Dual-Site Holodeck Deployment (Holodeck 9.1)
Dual-site is materially reworked in 9.1. New-HoloDeckConfig now AUTO-GENERATES both Site A and Site B configuration files in one pass, with Site A automatically loaded into the PowerShell session. BREAKING: Import-HoloDeckConfig now takes a MANDATORY -Site a|b — the Site B instance is brought up in a separate PowerShell session via 'Import-HoloDeckConfig -ConfigID <id> -Site b' before New-HoloDeckInstance -Site b. Verified topology defaults: Site A CIDR 10.1.0.0/20, VLANs 0 and 10-25, -VLANRangeStart 10, >=16 consecutive VLANs, domain site-a.vcf.lab; Site B CIDR 10.2.0.0/20, VLANs 40-58, -VLANRangeStart 40, >=19 consecutive VLANs, domain site-b.vcf.lab. MasterCIDR must be /20. Canonical sequence: New-HoloDeckNetworkConfig -Site a, then -Site b, then Set-HoloRouter -dualsite, then New-HoloDeckInstance per site. WARNING (verbatim): if custom CIDR/VLAN ranges are used in New-HoloDeckNetworkConfig, the same values must be passed to New-HoloDeckInstance or they are overwritten by defaults. 9.1 fixed several Site B defects: VLAN range start handling, depot alias configuration and search paths, and VCF Installer Site B network port-group failures. CAPACITY CEILING: the Holodeck 9.1 docs recommend 96 CPU / 1536 GB / 5 TB for dual-site VCF 9.1, well beyond a 1 TB host; the dual-site VVF 9.1 recommendation (48 / 1024 GB / 4 TB) sits exactly on the RAM line with no headroom. Soft validation will let either start anyway and degrade — size before you deploy, do not rely on pre-checks to stop you. SHARED HOST: one HoloRouter per Holodeck instance is required (maintainer, GitHub Discussion #178); never run Set-HoloRouter on a HoloRouter that belongs to another instance, and give each instance on a shared physical host its own HoloRouter and -VLANRangeStart.
Objectives
- Design a non-overlapping IP addressing scheme for a dual-site Holodeck 9.1 topology, starting from the auto-generated Site A + Site B config pair
- Deploy the Site B management domain with cross-site routing on Holodeck 9.1
- Establish and validate cross-site routing through the HoloRouter (Set-HoloRouter -dualsite, one VRF per site), and explain why an NSX VNA cluster inside a site does not change that routing path
- Validate bidirectional management plane reachability and DNS cross-resolution using the 9.1 Technitium DNS service
- Map a dual-site lab topology to production VCDX disaster recovery and stretched cluster architecture using the RCAR framework
Prerequisites
Single Holodeck 9.1 management domain (VCF 9.1.0.0) fully deployed and healthy (holodeck91-03 completed). Physical host must have the documented dual-site headroom for Site B (Holodeck 9.1 docs: dual-site VCF 9.1 = 96 CPU / 1536 GB RAM / 5 TB SSD-NVMe; 768 GB / 64 CPU / 2.5 TB was the VCF 9.0 dual-site row). Site A instance in a stable state — snapshot taken. No existing dual-site configuration active.
Prior labs: holodeck91-03
Required skills:
- VCF management domain architecture (SDDC Manager, vCenter, NSX)
- IP address planning and CIDR notation
- BGP routing concepts (or comfort with static route configuration)
- vCenter linking in VCF Operations (and why Enhanced Linked Mode is deprecated in VCF 9.x)
- NSX federation and multi-site networking
- Disaster recovery and RTO/RPO analysis
Lab Environment
Dual Holodeck 9.1 instances on a single physical host. Site A (production) and Site B (DR) use the non-overlapping CIDR/VLAN allocations from the auto-generated 9.1 config pair (Holodeck 9.1 defaults: Site A 10.1.0.0/20 with VLANs 0 and 10-25; Site B 10.2.0.0/20 with VLANs 40-58). Cross-site routing is provided by the HoloRouter dual-site function (Set-HoloRouter -dualsite configures a separate VRF per site and routes between the Site A and Site B VLANs — Holodeck 9.1 docs, Holorouter Overview). A VNA cluster is an optional NSX component inside a site's management or workload domain (distributed networking, e.g. Supervisor in Distributed mode); it does not replace the inter-site router.
NSX and SDDC Manager use the Holodeck master password listed in the 9.1 docs Post Deployment credentials table; no separate or doubled password is documented for API calls. Documented 9.1 behaviour: Import-HoloDeckConfig takes a mandatory -Site a|b and must be re-run at the start of each new PowerShell session (Holodeck 9.1 release notes and cmd_reference).
Mermaid source
graph TB
PHY[Single Physical ESXi Host] --> RTR[HoloRouter 9.1 dual-site - VRF per site]
subgraph SiteA[Site A - Production]
SDDC_A[SDDC Manager A]
VC_A[vCenter A]
NSX_A[NSX Manager A]
end
subgraph SiteB[Site B - DR]
SDDC_B[SDDC Manager B]
VC_B[vCenter B]
NSX_B[NSX Manager B]
end
RTR -->|Cross-site routing| SiteA
RTR -->|Cross-site routing| SiteBIP Addressing
| Network | Purpose | VLAN |
|---|---|---|
10.1.0.0/20 | Site A supernet (Holodeck 9.1 default; custom CIDRs must be /20) | 0 and 10-25 (default -VLANRangeStart 10; a custom start needs >=16 consecutive VLANs) |
10.2.0.0/20 | Site B supernet (Holodeck 9.1 default; custom CIDRs must be /20) | 40-58 (default -VLANRangeStart 40; a custom start needs >=19 consecutive VLANs) |
Credentials
| System | Username | Password |
|---|---|---|
| Physical ESXi Host | root | Set during ESXi installation |
| HoloRouter 9.1 services (Webtop :30000 now authenticated; Authentik SSO :9443) | verify against live 9.1 toolkit | Webtop is behind authentication in 9.1; Authentik defaults are an open question — capture on first live deployment |
| SDDC Manager / vCenter (both sites) | administrator@vsphere.local | Set in the VCF bring-up spec per site |
| NSX Manager (both sites) | admin | Set in the VCF bring-up spec per site (Holodeck default: the master password listed in the 9.1 docs Post Deployment credentials table; there is no separate API password) |
Tasks
Task 1 Plan the dual-site topology and IP addressing scheme
availabilityIn VCDX design, topology and addressing are foundational. 9.1 changes the starting point: the toolkit hands you an auto-generated Site A + Site B config pair, so planning becomes a review-and-align exercise rather than a from-scratch design — which makes collision analysis on shared hosts MORE important, not less. This task mirrors the VCF Planning and Preparation Workbook exercise for multi-site deployments.
Document your Site A baseline (from holodeck91-03) in a design worksheet: management CIDR, per-function subnets, and VLAN allocations as actually deployed.
Review the auto-generated Site B configuration produced by New-HoloDeckConfig (confirmed 9.1 behavior: both sites generated in one pass). Verify every Site B CIDR and VLAN is non-overlapping with Site A AND with anything else on the physical network.
Document cross-site routing requirements: both sites must reach each other's management plane (SDDC Manager, vCenter, NSX). Record the gateway/router interfaces for each direction. The HoloRouter exposes one VRF per site after Set-HoloRouter -dualsite; list the per-site gateways with Get-HoloDeckSubnet -Site a / -Site b.
Analyze physical resource requirements: Site B roughly doubles the footprint (Holodeck 9.1 docs, recommended for dual-site VCF 9.1: 96 CPU, 1536 GB RAM, 5 TB disk on SSD/NVMe; soft validation only warns below these values). Verify actual per-site consumption (plus GitOps/VNA overhead) and confirm the physical host has headroom; document contention risks if not.
Create a failure domain analysis: on a single physical host, both sites share one failure domain. Document how this differs from production dual-site and how it affects the RTO/RPO strategy developed in Task 5.
Validation Gate
Check: Review the worksheet: (1) Site A/B CIDRs non-overlapping, (2) VLAN ranges non-overlapping, (3) cross-site routing table complete, (4) resources verified against capacity, (5) failure domain analysis documented
Expected: All five points pass. Ready to deploy Site B.
Common Errors
Task 2 Deploy the Site B Holodeck instance with cross-site network configuration
availabilityExecuting the dual-site deployment is the operational foundation for VCDX multi-site architecture discussions: two independent VCF bring-ups on shared infrastructure, with resource contention and ordering dependencies. The 9.1 sequence is documented (Holodeck 9.1 docs, Dual Site Deployment): New-HoloDeckNetworkConfig per site, Set-HoloRouter -dualsite, New-HoloDeckInstance -Site a, then -Site b from a second PowerShell tab; what 9.1 adds over 9.0 is the two-config-file New-HoloDeckConfig and the -Site parameter.
Open your management session (PowerShell workstation or the authenticated 9.1 Webtop) and navigate to the Holodeck runtime directory.
Locate the Site B configuration created by your New-HoloDeckConfig run (see holodeck91-01/91-05) and load it in a second PowerShell tab with Import-HoloDeckConfig -ConfigId <id> -Site b. The per-site network configuration step still exists in 9.1: New-HoloDeckNetworkConfig -Site b -MasterCIDR 10.2.0.0/20 (custom: /20 CIDR plus -VLANRangeStart).
Confirm the Site A network configuration in use matches what was deployed in holodeck91-03 (no drift).
Apply any Site B adjustments decided in Task 1 (CIDR/VLAN alignment, resource sizing). Note: Site B port-group issues were fixed in 9.1 — if you still hit port-group errors, capture them as new 9.1 findings rather than assuming 9.0 causes.
Generate the per-site network configs and enable dual-site routing on the HoloRouter, in the documented 9.1 order: New-HoloDeckNetworkConfig -Site a -MasterCIDR 10.1.0.0/20, then New-HoloDeckNetworkConfig -Site b -MasterCIDR 10.2.0.0/20, then Set-HoloRouter -dualsite (configures a separate VRF table per site). Set-HoloRouter -dualsite is retained in 9.1 (cmd reference).
Review the Site B configuration one final time: site identifier, management CIDR, resource sizing within the contention envelope planned in Task 1.
Deploy Site B from the second PowerShell tab: New-HoloDeckInstance -Site b [same options as Site A] (custom ranges: -CIDR <site-a-cidr>,<site-b-cidr> -VLANRangeStart <a>,<b>). GitOps path: trigger the Site B pipeline run in GitLab with the Site B inputs.
Verify the router's cross-site routing state. FRR remains the routing daemon on the 9.1 HoloRouter (Set-HoloRouter cmd_reference: 'FRR for BGP routing'), so inspect its routing table for both site supernets with vtysh on the HoloRouter, and compare the BGP settings with Get-HoloDeckBGPConfig (9.1 cmd reference).
Start the Site B VCF 9.1.0.0 deployment and monitor to completion, watching physical host memory/CPU for contention with Site A throughout (Academy-measured 9.0 baseline: 90-150 minutes; official docs publish only 9.0.0.0 timings — ManagementOnly 4-5 h — so treat Academy numbers as environment-specific; verify 9.1 timing).
Validation Gate
Check: Run Get-HoloDeckInstance -InstanceID <id> from each site's PowerShell session (after Import-HoloDeckConfig -ConfigId <id> -Site a|b); it reads the instance's output.json: both sites should report a completed VCF 9.1.0.0 deployment
Expected: Both Holodeck instances (Site A and Site B) report a completed VCF 9.1.0.0 deployment
Common Errors
Task 3 Configure cross-site networking and validate bidirectional reachability
manageabilityCross-site networking is where the architecture comes alive: Layer 3 design, failure domain isolation, and management traffic engineering. On 9.1 the DNS layer changed (Technitium replaces the dnsmasq-era path), while routing between sites stays on the HoloRouter (one VRF per site) — a VNA cluster inside a site adds distributed stateful services for VPCs, not inter-site routing. This task doubles as the re-baselining exercise for every 9.0-era networking assumption.
Verify cross-site routes on the 9.1 router function: both site supernets must be reachable in each direction. Inspect with vtysh on the HoloRouter: FRR is retained as the BGP daemon in 9.1 (Set-HoloRouter cmd_reference).
Test reachability from Site A to the Site B management plane: ping the Site B SDDC Manager and vCenter IPs.
Test reachability from Site B to the Site A management plane: ping the Site A SDDC Manager and vCenter IPs.
Configure DNS cross-resolution between sites. On 9.1 this is done in Technitium DNS (admin UI on :5380, behind the HTTPS reverse proxy) — the 9.0-era method of editing dnsmasq configuration does NOT carry forward. Exact zone/record workflow: verify against live 9.1 toolkit, then add each site's management FQDNs to the other site's resolution path.
Validate SDDC Manager awareness of both sites from each site's UI (Inventory > Workload Domains). Document any licensing constraints on multi-domain configuration.
Optional (capacity permitting): deploy an NSX Global Manager, make it active, and add both sites' NSX Managers as locations under System > Location Manager on the Global Manager (Add On-Prem Location; VCF 9.0.1 docs p.2301). Size the Global Manager appliance into the dual-site host budget before you start, and check that your licensing covers NSX Federation.
Test management plane operations across sites (e.g. vCenter inventory visibility, test object creation) to prove bidirectional operational reachability.
Validation Gate
Check: Three checks: (1) Site A pings Site B management IPs, (2) Site B pings Site A management IPs, (3) cross-site FQDNs resolve in both directions via Technitium
Expected: All three pass. Bidirectional IP and DNS reachability confirmed.
Common Errors
Task 4 Validate dual-site management plane operations
manageabilityPost-deployment validation proves the dual-site architecture is operational and ready for VCDX-level defense: single-pane-of-glass management, per-site visibility, and cross-site link health. These UI-driven checks are largely version-agnostic; note any 9.1 UI differences as enrichment material.
From the (authenticated) Webtop or your workstation browser, open SDDC Manager A and verify the management domain shows Active with all hosts and clusters healthy.
Open SDDC Manager B and verify the Site B management domain shows Active.
Open vCenter A > Hosts and Clusters: verify all nested hosts connected, vSAN health green, HA/DRS enabled.
Open vCenter B > Hosts and Clusters: verify the same for Site B.
Optional: link the two sites' vCenter instances in a vCenter Group in VCF Operations (VCF Operations 9.1: Operate > Configurations > Logical Groupings > vCenter Linking; in 9.0 the path is Infrastructure Operations > Configurations > Logical Groupings > vCenter Linking — Broadcom KB 444621) using FQDNs (requires the Task 3 DNS cross-resolution) and a common identity provider configured on both vCenter instances, with the IdP user granted permissions on each vCenter individually (VCF 9.0 docs, vCenter linking prerequisites, p.2842). Do not configure Enhanced Linked Mode: it is deprecated from vCenter 9.0, kept only for the upgrade of existing infrastructure, and the recommended alternative is grouping in VCF Operations (VCF 9.0 release notes, Product Support Notes).
Open each site's NSX Manager > System > Fabric > Nodes and verify all host transport nodes show Success/Up.
Test cross-site management visibility (VCF Operations vCenter Group inventory or per-site UIs) by creating a test object at Site A and confirming visibility per your chosen management model.
Document dual-site operational readiness: SDDC Managers healthy, vCenters healthy, NSX healthy, cross-site ping and DNS confirmed, management traffic flowing. Add a 9.1-specific appendix noting every place the live experience diverged from this stub.
Validation Gate
Check: Verify: (1) both SDDC Managers Active, (2) both vCenters healthy with green vSAN/HA/DRS, (3) vCenter linking in VCF Operations connected if configured, (4) all NSX transport nodes up at both sites
Expected: All management components dual-site healthy; cross-site operations functional
Common Errors
Task 5 Design exercise — map dual-site Holodeck to production DR architecture using RCAR
recoverabilityThis is the VCDX defense moment: articulate the architectural decisions, trade-offs, and risk mitigations of the working dual-site lab using the RCAR framework. The 9.1 twists worth defending: dual-site config generation is now the toolkit default — discuss what 'DR rehearsal out of the box' changes about lab-to-production mapping — and the Distributed (VNA) versus Centralized (NSX Edge) networking choice inside each site, which changes the per-site failure domain but not the inter-site router.
Define the production DR scenario your dual-site lab represents: primary site, DR site, business objectives, RPO/RTO targets.
Create the RCAR analysis skeleton: Requirements, Constraints, Assumptions, Risks.
REQUIREMENTS: list at least 5 (replication capability, network failover, cross-site management, RPO/RTO, encrypted management traffic).
CONSTRAINTS: list at least 4 (single physical host, lab latency vs WAN reality, licensing, L2 adjacency limits; add 9.1-specific unknowns such as VNA sizing minimums).
ASSUMPTIONS: list at least 4 (replication licensing, cross-site latency/bandwidth, Site B readiness, manual vs automated DNS failover).
RISKS: list at least 6 with impact and mitigation (cross-site link failure, split-brain vCenter authority, RPO breach from replication lag, full site power loss, missing management-plane backups, operator error during failover).
Map the 5 VCDX design quality dimensions (Availability, Manageability, Performance, Recoverability, Security) to your dual-site design, including trade-offs. For Security, incorporate the 9.1 posture: HTTPS-everywhere reverse proxy, Vault-managed secrets, Authentik SSO — and contrast with the 9.0-era open-services model.
Create a failover decision tree covering at least: link-only failure (do not fail over), complete site power loss (fail over), partial management-plane failure (manage from surviving components), Site A vCenter down with hosts and workloads up (restore vCenter, do not fail over), and a single host failure inside a site (vSphere HA restarts VMs, do not fail over).
Write (do not execute) a failover runbook with step timings and estimated total RTO; compare against your target.
Validation Gate
Check: All 9 steps completed: scenario, RCAR worksheet, four RCAR sections, quality matrix, decision tree, runbook with RTO estimate
Expected: Design documentation comprehensive enough for VCDX oral defense with detailed follow-up answers available for any section
Common Errors
Final Validation
A complete dual-site Holodeck 9.1 deployment (Site A and Site B, both VCF 9.1.0.0) is operational with validated cross-site routing, Technitium-based DNS cross-resolution, management plane visibility, and a production-ready DR design documented via RCAR.
✓ Two Holodeck instances running (Site A and Site B) with non-overlapping ranges → Get-HoloDeckInstance for each site reports a completed VCF 9.1.0.0 deployment
✓ Cross-site IP reachability in both directions → Management-plane pings succeed Site A <-> Site B
✓ Cross-site DNS via Technitium → FQDNs for each site resolve from the other site
✓ Both SDDC Managers show management domain Active → Inventory > Workload Domains healthy at both sites
✓ Both vCenters healthy; NSX transport nodes up at both sites → Green health indicators across both management planes
✓ RCAR design exercise completed → Requirements, Constraints, Assumptions, Risks, quality matrix, decision tree, and runbook documented
Cleanup / Restore
Snapshot: holodeck91-06-complete
• Take snapshots of all Holodeck VMs from both sites, named 'holodeck91-06-complete'
• Document the dual-site topology as actually deployed: IP addressing, FQDN mappings, cross-site routing, and every point where 9.1 diverged from this stub
• Archive the RCAR design exercise output for VCDX preparation
• If on a shared physical host: record this instance's VLAN ranges (-VLANRangeStart) and HoloRouter external IP in the shared registry so other instances, each with its own HoloRouter, avoid collisions
Design Reflection (VCDX)
Dual-site is the VCDX differentiator. A panelist probes: why dual-site vs single-site; how sites are decoupled operationally; latency-budget assumptions; split-brain handling; validated RTO/RPO. The 9.1-specific angles to defend: (1) dual-site auto-generation makes DR rehearsal a default rather than an advanced configuration — what does that change about your design process and your collision-avoidance controls? (2) inside each site, Distributed (VNA cluster) versus Centralized (NSX Edge cluster) networking is a design-decision axis — when would you model with a VNA cluster, and what does it change in the per-site failure domain (the inter-site HoloRouter path stays the same)?
Requirements
- Replicate application VMs Site A to Site B within a defined RPO
- Cross-site management plane reachability (IP + DNS) in both directions
- Per-site SDDC Manager lifecycle independence
- Failover RTO within the defined target including detection, DNS update, and power-on sequencing
- Encrypted, authenticated cross-site management traffic (9.1 HTTPS-everywhere posture)
Constraints
- Single physical host = shared failure domain for both sites (lab-only constraint)
- Single HoloRouter provides the inter-site routing (one VRF per site); it is a shared transit and DNS layer for both sites
- Cross-site latency in the lab (<2ms) does not represent WAN reality (50-150ms)
- Licensing limits on NSX federation, replication, and multi-domain configuration
- A HoloRouter serves exactly one Holodeck instance (one HoloRouter per instance; maintainer, GitHub Discussion #178): on a shared physical host each instance needs its own HoloRouter and its own -VLANRangeStart
Assumptions
- The auto-generated Site B config can be adjusted safely before deployment
- The Holodeck 9.1 recommended dual-site VCF 9.1 footprint (96 CPU, 1536 GB RAM, 5 TB disk) is available on the physical host; pre-checks only warn below it
- Technitium DNS supports the cross-site record configuration this design requires
- Manual DNS failover is acceptable to the business scenario chosen
Risks
- Cross-site link failure stalls replication — MITIGATION: latency/lag monitoring with explicit failover thresholds
- Split-brain vCenter authority post-failover — MITIGATION: runbook declares the sole authoritative site and isolation steps
- Auto-generated Site B collides with shared-host tenants — MITIGATION: pre-deploy allocation review and registry
- 9.0-era router workarounds applied blindly to the 9.1 stack corrupt router state — MITIGATION: treat all 9.0 fixes as unverified; re-diagnose from live logs
- Resource exhaustion during Site B bring-up destabilizes Site A — MITIGATION: continuous memory monitoring, conservative sizing, pause/resume plan
- Missing management-plane backups discovered during a drill — MITIGATION: scheduled SDDC Manager and vCenter backups stored off-instance
Self-Assessment Discussion Prompts
- 9.1 auto-generates the Site B configuration. What does 'DR by default' change about your design review process compared to 9.0's manual dual-site workflow?
- Where does the failure domain boundary sit in this lab vs production, and how does the choice of Distributed (VNA) vs Centralized (NSX Edge) networking inside a site move it — given that the inter-site router is unchanged?
- How do you prevent two vCenters from independently managing the same VM during a network partition?
- Compare stretched vSAN synchronous replication vs separate clusters with async replication for this scenario's RPO/RTO/cost trade-offs
- If cross-site latency were 50ms instead of <2ms, which components could no longer be stretched and why?
- The 9.1 router stack is HTTPS-everywhere with authenticated services. How does that change the security section of your dual-site defense compared to the 9.0-era open-services router?
References
- Holodeck 9.1 Release NotesTier 1 — Official
Official 9.1 release notes: dual-site auto-generation, Site B port-group fixes, VLAN range fixes, VNA cluster support, the 9.1 services stack and the mandatory Import-HoloDeckConfig -Site parameter. - Announcing the General Availability of Holodeck 9.1 (VMware Cloud Foundation blog, 1 July 2026)Tier 1 — Official
Official GA announcement: New-HoloDeckConfig now defaults to creating two config files (Site A and Site B) to avoid race conditions when deploying dual site in parallel, and Import-HoloDeckConfig gains a -Site parameter that loads the correct site config. It does not cover shared-host cautions; confirm those against vmware.github.io/Holodeck/9.1/. - Holodeck Toolkit GitHub RepositoryTier 1 — Official
Verify 9.1 dual-site documentation and cmdlet reference here once published. - VMware Cloud Foundation Multi-Site DocumentationTier 1 — Official
Locate the VCF 9.1 multi-site deployment guide for production-side design mapping. - Community blog — VCF Multi-Site and DR Design PatternsTier 3 — Background reading
Community deep dives on multi-site VCF and stretched cluster designs.