ADR-065: Windows Desktop OS fallback → VDI Linked Clone
Date: 2026-02-23 Status: Accepted
Context
Analysis of a real 1,483-VM LiveOptics export showed that 92 % of VMs (1,366)
fell through to the os_fallback confidence bucket, meaning no VM name pattern
matched and only the OS field was used. Of those, ~904 VMs had a Windows 10,
Windows 11, or Windows 7 guest OS.
The previous rule (priority 905) classified all Windows Desktop OS VMs as
Virtual Machines / VMware / Hyper-V / KVM - No Database, File nor Email
with a DRR of 5. While defensible, this misrepresents the actual workload:
desktop OS VMs in a datacenter are overwhelmingly VDI endpoints (Citrix PVS,
VMware Horizon linked clones, Citrix MCS).
Decision
Change the "Windows Desktop (OS fallback)" rule (priority 905) to classify as:
- Category:
VDI - Subcategory:
Linked Clone / PVS (Citrix)(DRR = 4)
The OS pattern (windows 10|windows 11|windows 7) is unchanged.
Additionally, add a new "VDI Generic" rule (priority 224, name-based) covering
explicit VDI infrastructure keywords: VDI, DESKTOP, RDS, UAG,
LOGINVSI, LOGINENTERPRISE.
Rationale
- In enterprise VMware environments, Windows 10/11 VMs running in a datacenter are almost exclusively VDI linked clones or MCS clones. Physical desktops are not inventoried in RVTools or LiveOptics.
- VDI Linked Clone (DRR=4) is a conservative estimate; Full Clone (DRR=8) and Instant Clone (DRR=6) are higher. Using the lower value maintains the pre-sales defensibility principle (ADR-005).
- This change improves the classification rate from ~8 % rule_match to ~75 % on the reference file, making the report far more actionable.
Consequences
- Positive: ~900 Windows Desktop VMs now get a more accurate workload category and DRR in typical enterprise files.
- Positive: Engineers reviewing VDI-heavy environments see a realistic capacity estimate, not a generic "Virtual Machines" bucket.
- Negative: Any genuine Windows 10/11 server-role VM (edge case) will be mis-classified as VDI; engineers can correct via the review grid.
- Pattern: OS-fallback rules should reflect the most probable datacenter usage of that OS, not the broadest possible category.