Data Center Architecture Decision Workshop
Compare four architecture candidates through an explicit local preference rubric, requirement gaps and migration review.
Beta · Local editorial preference rubric; not an engineering feasibility, capacity or availability verdict. No hidden AI.
Only explicit local save stores inputs in this browser. Exports contain planning inputs. No live collection, credentials or input-derived analytics.
Architecture comparison
Complete the workshop inputs and compare architectures. Ratings express preference, not capacity.
5. Rationale and migration review
Why not VXLAN?
If applications accept routed boundaries and no overlay segmentation is required, VXLAN introduces VTEP/VNI/MTU and control-plane work without a declared requirement. First document the missing capability, then justify the added layer. Preference scores alone cannot establish cost or benefit.
S-JUNIPER-EVPN
When simple L3 is enough
A routed fabric can be a candidate when applications tolerate subnet boundaries and no cross-rack L2 service is required. Verify addressing, ECMP behavior, failure paths and actual link capacity. Counts here do not validate any of those engineering properties.
S-RFC7938
When EVPN adds value
Required overlay segmentation or within-fabric L2 extension can justify evaluating EVPN/VXLAN. Validate service-to-VNI/VRF mapping, supported routes, MTU and multivendor behavior. A positive rating is not evidence that a particular platform meets the requirement.
S-JUNIPER-EVPN
When controller/fabric complexity is justified
Evaluate a controller fabric when consistent policy coordination and lifecycle governance are valuable and the team can operate the chosen model. ACI is one vendor-specific example. Compare licensing, APIs, recovery, ecosystem coupling and training before selecting; do not assume every controller fabric has identical behavior.
S-CISCO-ACI
Migration considerations
Migration difficulty remains UNASSESSED: the current topology, maintenance windows and application dependencies were not collected. Inventory gateways, addressing and policy; test coexistence, failure recovery and rollback. Plan a pilot with measurable acceptance criteria and operational ownership before expanding.
S-LOCAL-RUBRIC
Questions
Does a high score prove capacity?
No. Counts add zero points and do not validate links, tables, licenses or topology.
Why is migration unassessed?
Existing architecture, maintenance windows and application dependencies were not collected. Review the candidate-specific migration considerations.
Sources and method
FitScore = sum of ten 0–4 ratings / 40 × 100. Every weight is 1. S-LOCAL-RUBRIC owns arbitrary editorial ratings; primary references support architectural principles only.
- RFC 7938 — Use of BGP for Routing in Large-Scale Data Centers
Sections 3–4 discuss traditional and Clos designs, and L3-only networks with L2 domains confined toward the edge. Informational operational experience, not a universal size threshold.
- Juniper — EVPN with VXLAN Data Plane Encapsulation
EVPN provides control-plane learning; VXLAN supplies an L2 overlay over an IP network. Overlay segmentation may serve multitenancy and required L2 connectivity; platform support remains vendor-specific.
- Cisco Application Centric Infrastructure Design Guide
ACI illustrates controller-coordinated policy and tenant/VRF/bridge-domain constructs. It is a vendor-specific example, not a requirement for every controller fabric. Validate controller lifecycle, policy dependencies and licensing separately.
Engineering workflow
Fabric Capacity Planner — separate bandwidth and ECMP calculationRelated resources
Fabric Capacity Planner — separate bandwidth and ECMP calculation