I85PREP
⌘ K
Cisco DCID 300-610 study materials
Cisco DCID 300-610 study resources

Cisco DCID 300-610: Data Center Network Design

Cisco DCID 300-610 · Work through your exam, one topic at a time.

All guide sections

Data Center Network Design

The Data Center Network Design domain carries roughly a quarter of the 300-610 exam and asks you to convert requirements-scale, L2 stretch, multi-tenancy, failure domains-into a validated spine-leaf design. Expect vPC, VXLAN EVPN, underlay routing, and legacy L2 extension (OTV, FabricPath) questions where the answer depends on which technology the requirement actually forces.

Core design rules

  • Underlay first. Build a routed spine-leaf with eBGP (unique private ASN per leaf, spines as a single AS or AS-set), /31 links, and ECMP. The overlay (VXLAN EVPN with the BGP EVPN address family) rides on top and never depends on STP.
  • vPC fundamentals. The peer-link is a port channel carrying all vPC VLANs and should be at least 10 Gb, typically 2x100 Gb; peer-keepalive belongs on a separate L3 path (mgmt0 or a dedicated link), never on the peer-link. Both peers share a vPC domain ID, role priority, and system MAC.
  • VTEP and gateway. Each leaf needs a dedicated NVE loopback; distributed anycast gateway uses the same virtual MAC/IP on every leaf, with ARP suppression to limit flooding.
  • MTU. 9216 jumbo end-to-end, including the peer-link, or VXLAN encapsulation will silently drop large frames.
  • L2 extension. For new designs prefer EVPN multi-site; OTV is the answer only when extending to a site without VXLAN or when the question is explicitly about OTV adjacency server or site VLAN.

Worked example

A leaf pair front-ends 48 servers, each dual-homed with 2x25G (one link per leaf). Requirement: survive a single leaf failure without host-visible oversubscription. Per-leaf host bandwidth = 48 x 25G = 1.2 Tb/s. If leaf A fails, leaf B carries all 48 servers, so the peer-link must absorb 1.2 Tb/s of redirected traffic plus orphan-port traffic. Therefore size the peer-link at 12x100G, or accept and document an oversubscription ratio. Uplinks to the spines are sized separately from north-south demand, since vPC does not add uplink bandwidth per flow.

Traps

  • Putting peer-keepalive on the peer-link, or letting both peers be STP root.
  • Reusing a vPC domain ID across different pairs.
  • Assuming vPC provides an active/active L3 path for a single flow-it is per-flow hashing.
  • Forgetting orphan ports (single-homed devices) in the failure design.
  • Designing VXLAN without verifying underlay MTU and the multicast/replication mode.