Arista Networks, Inc.

Amit Dattatray Ranpise

Rajesh Kumar Semwal

Alton Lo

Isidoros Kouvelas

Kallol Mandal

Nitin Singh

Kamlesh Raghuvanshi

Madhu Sudan

Venkit Kasiviswanathan

Anda Wang

Techniques disclosed herein provide a method for efficiently propagating address resolution reply messages. A first router in a first network receives an address resolution request message from a second router in a second network. The first router generates an entry for the address resolution request message and stores the entry in a pending address resolution requests table. When the first router receives a route advertisement, it extracts a network layer address from the route advertisement and determines whether the pending address resolution requests table includes an entry for the network layer address. If so, the router extracts a link layer address from the route advertisement and generates an address resolution reply message comprising the network layer address and the link layer address. The router then transmits the address resolution reply message to the second router.

Efficient ARP packet propagation

In response to receiving an ASR message, a VTEP generates a specially modified control plane message advertising the IP-to-MAC binding of the ASR message. The control plane message may be modified to indicate that it is not to be used for MAC learning. The control plane message is advertised over the network. When an intended recipient receives the message, it uses that message just for the IP-to-MAC binding. When an unintended recipient receives the message, it may drop it as invalid.

Efficient ARP bindings distribution in VPN networks

Techniques described herein provide for fast updating of a forwarding table in a single active multihoming configuration. A first network device that is not connected to an ethernet segment (ES), receives a plurality of ethernet segment (ES) routes (e.g., EVPN type-4 routes) from a plurality of network devices that are connected to a host via the ES. When connectivity is lost to the on a designated forwarded for the ES, t the first network device performed a designated forwarding election algorithm based on the plurality of the received ES routes, to identify that a second network device of the plurality of network devices is designated as a new forwarding device. The first network device modifies an entry in a forwarding table to indicate that the host is now reachable via the second network device.

Multihoming optimizations for fast failover in single-active networks

Systems and methods are provided herein for a mechanism for faster convergence of network traffic after a network device's link is interrupted by leveraging the withdrawal of the ethernet virtual private network (EVPN) auto discovery (AD) route. This may be accomplished by a first device checking an ethernet segment identifier (ESI) status flag before generating an entry in the first device's forwarding table, where the entry is based on an IP route for a host received by a second network device. In response to receiving a withdrawal of an EVPN AD route from the second device, the first device may update the ESI status flag to indicate that the host on the ethernet segment (ES) is reachable only via the third device and update the entry that was based on the IP route for the host received by the second network device to prevent sending traffic to the host via the second device.

Systems and methods for convergence of network traffic after an interruption of a network device's link

Network identifiers are extracted from route advertisements. A table associates virtual network identifiers with provider edge devices. When a virtual network identifier extracted from a route advertisement matches a virtual network identifier in the table, the route advertisement is propagated to the provider edge devices associated with that virtual network identifier in the table. The route advertisement is not propagated to provider edge devices not associated with that virtual network identifier in the table.

Filtering advertising of route advertisements based on virtual network identifiers

Techniques disclosed herein provide a method for configuring a network in DCI environment. An EVPN session is established between a first gateway device of a first network, and a second gateway device of a second network that are linked by L2 DCI link. An ESI is allocated for that EVPN session. A label is created for every combination of the ESI and media access control virtual routing and forwarding table (MAC VRF) that is locally configured at the first gateway device. An EVPN path is received for a host in the first network that is associated with MAC VRF. The path in imported the first MAC VRF by the first gateway device and exported via the inter-DCI EVPN session. The second gateway device identifies a label for MAC and re-exports it in local EVPN session with the identified label.

Systems and method for using DCI-ESI to build highly scalable EVPN DCI networks

Techniques described herein provide for expediting routing convergence in EVPN with multihomed ethernet segment when one of the redundant devices loses connection to the ethernet segment. When a first redundant device receives EVPN auto discovery (AD) route advertising an ethernet segment from a second redundant device, it creates an entry for a forwarding table. The entry has, for an advertised MAC address, a local identifier of the ethernet segment (marked as active next hop) and the identifier of the second network device (marked as backup next hop). When a packet for that MAC address is received, the first redundant device routes the data using data from the entry. In particular, the first redundant device uses the local identifier of the ethernet segment as next hop when the ethernet segment link is active; and uses the identifier of the second network device as next hop when the ethernet segment is down.

EVPN multihoming optimizations for ethernet segment connection interruptions

Techniques described herein provide for fast updating of a forwarding table in a single active multihoming configuration. A first network device that is not connected to an ethernet segment (ES), receives a plurality of ES routes (e.g., EVPN type-4 routes) from a plurality of network devices that are connected to a host via the ES. When connectivity is lost to a designated forwarded for the ES, the first network device performs a designated forwarding election algorithm based on the plurality of received ES routes, to identify that a second network device of the plurality of network devices is designated as a new forwarding device. The first network device modifies an entry in a forwarding table to indicate that the host is now reachable via the second network device.

Techniques are presented herein for managing the processing of network data units (NDUs) received by a network device. More specifically, the techniques relate to updating NDU processing information in a network device in response to a link becoming active, where the updating reduces the NDU loss when the NDU processing information is being updated.

Hitless linkup of ethernet segment

Techniques described herein relate to a method for generating a table for forwarding traffic between networks that are implemented using different types of control planes. In one or more embodiments, the method comprises: receiving a first route advertisement for a first network; extracting a first next hop identifier-multiprotocol label switching (MPLS) label combination; receiving another route advertisement for the first network; and in response to determining that a local label has not been assigned to the first next hop identifier-MPLS label combination: generating a first entry and adding the first entry to a forwarding table of the gateway device.

Interconnecting networks operating different types of control planes

Virtual network identifiers are extracted from route advertisements. A table associates virtual network identifiers with provider edge devices. When a virtual network identifier extracted from a route advertisement matches a virtual network identifier in the table, the route advertisement is propagated to the provider edge devices associated with that virtual network identifier in the table. The route advertisement is not propagated to provider edge devices not associated with that virtual network identifier in the table.

Systems and methods for generating a route advertisement including a sequence number associated with a network layer address. An illustrative method includes receiving a first route advertisement advertising a path to a primary device, the first route advertisement including a network layer address, a first link layer address, and a first sequence number associated with the network layer address; receiving a gratuitous address resolution message from a standby device, the gratuitous address resolution message including the network layer address and a second link layer address; generating a second route advertisement advertising a path to the standby device, the second route advertisement including the network layer address, the second link layer address, and a second sequence number associated with the network layer address, wherein the second sequence number is incremented by a predetermined increment value over the first sequence number; and transmitting the second route advertisement.

Fast failover and recovery for overlay networks

Techniques for implementing neighbor equivalence groups on a network device are provided, where a neighbor equivalence group is a group of peers of the network device that communicate identical control plane state information for a given network protocol to the network device. In the context of Border Gateway Protocol (BGP), these techniques can include (1) creating, by the network device, a neighbor equivalence group for a set of BGP peers that advertise the same BGP paths to the device, (2) maintaining, by the network device, a single path database for the neighbor equivalence group (rather than one path database for each BGP peer in the group), and (3) immediately processing by the network device, BGP path update/withdrawal messages received from any of the BGP peers in the group against the single path database, without waiting for the same message to be received from every peer.

Growing community of inventors

San Jose, CA, United States of America

Amit Dattatray Ranpise