Parent folderOpen Confluence

RWS Architecture article

DC3.0 Network and Supporting Services

Deliver lane-scoped Phase 1 networking: Location A single-ACI-pod AI lane and Location B VXLAN BGP EVPN spine-leaf target lane, plus supporting services for DNS, load balancing, fi

  1. Typeworkpackage
  2. Statusplanned
  3. Domainnetwork
On this page
  1. WP-01 DC3.0 Network and Supporting Services
  2. Summary
  3. Scope
  4. Architecture Context
  5. Decision Context
  6. Dependencies
  7. Acceptance Criteria

WP-01 DC3.0 Network and Supporting Services

Summary

Deliver lane-scoped Phase 1 networking: Location A single-ACI-pod AI lane and Location B VXLAN BGP EVPN spine-leaf target lane, plus supporting services for DNS, load balancing, firewalling, and identity-aware access.

Scope

Deliver lane-scoped Phase 1 networking: Location A single-ACI-pod AI lane (strictly AI use case only) and Location B VXLAN BGP EVPN spine-leaf target lane (L3VNI-per-VRF, L2VNI-per-VLAN segmentation, Anycast GW), plus supporting services for DNS, load balancing, firewalling, and identity-aware access.

  • Enforce explicit service-versus-transport boundaries in design and validation artifacts (L3VPN service, SR-MPLS transport, EVPN control-plane, VXLAN dataplane).
  • Implement default L3 interconnect between datacenter EVPN VRFs and backbone L3VPN VRFs.
  • Treat any cross-site L2 stretch request as exception-only with ADR-level approval.

Architecture Context

Decision Context

Dependencies

Acceptance Criteria

  • Location A single ACI pod lane is operational for the AI cluster use case only, with documented controls preventing non-AI tenant onboarding.
  • Location A and Location B lane boundaries are documented, approved, and auditable.
  • MTU validation shows no fragmentation for in-scope east-west and north-south paths, including encapsulation budgets for EVPN/VXLAN and OpenShift overlays.
  • Network-path benchmarks for storage traffic (replication and client data paths) meet target transport SLOs.
  • BFD failure detection meets the defined p95 target for in-scope adjacency tests in the approved failover test profile.
  • BGP reconvergence for the approved failover test profile (for example single-link loss and single-node failure scenarios) meets p95 < 2s, with scenario definitions and measurement method captured in evidence artifacts.
  • EVPN control-plane health (show bgp l2vpn evpn) and dataplane health (show nve peers and show nve vni) pass change gates before close.
  • Production-target multihoming evidence confirms ESI state stability and expected PIP/VIP transition behavior in scoped failover and recovery tests.
  • Scoped routed workload tests confirm no unintended node-level NAT on CUDN-first paths and preserve workload source identity for policy and observability controls.
  • No unauthorized route leaks are observed in scoped validation evidence; negative tests for defined tenant pairs confirm RT isolation remains intact.
  • Network control evidence for cutover is published in the release evidence bundle and approved by network and platform owners.