Day-2 operations

Protection and Recovery: DR Between Two VCF 9.1 Sites — Build Guide

This guide sets up VCF Protection and Recovery 9.1 for disaster recovery of workloads between two VCF 9.1 sites: a protected site where the workloads run, and a recovery site they fail over to. It covers the licence decision, choosing a replication type, deploying and pairing the appliances, replication, inventory mappings, recovery plans, and testing, failover and failback.

Not covered here:

  • Cyber / ransomware recovery and the clean room.
  • Local vSAN snapshots for operational recovery.
  • Multi-tenant DR through VCF Automation.
  • DR of the management components themselves (VCF Operations, VCF Operations for networks). Broadcom covers that in its own runbook, VCF Fleet Disaster Recovery, which uses the same appliances; see §12.

Evidence level: sourced from Broadcom TechDocs (the Protection and Recovery 9.1 documentation, pages last updated 2026-10-20, and its 9.1 release notes), the configmax tool and the Ports and Protocols tool (both read 2026-11-04). Not lab- or field-verified in this repo. Steps for recovery plans, mappings and failover are summarised from the TechDocs Disaster Recovery section; read the page for your build before running them.


The sequence, end to end

  1. Decide the licence: replication only, or orchestrated DR
  2. Names, then and now
  3. Choose the topology and the replication type
  4. Plan: IPs, DNS, capacity, networks, firewall (incl. static routes for a routed replication subnet)
  5. Deploy the appliance on both sites
  6. Pair the sites
  7. Configure replication
  8. Inventory mappings
  9. Protection groups and recovery plans
  10. Test, migrate, recover, reprotect
  11. Monitor and back up the configuration
  12. The management components: VCF Fleet Disaster Recovery
  13. Upgrading from 9.0.x
  14. Limits, caveats and known issues

Steps 1–4 are planning. The licence decision in step 1 sets what you can build, so take it first.


1. Decide the licence

From Welcome to Protection and Recovery and, for Site Recovery Manager, KB 461219:

You wantLicenceNotes
Replication with an RPO above 5 minutes, recovered by handIncluded with VCF”replicate with RPO higher than 5 minutes using standalone vSphere Replication”
Local vSAN snapshotsIncluded with VCFOut of scope here
Orchestrated DR: recovery plans, non-disruptive tests, planned migration, failover, reprotectIncluded with an active VCF or VVF subscription since 2 November 2026; not with perpetual licencesSite Recovery Manager only; see 1.1 for the key you need until 9.1.3
An RPO under 5 minutes with vSAN snapshots and replicationAdvanced Cyber Compliance (per core)Not changed by KB 461219
Cyber recovery (clean room, isolated recovery)Advanced Cyber ComplianceOut of scope here; not changed by KB 461219

Replication without a recovery plan protects the data, not the service: failover then means registering and powering on every VM by hand, in the right order, with the right networks. For anything beyond a handful of VMs, plan for orchestrated DR: with a subscription it no longer costs extra.

1.1 Site Recovery Manager is now included

The sources disagree, so here is the timeline:

DateSourceSays
Last updated 20 October 2026Welcome to Protection and Recovery”If you want to use disaster recovery without cyber recovery, you can purchase a standalone VMware Site Recovery Manager license for VCF, VMware vSphere Foundation (VVF) or legacy vSphere software”
Effective 2 November 2026KB 461219”VMware Site Recovery Manager (SRM)—formerly VMware Live Site Recovery (VLSR)—is no longer sold standalone. SRM is now included in VMware Cloud Foundation (VCF) and VMware vSphere Foundation (VVF). All existing active VCF and VVF subscription customers are entitled to SRM.”

The KB is the newer source; follow it. What it means in practice:

  • Who: active VCF and VVF subscriptions only. “Customers with perpetual VMware Cloud Foundation licenses, with or without active Support and Subscription (SnS), are not entitled” to it.
  • What: only Site Recovery Manager, the orchestration layer. The KB says nothing about vSAN replication with an RPO under 5 minutes or cyber recovery; those still need Advanced Cyber Compliance.
  • From 9.1.3: the entitlement comes with the subscription automatically (the KB expects 9.1.3 “between February 2027 - April 2027”), and then needs the protection and recovery appliance at 9.1.3 and vCenter at 9.1 or later.
  • Until then: a term-limited key. Up to 10,000 protected VMs, expires 2 May 2027. Get it from the Support Portal, not the Licensing Portal: My Downloads → VMware Cloud Foundation (or VMware vSphere Foundation) → VMware Cloud Foundation 9 → version 9.1 or later → search Protection and Recovery → View Group → under Protection and Recovery (Key) → View Group → download the PDF that contains the key.
  • The key replaces existing ones. It “cannot be split, combined, upgraded, or downgraded”, so you replace any existing SRM keys in vCenter with it, including in an existing site pair. It is a version 8.1 key, which “works on all supported SRM versions”.
  • Plan the upgrade before 2 May 2027: move the appliances to 9.1.3 and vCenter to 9.1 or later before the key expires, or DR stops working.

2. Names, then and now

The 9.1 release notes, verbatim: “VMware Live Recovery has been renamed and integrated into VMware Cloud Foundation (VCF) as Protection and Recovery.” (9.1 release notes). Older documents, licences and support cases still use the old names.

Older nameVCF 9.1 nameWhat it is
VMware Live RecoveryProtection and RecoveryThe product family
VMware Live Site Recovery / Site Recovery Manager (SRM)Protection and recovery serverOrchestrates recovery plans. Site Recovery Manager is still the name of its licence
vSphere Replication appliancevSphere Replication, inside the applianceHost-based replication
vSAN Data ProtectionvSAN snapshots and replicationSnapshots and vSAN-to-vSAN ESA replication
Separate SRM and vSphere Replication appliancesOne protection and recovery applianceSince 9.1 (converged from 9.0.2.3 and earlier)

3. Choose the topology and the replication type

3.1 Topology

Shared Recovery Site lists what’s supported: 1:1, fan-in / shared recovery site (N:1), fan-out / shared protected site (1:N) and many-to-many (N:N). This guide follows the 1:1 case: one protected and one recovery site.

For anything other than 1:1:

  • Each protection and recovery pair shares an extension ID, set when you install. “Each pair of protection and recovery instances in the shared recovery site configuration must use a different protection and recovery extension ID.” If the first pair used the default ID, every later pair needs its own custom ID. Choose the IDs before the first deploy.
  • The extra instances at the shared site are scale-out appliances, which hold only the protection and recovery server. You can deploy up to nine per vCenter (Deploy Protection and Recovery).

3.2 Replication type per workload

TypeSourceTargetWhen
vSphere Replication (enhanced)Any datastoreAny datastoreThe default. Host-based, per VM, no storage-array dependency
vSAN replication (vSAN snapshots and replication)Any storage, since 9.1vSAN ESA cluster onlyImmutable snapshots, daily / weekly / monthly schedules, tag-based protection groups; RPO under 5 minutes needs Advanced Cyber Compliance
Array-based replicationArray LUNs / volumes through a Storage Replication Adapter (SRA)The paired arrayLarge datastores already replicated by the array; check the array and SRA in the compatibility guide

Two rules that shape the design:

  • vSAN replication needs a vSAN ESA recovery cluster. From Create Site and Cluster Pairing: “you must pair two clusters, one on the local site and one on the remote site. The target cluster must be a vSAN ESA cluster.” If the recovery cluster isn’t vSAN ESA, you use vSphere Replication or array replication.
  • vVols are on the way out. “Protection and Recovery 9.1 is the last release that supports vSphere Virtual Volumes. It is recommended to migrate your Virtual Volumes replications to array-based replication.” (9.1 release notes). Don’t design new DR on vVols.

Array-based replication doesn’t support NFS 4.1 datastores. vSphere Replication does (9.1 release notes).


4. Plan: IPs, DNS, capacity, networks, firewall

4.1 Appliances

One appliance per vCenter, on both sites. From System Requirements:

AppliancevCPURAMDiskNIC
Protection and recovery appliance824 GBabout 800 GB (5 disks)1 Gbit
Scale-out appliance (only for > 5,000 VMs or a shared site)48 GBabout 110 GB (5 disks)1 Gbit
  • FQDN and IP per appliance, with forward and reverse DNS. TechDocs: “Use fully qualified domain names (FQDN) whenever possible.”
  • ESX 9.0 or later for VCF 9.1.x (Deploy Protection and Recovery).
  • The recovery-site vCenter and appliance must not depend on the protected site: “During a disaster recovery, protection and recovery and vCenter must be up and running on the recovery site.” (Deployment Models). That includes DNS, NTP, AD and the identity provider at the recovery site.

Example values: pnr01.sfo.rainpole.io (protected), pnr01.lax.rainpole.io (recovery).

4.2 Capacity at the recovery site

Size the recovery cluster for the VMs you’ll fail over, plus the space the replicas and their snapshots use. During a test, the test VMs run next to the replicas, so plan for both.

4.3 Networks at the recovery site

Decide how recovered VMs get their network, before you set up mappings:

  • Re-IP: the VMs come up in different subnets at the recovery site, and the recovery plan changes their IPs (IP customisation). Needs DNS updates and clients that use names.
  • Same subnets: the recovery site has the same subnets, as NSX segments or VPC subnets that are kept down or isolated until failover. No re-IP, but the routing has to move at failover.
  • A test network that is isolated from production, for non-disruptive tests.

4.4 Firewall flows

From the Ports and Protocols tool (products Protection and Recovery and vSphere Replication; the 9.1.0 and 9.1.1 listings are identical, read 2026-11-04). The picture shows the same flows per site; the table below is the reference.

Firewall flows between the protected site and the recovery site: appliance to appliance on 443 and 5480, each appliance to the remote vCenter on 443, protected ESX to recovery ESX on 32032 for enhanced replication data, protected ESX to the recovery vSphere Replication server on 31031 and 32032 for legacy replication, and the intra-site flows per site

SourceDestinationPort(s)ProtoPurpose
Appliance (protected)Appliance (recovery)443TCPSite-to-site management
ApplianceLocal and remote vCenter443TCPManagement
Local applianceRemote appliance5480TCPvSphere Replication management
ESX (protected site)ESX (recovery site)32032TCPEnhanced vSphere Replication traffic
ESX (protected site)vSphere Replication server (recovery site)31031, 32032TCPLegacy vSphere Replication traffic
vSphere Replication server (recovery)ESX (recovery, same site)902TCP + UDPIntra-site replication management
AppliancevSphere Replication server8123TCPIntra-site management
ApplianceLocal vCenter9084TCPIntra-site
Recovery-site applianceRecovery-site ESX902, 443TCPInternal services
Browser / admin workstationAppliance5480, 443TCPAppliance management, UI
Appliancevcsa.vmware.com443TCPDiagnostics (optional)

Enhanced replication sends the data from the protected site’s ESX hosts straight to the recovery site’s ESX hosts. The rule you most often need to request is therefore between the two sites’ ESX networks, not to an appliance. Add these flows to 07-firewall-ports.md §H.

4.5 A replication network in its own subnet: static routes

If replication runs over its own VLAN and subnet at each site, and the two sites’ replication subnets are different and routed, you need static routes. Tagging the VMkernel adapter is not enough.

Why: vSphere Replication runs only on the ESX default TCP/IP stack. “You cannot enable the vSphere Replication traffic on a custom TCP/IP stack on the ESXi host” (Isolating the Network Traffic of vSphere Replication, 9.0; KB 432747). The default stack has one default gateway, normally the management gateway. The vSphere Replication / vSphere Replication NFC tag says which vmk the service uses; it does not change routing. A destination outside the local replication subnet goes to the default gateway, so the traffic leaves via management (KB 430898). Broadcom’s answer is tagging and static routes (KB 432747).

Field-verified: setting Override default gateway for this adapter on the replication vmk (VMkernel adapter → Edit → IPv4 settings) did not fix it; replication still needed static routes. Broadcom doesn’t mention that option for replication either.

Where the routes go:

WhereRoute toViaNotes
Every ESX host, protected sitethe recovery site’s ESX replication subnet and the recovery site’s appliance replication IP (both TCP 32032)the local replication gatewaythe gateway must be in a subnet where the host has a vmk (KB 308786); why both: see below
Every ESX host, recovery sitethe protected site’s replication subnetthe local replication gatewaythe reverse routes, for reprotect and failback (isolation page)
Appliance replication NIC, both sitesthe other site’s ESX replication subnet(s)the replication NIC’s gateway”Source VRMS appliance must have routes for the target ESXi cluster” and the reverse (KB 384776); missing routes show as error 31031 (KB 395034)

ESX hosts. The vSphere Client has no screen for host static routes (only the stack’s default gateway, which would move management too). Use esxcli or PowerCLI (PowerShell 7 and VCF PowerCLI on the jump host: 42-jump-host-powershell.md):

# per host
esxcli network ip route ipv4 add -n 172.28.20.0/24 -g 172.27.20.1
esxcli network ip route ipv4 list
vmkping -I vmk3 172.28.20.11      # replication vmk -> remote host's replication vmk
# whole cluster (PowerCLI)
Get-Cluster sfo-w01-cl01 | Get-VMHost |
  New-VMHostRoute -Destination 172.28.20.0 -PrefixLength 24 -Gateway 172.27.20.1 -Confirm:$false

Routes are per host: add them to your host-prep for every host you add later (cluster expansion, host replacement), or new hosts replicate over management.

Appliance. The appliance management interface (VAMI) supports only one default gateway (KB 384776). Add routes in the NIC’s systemd-networkd file (Configure a Static Route on an Additional VM Network Adapter, 9.0):

Enhanced replication talks to the appliance too. The data goes ESX to ESX, but the source host’s hbr-agent first logs in to the recovery site’s replication appliance on TCP 32032 (the brokered server). KB 393268 shows it failing as “Failed to login to brokered server: ##.##.##.## : 32032”, alongside “Ensure ports 32032 are open between source and destination ESXI hosts”. So a healthy enhanced replication has two kinds of 32032 sessions from a source host:

SessionWhatVolume
source ESX → recovery-site appliancelogin / brokering (control)small
source ESX → recovery-site ESXthe replication datathe bulk

A session to the appliance on 32032 is therefore not proof of legacy replication. Only appliance sessions and no sessions to recovery-site hosts means the data is not going host to host: either a legacy replication, or the host-to-host path is broken (32032 not open between the hosts, KB 392287; or a missing route). Legacy replication uses the appliance as the data target (31031, 32032).

# /etc/systemd/network/10-eth1.network   (eth<N> = the replication NIC)
[Route]
Gateway=172.27.20.1
Destination=172.28.20.0/24

Then systemctl restart systemd-networkd and check with ip r. Add GatewayOnLink=yes in [Route] if the route doesn’t validate. Appliance upgrades can drop these routes: back them up before an upgrade and re-add them afterwards (KB 384776).

Tags: which vmk replication uses

There is no setting in Protection and Recovery that picks a vmk. ESX picks it from the VMkernel service tags (vSphere Client: Host → Configure → VMkernel adapters → Edit → Enabled services):

SiteTag on the replication vmkEffect
Source (protected)vSphere Replicationoutgoing replication data leaves via this vmk (source)
Target (recovery)vSphere Replication NFCincoming data is written to the datastore via this vmk (target)
  • Tag both services on both sites: after reprotect the roles swap.
  • No tag means replication falls back to vmk0 (KB 384776).
  • The tag picks the source vmk; the routing table still picks the gateway. Tag and route together, as above.
  • With enhanced replication (ESX to ESX, TCP 32032) the tags and routes are all there is. For legacy replication, or to separate the appliance’s own traffic, give the appliance a second vNIC in the replication network (a third for NFC only if you want) (isolation page), set that NIC’s IP as the address for incoming replication traffic in the appliance management interface (https://<appliance>:5480; older vSphere Replication called the field IP Address for Incoming Storage Traffic, the 9.1 label is not checked here), and add its routes as shown above.
  • Changing tags or routes affects new sessions. Check with step 5 below whether the local IP has moved to the replication vmk; if a replication still shows vmk0, a pause/resume or a sync starts a new session (not tested here, not in Broadcom’s docs).

Verify from the ESX Shell (SSH). The tools are standard in the ESX shell; option sets vary by build, so check traceroute --help / nc -h.

# 1. Routes and vmk IPs
esxcli network ip route ipv4 list
esxcli network ip interface ipv4 get

# 2. Which path does ESX itself pick? (no -i/-s = the default-stack routing)
traceroute 172.28.20.11
traceroute -i vmk3 172.28.20.11            # forced out of the replication vmk

# 3. Reachability and MTU via the replication vmk
vmkping -I vmk3 172.28.20.11
vmkping -I vmk3 -d -s 8972 172.28.20.11    # jumbo frames end to end

# 4. Is the replication port open? (from the replication vmk's IP)
nc -z -v -w 3 -s 172.27.20.21 172.28.20.11 32032

# 5. Which local IP do live replication sessions use?
esxcli network ip connection list | grep 32032

Reading the results:

  • A plain traceroute whose first hop is the management gateway means the static route is missing or wrong.
  • nc succeeded = path and port open; timeout = firewall, missing route or asymmetric return path; refused = path works, nothing listens. The ESX firewall filters outbound too: if nc times out, check esxcli network firewall ruleset rule list | grep 32032 before blaming the network.
  • In step 5, a local address equal to the vmk0 IP means replication still runs over management; after the route is in place it should be the replication vmk’s IP.
  • Also in step 5, look at the remote addresses: for an enhanced replication expect one to the recovery-site appliance (broker) and ones to recovery-site ESX hosts (data). Test both paths: nc -zv -s <repl-vmk-ip> <recovery-esx-repl-ip> 32032 and the same to the appliance’s replication IP.

5. Deploy the appliance on both sites

Protection and Recovery isn’t deployed from VCF Operations: TechDocs documents only the OVF wizard, and neither VCF Operations release note for 9.1.0 or 9.1.1 lists it as a lifecycle-managed component. Per Deploy Protection and Recovery:

  1. Download the OVA (or the ISO, and mount it) from the Broadcom Support Portal. For a shared site you also need the scale-out OVA.
  2. In the vSphere Client on the protected site: right-click a host → Deploy OVF Template → the OVA (Protection-and-recovery-9.1.x.x.<build>_OVF10.ova) or an online URL.
  3. Name and folder (the name must be unique in the folder), cluster or host, EULA, datastore and storage policy, network.
  4. Customize template: network settings (DHCP or static), passwords, and SSH if you want it.
  5. Finish, and repeat on the recovery site.

Then in the appliance management interface (https://<appliance>:5480), connect each appliance to its own vCenter. If the vCenters are in Enhanced Linked Mode, pick the right one from the list, because you’ll see both (Deployment Models). For a shared site, set the extension ID here.

You can deploy only one protection and recovery appliance per vCenter, plus up to nine scale-out appliances (Deploy Protection and Recovery). The appliance uses its embedded vPostgres database only. Once both sites are configured, the protection and recovery plug-in and the vSAN snapshots and replication plug-in appear in the vSphere Client (Deploying and Configuring the Appliance).


6. Pair the sites

The recommended path, per Create Site and Cluster Pairing (you must be in the vCenter Administrators group):

  1. vSphere Client → the vCenter → Configure → Protection and Recovery → Pair Sites.
  2. Enter the remote vCenter host name and credentials → Next.
  3. Accept the vCenter certificate, then the protection and recovery certificate.
  4. Review → Pair.
  5. For vSAN replication, also pair the clusters: one local, one remote. The remote cluster must be vSAN ESA.

If you only use vSphere Replication or array-based replication, you can pair from the protection and recovery UI instead.

For two-factor authentication (RSA SecurID or Smart Card), the vCenters on both sites must be in Enhanced Linked Mode and direct replication partners, and you pair with the vCenter administrator credentials (9.1 release notes).


7. Configure replication

vSphere Replication, per VM or group of VMs:

  • Pick the target datastore and storage policy at the recovery site.
  • Set the RPO: above 5 minutes is included with VCF, under 5 minutes needs Advanced Cyber Compliance.
  • Point-in-time instances, if you need to roll back further than the last replica.
  • Quiescing, if the guest supports it.

Seeding: 9.1 can start a new replication from an existing replica and send only the changes, which saves the full first sync over the WAN (9.1 release notes).

vSAN replication:

  • Protection groups with snapshot schedules (daily / weekly / monthly) and retention.
  • Immutable snapshots where you need them.
  • Membership by VM tag, so new VMs with the tag are protected without a manual step.

Array-based replication:

  • Install the array vendor’s SRA on both appliances.
  • Register the array managers and enable the array pairs; the replicated datastores then appear.

Upgraded environments: after an upgrade to 9.1, every vSphere Replication replication must be reconfigured to enhanced replication (9.1 release notes).


8. Inventory mappings

Mappings tell the recovery site where things go. Set them once per site pair, in both directions if you’ll fail back:

MappingMapsWatch out for
NetworkProtected port group / segment → recovery port group / segmentNSX segments and VPC subnets at the recovery site must exist first
Test network→ an isolated network for testsTests must never touch production
FolderVM folders
ResourceClusters / resource pools
Storage policyVM storage policiesNeeded for vSAN and vSphere Replication targets
Placeholder datastoreWhere placeholder VMs live at the recovery siteSmall, but must exist on each site for failback

9. Protection groups and recovery plans

  • Protection group: a set of VMs protected together, by one replication type. Up to 1,500 VMs per group and 500 groups in total (configmax, see §14).
  • Recovery plan: one or more protection groups, plus:
    • priority groups (1 to 5 in Site Recovery Manager) and VM dependencies, so databases start before the application servers that use them
    • IP customisation for re-IP recovery (§4.3)
    • scripts and prompts at the points where people have to act (DNS change, load balancer, a sign-off)

Build one recovery plan per application or per business service, so you can test and fail over one service without the others. Keep a site-wide plan for a full disaster. 250 recovery plans are allowed, with up to 2,000 VMs and 250 protection groups per plan.


10. Test, migrate, recover, reprotect

Recovery plans run in four ways. This is the long-standing Site Recovery Manager workflow, and the 9.1 server keeps it. The 9.1 pages sit under Disaster Recovery → Using Protection and Recovery (Disaster Recovery); read the procedure for your build before running one.

OperationProtected siteWhat happens
TestKeeps runningRecovers copies of the VMs into the test network; Cleanup removes them afterwards
Planned migrationUpSyncs the latest changes, shuts the VMs down cleanly and recovers them. No data loss; stops on errors
Disaster recoveryDown or unreachableRecovers from the last replica; carries on past errors on the protected side
ReprotectBack upReverses the direction, so the old recovery site becomes the protected site

Failback is a reprotect followed by a planned migration back.

A routine that makes failover work when it’s needed:

  1. Test after every change to a recovery plan, and on a schedule (for example quarterly), and keep the test report.
  2. Run a planned migration at least once before go-live, during a maintenance window.
  3. After a real failover, reprotect as soon as the old site is back, so the workloads are protected again.

At most 10 recovery plans run at once, and 2,000 VM recoveries can start at the same time (configmax).


11. Monitor and back up the configuration

11.1 Monitor

VCF Operations has Protection and Recovery dashboards (Protection and Recovery Dashboards) and management packs for the protection and recovery server, vSphere Replication and vSAN snapshots and replication (see the release notes index). Watch for:

  • replications not meeting their RPO
  • recovery plans whose last test is old or failed
  • placeholder VMs missing at the recovery site

Automation: 9.1 is the last release with the VCF Operations orchestrator plug-ins for protection and recovery and for vSphere Replication. New automation should use the Protection and Recovery REST API Gateway and the DR REST plug-in (9.1 release notes).

11.2 Back up the configuration (export / import)

The appliance has no built-in scheduled backup of its own configuration, the way SDDC Manager, vCenter or NSX do. What Broadcom gives you is the Configuration Import/Export Tool (Exporting and Importing Protection and Recovery Configuration Data). It ships with the appliance in /opt/vmware/impex and is also available as a standalone .jar.

What it exports (to one XML file): version and build, local and remote site names, inventory mappings, placeholder datastores, advanced settings, array managers with SRA information, protection groups, recovery plans, power-on and power-off settings and shutdown actions. Not in that file: the replications themselves. vSphere Replication has its own export tool (/opt/vmware/vr-impex-tool), see Exporting and Importing Replication Configuration Data (9.0.3 page; no 9.1 page found yet).

Prerequisites:

  • Export: a working site pair, with the appliance up on both sites.
  • Import: a clean installation, registered with the same vCenter or one with the same inventory. Import the vSphere Replication data first, then the Protection and Recovery data.
  • Standalone tool: Java 21.0.x or later on 9.1 (the 9.0 docs say 11.0.x), and JAVA_HOME set. Use a service account or SSO administrator credentials; solution-user authentication is not supported.

Three ways to export:

RouteHowCredentials
UIvSphere Client → Site Recovery → Open Site Recovery → site pair → View Details → Summary → Export/Import SRM Configuration → Export → Download (UI export)your session
Script on the applianceAs root (su): sh /opt/vmware/impex/bin/export.sh, which writes to /opt/vmware/impex/exports/ (script export)none (script generated during pairing)
Standalone .jarjava -jar import-export.jar --exportInteractive (or --export with options, --exportPath <dir>) (syntax)service account or SSO admin

Automate it

The export is manual by default, but the documented way to automate it is a cron job on the appliance that runs the same script (cron export). As root on the appliance:

su
crontab -e
# hourly, Broadcom's example:
0 * * * * /usr/bin/sudo /bin/bash /opt/vmware/impex/bin/export.sh

Daily is usually enough; export after every recovery-plan change as well. Two things the cron job does not do for you:

  1. Get the file off the appliance. The exports land in /opt/vmware/impex/exports/ on the appliance you would be restoring. Copy them somewhere else, for example a second cron line that scps the newest file to the SFTP backup target from 08-backup-target.md, or pull them from that server on a schedule.
  2. Prune old files. Add a cleanup (for example find ... -mtime +14 -delete) so the exports directory doesn’t fill the appliance disk.

Run it on both sites: each appliance holds its own side of the pair. For vSphere Replication, schedule its tool the same way if you rely on it.

Untested here. The cron route is documented on the 9.0 pages; the 9.1 parent page describes the same tool in the same location. A cron entry or extra script on the appliance may not survive an appliance upgrade or redeploy, so check it after every upgrade. Test a restore (import into a clean appliance) at least once.

Restore (UI or standalone):

  • UI: on a clean appliance, Summary → Export/Import SRM Configuration → Import, browse to the XML. If a stage fails, download the CSV report (UI import).
  • Standalone, from /opt/vmware/impex: java -jar import-export.jar --importInteractive --path <exported.xml>. By default it retries VM recovery settings 5 times with a 10 000 ms delay; raise with --delay / --retries (import).
  • 9.1 caveat: the import applies the recovery settings of a protected VM only once, even if the VM is in several recovery plans. Check those VMs in every plan after an import (9.1 release notes).

12. The management components: VCF Fleet Disaster Recovery

This guide protects workloads. Recovering the fleet’s own components after an instance or site failure is a separate Broadcom runbook, VCF Fleet Disaster Recovery. It uses Protection and Recovery to fail over and fail back VCF Operations and VCF Operations for networks to another VCF instance in the same fleet. It has PowerShell automation through the VMware.CloudFoundation.InstanceRecovery module.

If you do both, they share the same appliance pair and limits: plan recovery plans, protected-VM counts and the recovery-site capacity for both together.

The 9.1 P&P workbook’s hidden Site Protection & DR sheet comes from the older validated design for protecting the management applications (separate vSphere Replication and VMware Live Recovery appliances, Aria naming). It doesn’t cover workload DR, and doesn’t match 9.1’s single appliance. Don’t transfer answers into it.


13. Upgrading from 9.0.x

From Convergence and Upgrade: “When using version 9.0.2.3 and earlier, you must converge to the combined protection and recovery appliance. Versions 9.0.3 and later support in-place upgrade.”

  • Converging from VMware Live Site Recovery 9.0.2.3, vSphere Replication 9.0.2.3 or vSAN Data Protection 8.0 U3 starts by deploying the 9.1 appliance on both sites (9.1 release notes).
  • All vSphere Replication replications move to enhanced replication.
  • Planned migration of array-replication groups from a 9.1 site to a 9.0.3 site isn’t supported, so upgrade both sites in the same window.
  • Check the paths in the Interoperability Matrix → Upgrade Path.

14. Limits, caveats and known issues

14.1 Scale limits

From configmax, product Protection and Recovery. 9.1.0 and 9.1.1 are the same except for vSAN replication.

Limit9.1.09.1.1
Protected VMs, total5,0005,000
VMs protected with vSphere Replication5,0005,000
VMs protected with array-based replication5,0005,000
VMs protected with vSAN replication1,0005,000
VMs protected with vVols replication500500
VMs on array replication with stretched storage1,0001,000
VMs per protection group1,5001,500
Protection groups, total500500
Recovery plans250250
Protection groups per recovery plan250250
VMs per recovery plan2,0002,000
Replicated datastores (array-based)500500
Recovery plans running at once1010
VM recoveries started at once2,0002,000

Above 5,000 protected VMs you add scale-out appliances (§3.1).

14.2 Caveats

From the 9.1 release notes, Caveats and Limitations. Check every protected VM against these:

CaveatEffect
VMs with persistent memory (PMem)Can’t be protected
Disks larger than 62 TBNo replication, and the VM won’t power on
4Kn native datastoresNot supported for vSphere Replication
Shared VMDKsCan’t be replicated
Storage policies with IO filtersNot allowed on source or target
Disk resize not a multiple of 512 bytesReplication fails until fixed
Encrypted VMsNeed vSphere 7.0 U2c or later
Uni-processor FTTurned off on the recovered VM
Flash Read CacheDisabled after recovery and the reservation set to 0; note it first
Replications within one vCenter (ROBO)Not supported with VMware Live Recovery activated
Array-based replication on NFS 4.1Not supported (vSphere Replication is)

References

ITQ – What's Next. VCF 9.1 deployment-planning and operations helper material. © 2026 Paul van Dieen. hollebollevsan.nl