric management system identifies VLAN information associated with the VLANs, automatically configures the primary I/O module using the VLAN information, and causes the VLAN information to be transmitted by the primary I/O module. A second fabric in the multi-fabric VLAN configuration system includes a leaf switch device that is coupled to the primary I/O module and that receives the VLAN information, and a second fabric management system that is coupled to the leaf switch device and that receives the VLAN information from the leaf switch device, and automatically configures the leaf switch device using the VLAN information.

Abstract: An automated Link Aggregation Group (LAG) configuration system includes a plurality of slave switch devices that are each coupled to an endhost device by at least one respective link. Each of the plurality of slave switch devices receives a Link Aggregation Group (LAG) communication from the endhost device, and forwards endhost device information in that LAG communication to a master switch device. The master switch device receives endhost device information from each of the plurality of slave switch devices and determines that each of the plurality of slave switch devices are coupled to the endhost device. In response, the master switch device sends a LAG instruction to each of the plurality of slave switch devices that causes the at least one respective link that couples each of the plurality of slave switch devices to the endhost device to be configured in a LAG.

Abstract: A port configuration migration system includes a primary I/O module connected to a server device via a secondary I/O module. A fabric manager system maps a virtual interface to a first downlink port on the primary I/O module that is connected to the secondary I/O module, with the virtual interface providing a virtual direct connection to the server device. The fabric manager system then configures the virtual interface with communication configuration information for the server device such that communications received via the first downlink port are transmitted using the virtual interface. The fabric manager system then receives a discovery communication from the server device via a second downlink port on the primary I/O module that is connected to the secondary I/O module, and remaps the virtual interface to the second downlink port such that communications received via the second downlink port are transmitted using the virtual interface.

Abstract: A Centralized Discovery Controller (CDC) uses built-in intelligence to determine or estimate whether a connection with a non-volatile memory express (NVMe) entity has been discontinued intentionally or whether a connection loss is rather transient, e.g., due to a temporary network issue. In the former case, the CDC sends out asynchronous event notifications (AENs) and communicates the absence of the NVMe entity in a get log page to indicate an administrative access control action, e.g., a user intervention due to a zoning change or the removal of an entity due to a hardware failure. In the latter case, the CDC creates an “unreachable” entry in the name server database that indicates a CDC connectivity failure but maintains the entity in the name server database despite the connection loss, refraining from sending out notifications to relevant (or impacted) entities to increase bandwidth, traffic stability, and network availability.

Abstract: An expander I/O module discovery/management system includes a secondary system chassis housing an expander I/O module coupled to a server device. The server device identifies the secondary system chassis and an expander I/O module port utilized by that server device, and then generates and transmits an expander I/O module reporting communication identifying the secondary system chassis and the expander I/O module port. A primary system chassis houses a switching I/O module coupled to the expander I/O module. The switching I/O module receives the expander I/O module reporting communication and determines that the secondary system chassis identified in the expander I/O module reporting communication is different than the primary system chassis. In response, the switching I/O module assigns a virtual slot to the expander I/O module, and assigns a virtual port associated with the virtual slot to the expander I/O module port identified in the expander I/O module reporting communication.

Abstract: An automated multi-fabric link aggregation system includes leaf switch devices that have leaf switch device downlink ports, that are included in a first network fabric, and that are aggregated to provide a first aggregation fabric. Each leaf switch device generates discovery communications including a first network fabric identifier for the first network fabric, and a first aggregation fabric identifier for the first aggregation fabric. The leaf switch devices then transmit the discovery communications via the leaf switch device downlink ports. I/O modules that have I/O module uplink port are included in a second network fabric and are aggregated to provide a second aggregation fabric. The I/O modules receive the discovery communications via each of the I/O module uplink ports, determine that each received discovery communication includes the first network fabric identifier and the first aggregation fabric identifier and, in response, automatically configure the I/O module uplink ports in a LAG.

Abstract: A secured network management domain access system includes a chassis housing a master I/O module that is configured to provide a network management domain, and a management module coupled to the master I/O module. The management module includes an enclosure controller coupled to the master I/O module via a first communication channel, and that retrieves master I/O module secured access information from the master I/O module via the first communication channel. The management module also includes a management service coupled to the enclosure controller via a second communication channel and to the master I/O module via a third communication channel, and that retrieves the master I/O module secured access information from the enclosure controller via the second communication channel, and performs validation operations with the master I/O module via the third communication channel such that the management service may securely access the network management domain via the master I/O module.

Abstract: An expander I/O module discovery/management system includes a secondary system chassis housing an expander I/O module coupled to a server device. The server device identifies the secondary system chassis and an expander I/O module port utilized by that server device, and then generates and transmits an expander I/O module reporting communication identifying the secondary system chassis and the expander I/O module port. A primary system chassis houses a switching I/O module coupled to the expander I/O module. The switching I/O module receives the expander I/O module reporting communication and determines that the secondary system chassis identified in the expander I/O module reporting communication is different than the primary system chassis. In response, the switching I/O module assigns a virtual slot to the expander I/O module, and assigns a virtual port associated with the virtual slot to the expander I/O module port identified in the expander I/O module reporting communication.