Virtual Desktop Platform
One person. One remote desktop. Prove it at home first, then design each cloud build separately.
Proxmox · prove access and recovery
A proposed single-user guest on the home server. Every device, route and restore remains to be verified.
1Reach the desktop
One person
Candidate devices
Private
connection
No publicdesktop portRoute & protocol: to decide
session
Proxmox home server
PlannedOne remote desktop
1 user · 1 sessionSynthetic demo files only
2Protect it two ways
Work files
File sync is not backupUser files
Planned test data
Versioned file backup
Tool, target & keys: to verify
Earlier version → new folder
Compare checksum; original kept
Whole desktop VM
Same-host copy ≠ off-host backupFull VM
Planned guest backup
Off-host VM backup
Destination & key recovery: to verify
New VM ID, network off
Console-only test; keep copy isolated
Two recovery layers can still share one failure domain. A shared NAS, its capacity, and independent key access must be reviewed.
3One step at a time — earn the next stage
Readiness check
- Owner authorises VM work
- Host room & rescue path
- Off-host backup approved
- OS · client · protocol chosen
Proxmox proof
First · planned- Build one desktop
- Test one client
- Test both restores
Acceptance
Needs evidence- Results & limits
- Recovery timings
- Repeatable runbook
Fresh AWS build
Only if accepted- New network & IAM
- Costs & backup
- Expiry & teardown
Azure comparison
Later- Linux VM desktop
- or Windows AVD
- Separate design & gates
Arrows show a proposed sequence. Cloud stages start only after the local proof is accepted; they are fresh builds, not transfers of the Proxmox VM.
Reusable design intent
Linux setup with cloud-init / Ansible, test method, evidence and runbook
Redesigned for each environment
Image · network · identity · licences · desktop access · backups · costs
Read the diagram
Readiness decisions & recovery detail
Before any guest is built
- Select one OS, client and protocol; do not assume a mobile app proves compatibility.
- Verify host capacity for both the pilot and restored copy, off-host storage, a rescue console, and permission for actual VM work.
- Keep encrypted-backup keys recoverable independently of the guest and its backup target.
Restore without a duplicate live identity
- Restore a new VM ID and disconnect its network before first boot.
- Check from the console; the restored hostname and machine identity must not collide with the original.
- Record failures and timings. Remove the test copy only under an agreed cleanup plan.
AWS · plan a private EC2 desktop
A fresh Linux build in a dedicated VPC. Region, size, protocol, access method and spending envelope are decisions to make.
1Reach a private cloud desktop
One person
One selected client
Re-test the chosen client against the cloud desktop stack.
Client & protocol: to testPrivate
access path
No publicdesktop port
VPN or reviewed overlay; private routing required
Access & egress: to designsession
AWS VPC
Proposed · private subnetEC2 Linux desktop
Fresh build · 1 userEncrypted EBS · no public IP
2Design both recoveries
Work files
Sync / versioning alone ≠ restore proofUser files
Synthetic data only
Versioned file repository
S3-backed option; tool & keys pending
Earlier file → new path
Compare checksum; original kept
EC2 & attached volumes
Snapshot creation ≠ recovery provenEC2 / EBS
Planned backup scope
AWS Backup plan
Vault, retention & KMS: to design
New EC2, isolated first
Quarantine subnet / SG selected before restore
Same-account backups share an account-compromise risk. Account separation, delete protection and independent key recovery remain decisions.
Estimate before provisioning
Region, EC2 hours, EBS, backup, access, endpoints / NAT, logs and traffic. A dated estimate is required; budget alerts are not a hard spending cap.
Give the pilot an expiry
Owner, scheduled stop and review date. Stopped compute can still leave billable storage, networking and recovery points.
Verify teardown
Inventory EC2, EBS, AMIs, snapshots, recovery points, IPs, access services and logs. Retain only explicitly agreed recovery evidence.
3Plan → prove → recover → decide
Entry gate
- Local recovery accepted
- Cloud scope authorised
- Cost estimate approved
- Expiry & cleanup agreed
AWS design
Decisions pending- VPC · route · IAM
- Image · keys · egress
- Backup & restore scope
Desktop proof
Planned- Fresh EC2 build
- One client / protocol
- Disconnect & reconnect
Recovery proof
Needs evidence- Earlier-file restore
- Isolated EC2 restore
- Timings & access checks
Decision & teardown
After acceptance- Actual costs & limits
- Retain / remove agreed
- Azure remains separate
AWS design decisions, restore caveats & sources
Decisions to settle before a build
- Choose image, architecture, instance size, protocol and a private client route; review software licensing and supported clients. Tablets remain pending.
- If Amazon DCV is selected, verify its EC2 licensing checks and required S3 reachability / IAM permissions; do not assume a private subnet has every required dependency.
- AWS Backup restores a new instance. Select quarantine networking and its IAM profile before restoring. User-data is not backed up; the original key pair is retained. Check identity and avoid contact with normal services until cleared.
- A security group is an allow-list. Remove unneeded outbound permissions and review AWS service-traffic exceptions; it is not a complete isolation boundary by itself.
Official references · checked 23 Oct 2026
- AWS Backup: restoring an EC2 instance
- VPC security-group rules and exceptions
- Amazon DCV licensing requirements
- AWS Budgets: tracking costs and alerts
These references describe service capabilities. They do not establish a deployed, licensed or tested configuration.
Azure · choose a desktop model first
Option A is a self-managed Linux VM. Option B is Windows Azure Virtual Desktop. They require separate access, identity, licensing and recovery designs.
1Option A · private Linux VM
One person
One chosen client
Validate the client against the chosen Linux desktop stack.
Client & protocol: to testPrivate
access path
No publicdesktop port
Private route and graphical protocol selected together
Route & identity: to designsession
Azure VNet
Proposed · private subnetLinux VM desktop
Fresh build · 1 userManaged disks · no public IP
2Plan recovery for Option A
Work files
Sync alone ≠ backupUser files
Synthetic data only
Versioned file repository
Blob-backed option; tool & keys pending
Earlier file → new path
Compare checksum; original kept
VM & managed disks
Recovery point ≠ restore provenLinux VM
Check supported disk / agent scope
Azure Backup plan
Vault, policy & key access: to design
Restore disks → isolated VM
Quarantine VNet / NSG before first boot
Subscription / account separation, retention and independent key access need review. A second recovery layer does not remove a shared account failure risk.
OR · Option B: Azure Virtual Desktop
A distinct managed desktop-access service to evaluate, if a Windows desktop is the goal.
Supported client
Windows App / supported client route; device coverage to verify
Entra identity
Tenant, entitlement, sign-in policy and supported join model: to design
AVD control plane
Workspace, app group and host pool type: to choose
Windows session host
Supported edition, licence eligibility, profiles and recovery: separate gates
Linux is not an AVD session-host option. The AVD lane needs its own network, profile and data-recovery design. Managed brokering does not prove workload backup or availability.
Compare complete costs
VM hours, disks, backup, access services, traffic and logs; add eligible Windows / AVD licensing and profile storage for Option B. No price is assumed.
Separate the evidence
A Linux VM result does not validate AVD. Compare the same tasks, client behaviour and recovery outcomes on whichever option is selected.
Plan expiry & cleanup
Agree a stop date and inventory. Review retained disks, IPs, vault recovery points, profiles, networking and logs before declaring teardown complete.
3Choose the model before the pilot
Comparison gate
- Earlier results reviewed
- Azure scope authorised
- One desktop model chosen
- Cost & expiry agreed
Azure design
Decisions pending- VNet · access · identity
- Linux image or AVD hosts
- Licence & key checks
Selected pilot
Future- One user / chosen client
- Fresh build & config
- Record access behaviour
Recovery proof
Needs evidence- Files / profiles in scope
- Isolated VM or host test
- Document recovery limits
Compare & close
After evidence- Cost & operational tradeoffs
- Keep / remove agreed
- Record teardown evidence
Azure model boundaries, recovery decisions & sources
Decisions to make explicit
- Choose one lane. Do not infer that Linux tools, guest images or clients transfer to Windows AVD.
- For the Linux plan, restore disks and establish quarantine networking before creating / booting the test VM. Review NSG defaults, effective rules, egress and network reachability.
- Azure VM backups cover supported attached disks; temporary disk data and memory state are not included. Review the support matrix and application consistency requirements.
- For AVD, check supported Windows editions, identity / join prerequisites, host pool model, licensing and client support. Design profile / user-data recovery separately.
Official references · checked 23 Oct 2026
- AVD prerequisites: operating systems, identity & licensing
- Azure Backup: VM and disk restore options
- Azure VM backup support matrix
- Azure VM backup FAQ: captured state and restore choices
Service documentation supports the planning choices; it does not establish implementation or recovery acceptance.