Arista ACE Specialist - Data Center key terms

A two-tier data center topology in which every leaf switch connects to every spine switch, giving uniform, predictable hop counts between servers. Leaves connect to hosts; spines only interconnect leaves and forward traffic between them.

The physical IP-routed fabric that carries encapsulated traffic between leaf and spine devices. It provides loop-free reachability, usually through a routing protocol such as OSPF or BGP, and does not carry tenant VLAN semantics itself.

A virtual network built on top of the physical underlay, using encapsulation to carry tenant Layer 2 or Layer 3 traffic between endpoints. VXLAN is the common encapsulation used in Arista data center designs.

A tunneling protocol that encapsulates Layer 2 frames inside UDP packets for transport across a Layer 3 network. A 24-bit VNI identifies each virtual network, allowing far more segments than traditional VLAN IDs.

A tunnel endpoint that encapsulates and decapsulates VXLAN traffic. On Arista switches the VTEP function is typically performed in hardware, mapping local VLANs to VNIs and learning remote MAC addresses over the tunnel.

A 24-bit identifier carried in the VXLAN header that distinguishes one virtual network segment from another. Each VNI is mapped to a local VLAN on the VTEP so that tenant traffic stays isolated.

A Layer 2 broadcast domain identified by a 12-bit tag inserted into Ethernet frames on trunk links. VLANs isolate traffic on shared switch ports and are commonly mapped to VNIs in an overlay design.

A switch interface configured to carry traffic for a single VLAN, typically connecting to an end host. Frames leaving the port are untagged, so the attached device needs no VLAN awareness.

A switch interface that carries multiple VLANs by tagging frames with 802.1Q headers. It typically connects switches or switch-to-hypervisor links where several broadcast domains must share one physical link.

Arista's multi-chassis link aggregation feature, in which two switches present a single logical LAG to downstream devices. Peers synchronize state over a dedicated peer link and use a peer-keepalive path to detect failures.

An IEEE standard protocol that negotiates and maintains link aggregation groups between two devices. It verifies that member links share compatible configuration and can detect a failed or misconfigured member.

A forwarding technique that load-balances traffic across multiple equal-cost paths by hashing packet header fields. In leaf-spine fabrics it allows all uplinks to be used simultaneously without blocking any link.

A Layer 2 protocol that prevents forwarding loops by electing a root bridge and blocking redundant ports. In routed leaf-spine fabrics it is often unnecessary in the underlay but still relevant to host-facing Layer 2 segments.

A path-vector routing protocol used in data center underlays and for connectivity to external networks. Arista supports eBGP designs where each leaf and spine has a unique autonomous system number for multipath routing.

A link-state interior gateway protocol that computes shortest paths using a shared topology database. It is a common choice for leaf-spine underlays because it converges quickly and supports equal-cost multipath.

A value prepended to a route to make it unique within a VRF context on a provider edge device. It does not select the VRF; it only ensures that overlapping prefixes from different tenants remain distinct.

A BGP extended community that controls which VRFs import or export a given route. Matching export and import targets is what actually leaks or isolates prefixes between routing contexts.

A virtual routing and forwarding instance that maintains its own separate routing table and forwarding state on a device. Multiple VRFs on one switch let overlapping address space serve different tenants or functions.

A design in which every leaf uses the same virtual MAC and IP address as the default gateway for a subnet. Hosts can move anywhere in the fabric and keep their gateway, avoiding re-ARP when they migrate.

A protocol that resolves an IPv4 address to a MAC address on a local broadcast domain. In overlay networks ARP requests may be flooded or handled by control-plane learning to reduce broadcast traffic.

A forwarding database that maps learned MAC addresses to switch ports and VLANs. The switch uses it to forward frames only out the port where the destination was last seen.

Arista's Extensible Operating System, a Linux-based network operating system with a modular, state-driven architecture. Its single-image design and open APIs support automation and consistent behavior across platforms.

The central state database in EOS that holds configuration and operational state for all agents. Processes publish and subscribe to state changes there, which is why EOS behavior is described as state-driven.

Arista's JSON-based programmatic interface for EOS, exposing CLI commands and configuration over HTTP or HTTPS. It lets automation tools run show commands and apply changes without screen-scraping a terminal.

Arista's management and telemetry platform that provides a single pane for configuration, change control, compliance, and streaming state across many EOS devices. It supports both on-premises and cloud-hosted deployments.

The web interface of CloudVision where operators view inventory, topology, alerts, and device state, and where they build and review configuration changes before deployment.

A CloudVision workflow that packages proposed configuration changes into a reviewable task, optionally requiring approval before execution. It records who changed what and supports rollback.

A push model in which network devices periodically send structured operational data to a collector rather than waiting to be polled. It gives near-real-time visibility into interface, queue, and system counters.

A gRPC-based protocol for retrieving, setting, and subscribing to configuration and state on network devices using modeled paths. It is widely used for streaming telemetry and intent-based automation.

A boot process in which a new switch obtains its software image and startup configuration automatically from a server or CloudVision, with no manual console interaction. It relies on DHCP options and a reachable provisioning service.

The active configuration currently in effect in device memory. Changes take effect immediately but are lost on reload unless the configuration is saved to startup.

The saved configuration file that a device loads during boot to become the running configuration. On EOS it is stored locally and can also be pulled from a central provisioning source.

A logical bundle of physical interfaces treated as one link for forwarding and redundancy. Member links must match in speed and configuration, and the bundle is often negotiated with LACP.

A switched virtual interface that gives a VLAN an IP address so the switch can route for that subnet or be managed on it. It is the typical host-facing gateway in Layer 3 leaf designs.

The maximum size of a frame or packet a link will carry without fragmentation. VXLAN adds encapsulation overhead, so underlay links usually need a larger MTU than the tenant payload requires.

An Ethernet frame larger than the standard 1500-byte payload, commonly up to 9000 bytes. Data center fabrics use them to reduce per-packet overhead for storage and overlay traffic.

A diagnostic tool that discovers the path packets take by sending probes with increasing time-to-live values and recording the routers that return errors. It helps locate where traffic is dropped or delayed.

A sampling-based traffic monitoring protocol that exports headers of sampled packets to a collector. It gives flow-level visibility with low overhead on high-speed interfaces.