Arista ACE Specialist - Data Center study materials
Arista ACE Specialist - Data Center study resources

Arista ACE Specialist - Data Center cheat sheet

Arista ACE Specialist - Data Center · Keep the key ideas within reach.

Quick reference

Data Center Networking Fundamentals

  • Leaf-spine (Clos) topology: every leaf connects to every spine; servers attach only to leaves; equal-cost paths between leaves are ECMP across spines.

  • Spine layer provides no server ports and no policy; it is a pure high-bandwidth transit layer, so oversubscription is engineered at the leaf uplinks.

  • Oversubscription ratio = total downlink (host-facing) bandwidth / total uplink bandwidth; 1:1 is non-blocking, 3:1 is common for general compute.

  • East-west traffic dominates modern data centers; design capacity for server-to-server flows rather than north-south client traffic.

  • VXLAN encapsulates L2 frames in UDP (dest port 4789) with a 24-bit VNI, allowing ~16M segments versus 4094 VLANs.

  • VXLAN requires an underlay providing IP reachability between VTEPs; the overlay carries tenant MACs and ARP, the underlay carries only VTEP loopbacks.

  • MLAG uses a peer link plus a shared LACP port-channel so two switches present one logical LAG to servers; it is not a loop-prevention protocol like STP.

  • Arista 7050X/7300 series are fixed/compact leaf-spine platforms; 7500R/7800R use deep buffers and high-density line cards for spine and border roles.

Quick reference

Arista EOS Configuration and Operations

  • EOS CLI modes: enable (>), configure (config), and interface/config submodes; 'show running-config' displays active config, 'write memory' or 'copy running-config startup-config' persists it.

  • Config sessions: 'configure session NAME' stages changes, 'show session-config diff' previews, and 'commit' applies atomically; 'abort' discards.

  • EOS supports multi-agent architecture: separate processes (e.g., routing protocols, LLDP) run as agents, so one protocol crash does not halt forwarding.

  • Management interfaces: Management1 is out-of-band; 'management api http-commands' enables eAPI (JSON-RPC over HTTP/HTTPS) for automation.

  • MLAG config requires matching domain-id, 'mlag peer-link' on both peers, and a unique local-interface per peer; peer link should be a dedicated port-channel.

  • 'show mlag' reports state (active/inactive), peer status, and consistency; 'show mlag config-sanity' validates peer settings.

  • ZTP (Zero Touch Provisioning) bootstraps a switch from DHCP options plus a startup script or config file fetched over HTTP/TFTP.

  • Event-driven features: 'event-handler' triggers scripts/CLI on log patterns or interface state changes; 'schedule' runs commands at intervals.

  • VLAN and trunk config: 'switchport mode trunk', 'switchport trunk allowed vlan'; MLAG member ports must match on both peers.

Quick reference

CloudVision and Network Automation

  • CloudVision is a multi-tier platform: CVP (on-prem appliance/VM) and CVaaS (cloud-hosted); both provide change control, telemetry, and compliance.

  • Change Control (formerly Change Management) workflows stage configs, run pre-checks, require approval, and support rollback to a prior snapshot.

  • CloudVision Studios provide intent-based, template-driven config generation (e.g., base config, VLAN, MLAG studios) with device tagging.

  • Streaming telemetry uses gNMI/gRPC push from EOS to CloudVision, giving near-real-time state versus periodic SNMP polling.

  • eAPI automation: send JSON-RPC 'runCmds' to /command-api; responses include 'result' array; use 'format':'json' for structured output.

  • Arista Python library 'pyeapi' wraps eAPI; 'jsonrpclib' or raw requests can call it directly; Ansible 'arista.eos' collection uses eAPI or CLI transport.

  • CloudVision topology and compliance dashboards show per-device config drift against defined tags/studios, flagging non-compliant devices.

  • Device tagging in CloudVision drives which studios and configlets apply; tags are hierarchical (e.g., campus/building/role).

Quick reference

Data Center Design, Routing, and Troubleshooting

  • Underlay routing commonly uses eBGP with unique ASNs per leaf/spine or OSPF; eBGP provides loop-free multipath and clear failure domains.

  • ECMP in EOS: 'maximum-paths' controls multipath count; hashing can be tuned with 'ip load-sharing' for L3/L4 fields.

  • Anycast gateway (VARP/Anycast GW) lets every leaf use the same virtual IP/MAC for a subnet, enabling optimal routing without traffic tromboning.

  • VARP config: 'ip virtual-router mac-address' plus 'ip address virtual' on the SVI; all leaves share the virtual MAC so hosts keep the same gateway after vMotion.

  • EVPN control plane: MP-BGP AFI 25 (L2VPN EVPN) carries MAC/IP routes (type 2), IMET (type 3), and Ethernet segment (type 4) for multihoming.

  • EVPN multihoming (ESI) replaces MLAG for active-active host attachment across leaves, using type-1/type-4 routes for DF election.

  • Troubleshooting order: verify physical (show interfaces status), then L2 (show mac address-table, show vlan), then L3 (show ip route, ping/traceroute), then overlay (show vxlan vtep, show bgp evpn).

  • 'show interfaces counters errors' and 'show interfaces counters rates' isolate CRC/alignment errors and congestion; CRC errors typically indicate cabling/optics faults.

  • Spanning-tree: EOS defaults to MSTP; rapid-pvst is available; in leaf-spine with MLAG/EVPN, STP is largely confined to the leaf pair.