Version: Next

Task 02: VMFleet Storage Performance Testing

DOCUMENT CATEGORY: Runbook SCOPE: Storage performance baseline PURPOSE: Deploy VMFleet, run storage tests, create performance baseline for customer handover MASTER REFERENCES:

Microsoft VMFleet GitHub

VMFleet 2.0 Quick Start Guide

VMFleet Workload Profiles

Status: Active VMFleet Version: 2.1.0.0 (April 2024)

Overview

VMFleet is Microsoft's official storage load generation tool for Storage Spaces Direct (S2D) environments. It deploys a fleet of VMs across all cluster nodes, each running DiskSpd to generate I/O load. This creates a realistic, distributed storage workload to validate performance and establish baselines.

Monitoring During Testing

VMFleet includes a built-in real-time dashboard (Watch-FleetCluster). For historical data and visualization, use the Azure Local Insights workbook configured in Phase 18: Monitoring & Observability. The Insights workbook provides IOPS, throughput, latency, and health metrics collected via Azure Monitor.

Purpose of This Step

Validate storage performance under load across all nodes
Document baseline metrics (IOPS, throughput, latency) for customer handover
Verify storage health under stress conditions
Generate reports for consolidated validation package

Maintenance Window Required

VMFleet generates significant I/O load. Run during a maintenance window when no production workloads are active.

Prerequisites

Infrastructure health validation completed (Step 1)
Windows Server 2022 Core VHD available (sysprepped)
At least 100GB free space on each node's Fleet volume
Maintenance window scheduled (4-6 hours)
No production VMs running during test

Variables from variables.yml

Variable Path	Type	Description
`platform.cluster_name`	String	Cluster name used in report headers
`compute.nodes[].name`	String	Node hostnames for per-node fleet VM distribution
`storage.pool_name`	String	Storage pool name for volume creation

Report Output

All results are saved to:

\\<ClusterName>\ClusterStorage$\Collect\validation-reports\
├── 02-vmfleet-storage-baseline-YYYYMMDD.csv
└── 02-vmfleet-storage-summary-YYYYMMDD.txt

Part 1: Install VMFleet Module

1.1 Install from PowerShell Gallery

# Install VMFleet module (current version: 2.1.0.0)
Install-Module -Name VMFleet -Force -Scope AllUsers

# Verify installation
Get-Module -Name VMFleet -ListAvailable
Import-Module VMFleet

# Check available commands
Get-Command -Module VMFleet

Expected output: Module version 2.1.0.0 (released April 2024) with commands including:

Install-Fleet - Create folder structure and copy binaries
New-Fleet - Deploy fleet VMs
Start-Fleet / Stop-Fleet - Control fleet operations
Remove-Fleet - Cleanup fleet VMs
Watch-FleetCluster - Real-time performance dashboard
Measure-FleetCoreWorkload - Run 4 pre-defined workload profiles
Start-FleetSweep - Run custom workload profiles
Get-FleetVolumeEstimate - Calculate recommended volume sizes

VMFleet Version History

Version	Release Date	Notes
2.1.0.0	April 2024	Current - Arc VM support, bug fixes
2.0.2.2	Jan 2022	Previous stable
2.0.0.0	Sept 2021	Major rewrite with `Measure-FleetCoreWorkload`

1.2 Verify DiskSpd Availability

VMFleet uses DiskSpd internally. Verify it's accessible:

# DiskSpd is bundled with VMFleet, but verify
$DiskSpd = Get-Command diskspd.exe -ErrorAction SilentlyContinue
if (-not $DiskSpd) {
 Write-Host "DiskSpd will be deployed by VMFleet during installation"
}

Part 2: Prepare Storage Volumes

2.1 Calculate Recommended Volume Sizes

VMFleet includes a helper command to calculate optimal volume sizes based on your cluster configuration:

# Calculate recommended volume sizes for your cluster
Get-FleetVolumeEstimate

Example output:

MirrorType VolumeSize Description
---------- ---------- -----------
2-way Mirror 500GB For 2-node clusters
3-way Mirror 350GB For 3+ node clusters
Mirror-Accel 400GB For mirror-accelerated parity

2.2 Create Collect Volume

The Collect volume stores VMFleet configuration, results, and the template VHD. Create it on shared storage visible to all nodes.

# Create Collect volume (single volume, accessible from all nodes)
# Recommended: 200GB minimum for VHD + results

$CollectVolumeName = "Collect"
$CollectVolumeSize = 200GB

# Check if volume already exists
$ExistingVolume = Get-ClusterSharedVolume | Where-Object { $_.SharedVolumeInfo.FriendlyVolumeName -like "*$CollectVolumeName*" }

if (-not $ExistingVolume) {
 # Create the virtual disk and volume
 New-Volume -StoragePoolFriendlyName "S2D on $((Get-Cluster).Name)" `
 -FriendlyName $CollectVolumeName `
 -FileSystem CSVFS_ReFS `
 -Size $CollectVolumeSize
 
 Write-Host "Created Collect volume: $CollectVolumeSize" -ForegroundColor Green
} else {
 Write-Host "Collect volume already exists" -ForegroundColor Yellow
}

# Verify volume path
$CollectPath = "C:\ClusterStorage\$CollectVolumeName"
Test-Path $CollectPath

2.3 Create Per-Node Fleet Volumes

Each node needs its own Fleet volume to host VMFleet VMs. Use the size from Get-FleetVolumeEstimate:

# Get cluster nodes
$Nodes = (Get-ClusterNode).Name

# Use volume size from Get-FleetVolumeEstimate or manual calculation
# Rule of thumb: ~10GB per fleet VM + overhead
$FleetVolumeSize = 500GB # Adjust based on VM count and resiliency

foreach ($Node in $Nodes) {
 $FleetVolumeName = "Fleet-$Node"
 
 # Check if volume exists
 $ExistingFleet = Get-ClusterSharedVolume | Where-Object { 
 $_.SharedVolumeInfo.FriendlyVolumeName -like "*$FleetVolumeName*" 
 }
 
 if (-not $ExistingFleet) {
 # Create Fleet volume for this node
 New-Volume -StoragePoolFriendlyName "S2D on $((Get-Cluster).Name)" `
 -FriendlyName $FleetVolumeName `
 -FileSystem CSVFS_ReFS `
 -Size $FleetVolumeSize
 
 Write-Host "Created Fleet volume for $Node" -ForegroundColor Green
 } else {
 Write-Host "Fleet volume for $Node already exists" -ForegroundColor Yellow
 }
}

# List all volumes
Get-ClusterSharedVolume | Format-Table Name, SharedVolumeInfo

2.4 Prepare Windows Server Core VHD

Copy a sysprepped Windows Server 2022 Core VHDX to the Collect volume:

# Source VHD location (update with your path)
$SourceVHD = "C:\ClusterStorage\Library\WindowsServer2022-Core-Sysprep.vhdx"

# Destination in Collect volume
$CollectPath = "C:\ClusterStorage\Collect"
$FleetVHD = "$CollectPath\FleetImage.vhdx"

# Copy VHD if not already present
if (-not (Test-Path $FleetVHD)) {
 Write-Host "Copying template VHD to Collect volume..." -ForegroundColor Yellow
 Copy-Item -Path $SourceVHD -Destination $FleetVHD -Force
 Write-Host "VHD copied successfully" -ForegroundColor Green
} else {
 Write-Host "Fleet VHD already exists at $FleetVHD" -ForegroundColor Yellow
}

# Verify VHD
Get-VHD -Path $FleetVHD | Format-Table VhdFormat, VhdType, Size, FileSize

Part 3: Install and Deploy VMFleet

3.1 Install Fleet Infrastructure

# Install VMFleet infrastructure
# This creates necessary folder structure and copies binaries

Install-Fleet -CollectVolumePath "C:\ClusterStorage\Collect"

# Verify installation
Get-ChildItem "C:\ClusterStorage\Collect" -Directory

Expected folders:

control - Fleet control scripts
result - Test results output
tools - DiskSpd and other tools
vhd - VHD storage

3.2 Create Fleet VMs

# Create the VM fleet
# Parameters:
# -VMs: Total number of VMs across cluster (4-8 per node recommended)
# -AdminPass: Local admin password for VMs
# -ConnectPass: Password for VM connection
# -BaseVHD: Path to template VHD

$TotalNodes = (Get-ClusterNode).Count
$VMsPerNode = 6 # Adjust based on available resources
$TotalVMs = $TotalNodes * $VMsPerNode

$AdminPassword = ConvertTo-SecureString "P@ssw0rd123!" -AsPlainText -Force

New-Fleet -BaseVHD "C:\ClusterStorage\Collect\FleetImage.vhdx" `
 -VMs $TotalVMs `
 -AdminPass $AdminPassword `
 -ConnectPass $AdminPassword `
 -DVDISO $null # No additional ISO needed

Write-Host "Created $TotalVMs VMFleet VMs across $TotalNodes nodes" -ForegroundColor Green

3.3 Verify Fleet Deployment

# Check fleet VMs
Get-FleetVM | Format-Table Name, State, ComputerName

# Verify VMs are distributed across nodes
Get-FleetVM | Group-Object ComputerName | Format-Table Name, Count

# Check all VMs are running
$RunningVMs = (Get-FleetVM | Where-Object State -eq "Running").Count
$TotalFleetVMs = (Get-FleetVM).Count

Write-Host "Fleet Status: $RunningVMs of $TotalFleetVMs VMs running" -ForegroundColor $(if($RunningVMs -eq $TotalFleetVMs){"Green"}else{"Yellow"})

Part 4: Start Fleet and Monitor Dashboard

4.1 Start the Fleet

# Start all fleet VMs and begin monitoring
Start-Fleet

# Wait for VMs to fully boot and respond
Write-Host "Waiting for fleet VMs to boot (60 seconds)..." -ForegroundColor Yellow
Start-Sleep -Seconds 60

# Verify fleet is ready
$FleetStatus = Get-FleetVM | Where-Object State -eq "Running"
Write-Host "Fleet ready: $($FleetStatus.Count) VMs online" -ForegroundColor Green

4.2 Launch Monitoring Dashboard

# Watch-FleetCluster provides real-time performance dashboard
# Run this in a separate PowerShell window for monitoring

Watch-FleetCluster

# Dashboard shows:
# - IOPS per node
# - Throughput (MB/s)
# - Latency (ms)
# - CPU utilization
# - Storage health

Dashboard Interpretation:

Metric	Description	Healthy Range
IOPS	I/O operations per second	Varies by workload
Throughput	MB/s read/write	Varies by workload
Latency	Average response time	< 10ms typical
CPU	Cluster CPU utilization	< 80% during test
Health	Storage pool status	All healthy

4.3 Azure Monitor Integration

If Azure Local Insights was configured in Phase 18, you can monitor VMFleet testing through the Azure portal:

Available Azure Workbooks During Testing

Workbook	What It Shows	Access Path
Azure Local Insights	Cluster health, storage IOPS/latency, node CPU/memory	Azure Portal → Azure Local → Cluster → Insights
Performance History	Historical storage metrics via `Get-ClusterPerformanceHistory`	PowerShell

Real-Time Performance Queries (Log Analytics)

// Storage performance during VMFleet testing
Perf
| where ObjectName == "Cluster CSV File System"
| where CounterName in ("Read Bytes/sec", "Write Bytes/sec", "Reads/sec", "Writes/sec")
| where TimeGenerated > ago(1h)
| summarize avg(CounterValue) by CounterName, bin(TimeGenerated, 1m)
| render timechart

// CPU utilization during test
Perf
| where ObjectName == "Processor"
| where CounterName == "% Processor Time"
| where InstanceName == "_Total"
| where TimeGenerated > ago(1h)
| summarize avg(CounterValue) by Computer, bin(TimeGenerated, 1m)
| render timechart

Capture Before/During/After Metrics

Run these queries before, during, and after VMFleet testing to document the performance impact and baseline. Export the results to include in the customer handover package.

Part 5: Run Core Workload Tests

5.1 Run Measure-FleetCoreWorkload

The Measure-FleetCoreWorkload command runs four industry-standard storage profiles:

# Run comprehensive storage performance tests
# This runs General, Peak, VDI, and SQL profiles sequentially

# Initialize report
$DateStamp = Get-Date -Format "yyyyMMdd"
$ReportPath = "C:\ClusterStorage\Collect\validation-reports"
$SummaryFile = "$ReportPath\02-vmfleet-storage-summary-$DateStamp.txt"
$CSVFile = "$ReportPath\02-vmfleet-storage-baseline-$DateStamp.csv"

New-Item -Path $ReportPath -ItemType Directory -Force -ErrorAction SilentlyContinue

$ReportHeader = @"
================================================================================
VMFLEET STORAGE PERFORMANCE BASELINE REPORT
================================================================================
Cluster: $((Get-Cluster).Name)
Date: $(Get-Date -Format "yyyy-MM-dd HH:mm:ss")
Fleet Size: $((Get-FleetVM).Count) VMs
Nodes: $((Get-ClusterNode).Count)
Generated By: $(whoami)
================================================================================

"@
$ReportHeader | Out-File $SummaryFile -Encoding UTF8

# Run core workload measurement
# Duration: ~30 minutes per profile (2+ hours total)
Write-Host "Starting core workload tests (estimated 2+ hours)..." -ForegroundColor Yellow

$CoreResults = Measure-FleetCoreWorkload -Verbose

# Save results
$CoreResults | Export-Csv -Path $CSVFile -NoTypeInformation
$CoreResults | Format-Table | Out-String | Add-Content $SummaryFile

Write-Host "Core workload tests complete. Results saved to $CSVFile" -ForegroundColor Green

5.2 Workload Profile Details

The Measure-FleetCoreWorkload command runs four industry-standard profiles with VM-CSV alignment testing at both 30% and 100%:

Profile	Block Size	Threads	Queue Depth	Read/Write	I/O Pattern	CPU Cap	Purpose
General	4K	1	32	100/0, 90/10, 70/30	Working set distribution	40%	Realistic mixed workload
Peak	4K	4	32	100/0	100% Random	None	Maximum IOPS (hero number)
VDI	8K read, 32K write	1	8	80/20	80% random, 20% sequential	None	Virtual desktop workload
SQL	8K OLTP, 32K Log	4/2	8/1	70/30 OLTP, 0/100 Log	Random OLTP, Sequential Log	None	OLTP database workload

Working Set Distribution

The General profile uses a realistic working set distribution (-rdpct95/5:4/10:1/85) where:

95% of I/O targets 5% of the file (hot data)
4% of I/O targets 10% of the file (warm data)
1% of I/O targets 85% of the file (cold data)

This simulates real-world data access patterns better than uniform random I/O.

5.3 Run Individual Workload Profiles

For targeted testing, run individual profiles:

# Peak IOPS test (4K random read)
Start-FleetSweep -b 4k -t 4 -o 32 -w 0 -r -d 300

# Throughput test (512K sequential read)
Start-FleetSweep -b 512k -t 2 -o 16 -w 0 -d 300

# Mixed workload (8K random, 70% read)
Start-FleetSweep -b 8k -t 4 -o 16 -w 30 -r -d 300

# Latency-focused (4K random read, low queue depth)
Start-FleetSweep -b 4k -t 1 -o 1 -w 0 -r -d 300

DiskSpd Parameter Reference:

Parameter	Description	Example
`-b`	Block size	`-b4k`, `-b8k`, `-b64k`, `-b512k`
`-t`	Threads per target	`-t4` = 4 threads
`-o`	Outstanding I/O queue depth per thread	`-o32` = 32 outstanding IOs
`-w`	Write percentage (0-100)	`-w0` = read only, `-w30` = 70/30 read/write
`-r`	Random I/O (omit for sequential)	`-r` = random
`-rs`	Mixed random/sequential	`-rs80` = 80% random, 20% sequential
`-d`	Duration in seconds	`-d300` = 5 minutes
`-Suw`	Disable software and hardware write cache	Recommended for accurate results
`-Z10m`	Random write buffer (defeats compression/dedup)	`-Z10m` = 10MB random buffer
`-rdpct`	Working set distribution	`-rdpct95/5:4/10:1/85` = hot/warm/cold
`-g<n>i`	IOPS limit per thread	`-g750i` = 750 IOPS cap

DiskSpd 2.2 Changes

DiskSpd 2.2 (June 2024) includes changes to the async I/O loop that improve latency measurement at high queue depths. If comparing results to older tests, results may need rebaselining.

Part 6: Collect and Analyze Results

6.1 Parse Results to CSV

# Get all result files
$ResultPath = "C:\ClusterStorage\Collect\result"
$ResultFiles = Get-ChildItem -Path $ResultPath -Filter "*.xml" -Recurse

# Parse results
$AllResults = foreach ($File in $ResultFiles) {
 [xml]$Xml = Get-Content $File.FullName
 
 $Result = $Xml.Results.TimeSpan.Iops
 
 [PSCustomObject]@{
 Timestamp = $File.LastWriteTime
 TotalIOPS = [math]::Round($Result.Total, 0)
 ReadIOPS = [math]::Round($Result.Read, 0)
 WriteIOPS = [math]::Round($Result.Write, 0)
 ReadMBps = [math]::Round($Xml.Results.TimeSpan.Throughput.Read / 1MB, 2)
 WriteMBps = [math]::Round($Xml.Results.TimeSpan.Throughput.Write / 1MB, 2)
 AvgLatencyMs = [math]::Round($Xml.Results.TimeSpan.Latency.Average, 2)
 MaxLatencyMs = [math]::Round($Xml.Results.TimeSpan.Latency.Max, 2)
 }
}

# Append to CSV
$AllResults | Export-Csv -Path $CSVFile -Append -NoTypeInformation

# Display summary
$AllResults | Format-Table -AutoSize

6.2 Generate Performance Summary

# Calculate aggregate statistics
$Stats = @"

================================================================================
PERFORMANCE SUMMARY
================================================================================

IOPS METRICS:
 Peak IOPS (4K Random Read): $($AllResults | Where-Object {$_.WriteIOPS -eq 0} | Measure-Object TotalIOPS -Maximum).Maximum
 Sustained IOPS (Mixed): $($AllResults | Measure-Object TotalIOPS -Average).Average | [math]::Round(0))
 Minimum IOPS: $(($AllResults | Measure-Object TotalIOPS -Minimum).Minimum)

THROUGHPUT METRICS:
 Peak Read Throughput: $(($AllResults | Measure-Object ReadMBps -Maximum).Maximum) MB/s
 Peak Write Throughput: $(($AllResults | Measure-Object WriteMBps -Maximum).Maximum) MB/s
 Average Throughput: $(([math]::Round(($AllResults | Measure-Object ReadMBps -Average).Average + ($AllResults | Measure-Object WriteMBps -Average).Average, 2))) MB/s

LATENCY METRICS:
 Average Latency: $(([math]::Round(($AllResults | Measure-Object AvgLatencyMs -Average).Average, 2))) ms
 Maximum Latency: $(($AllResults | Measure-Object MaxLatencyMs -Maximum).Maximum) ms
 Latency Target (< 10ms): $(if(($AllResults | Measure-Object AvgLatencyMs -Average).Average -lt 10){"PASS"}else{"REVIEW"})

"@

$Stats | Add-Content $SummaryFile
Write-Host $Stats

6.3 Expected Performance Ranges

Reference values for healthy Azure Local clusters:

Configuration	Peak IOPS (4K)	Throughput (MB/s)	Latency (ms)
2-node, NVMe	200,000+	3,000+	< 5
3-node, NVMe	400,000+	5,000+	< 5
4-node, NVMe	600,000+	8,000+	< 5
2-node, SSD	50,000+	1,000+	< 10
4-node, SSD	150,000+	3,000+	< 10

note

Performance varies based on drive type, count, and configuration. Use these as reference only.

6.4 Understanding Result Columns

The result.tsv file from Measure-FleetCoreWorkload contains detailed metrics:

Column	Description
RunLabel	Unique identifier for the test run
Workload	Profile name (General, Peak, VDI, SQL)
VMAlignmentPct	Percentage of VMs aligned to CSV owner (30% or 100%)
IOPS	Total I/O operations per second across all VMs
AverageCPU	Average cluster CPU utilization
AverageCSVReadIOPS	Read IOPS from CSV host perspective
AverageCSVWriteIOPS	Write IOPS from CSV host perspective
AverageCSVReadMilliseconds	Read latency from host
AverageCSVWriteMilliseconds	Write latency from host
AverageReadMilliseconds	Read latency from VM perspective
AverageWriteMilliseconds	Write latency from VM perspective
ReadMilliseconds50/90/99	Read latency percentiles
WriteMilliseconds50/90/99	Write latency percentiles
CutoffType	Why test stopped (No, ReadLatency, WriteLatency, Scale)

30% vs 100% Alignment

100% Alignment: All VMs are on the node that owns their CSV (best-case scenario)
30% Alignment: VMs are distributed across nodes (realistic production scenario)

Both values should be documented as they represent different operating conditions.

Part 7: Stop and Cleanup Fleet

7.1 Stop Fleet VMs

# Stop all fleet operations
Stop-Fleet

# Verify all VMs stopped
Get-FleetVM | Where-Object State -ne "Off" | Stop-VM -Force

# Check status
Get-FleetVM | Format-Table Name, State

7.2 Remove Fleet (Cleanup)

# Remove all fleet VMs and configurations
# WARNING: This deletes all fleet VMs and their disks

Remove-Fleet -Force

# Verify removal
$RemainingFleet = Get-VM -Name "Fleet*" -ErrorAction SilentlyContinue
if ($RemainingFleet) {
 Write-Host "Warning: $($RemainingFleet.Count) fleet VMs still exist" -ForegroundColor Yellow
} else {
 Write-Host "Fleet cleanup complete" -ForegroundColor Green
}

7.3 Cleanup Fleet Volumes (Optional)

# Remove Fleet volumes if no longer needed
# Keep Collect volume for future testing

$FleetVolumes = Get-ClusterSharedVolume | Where-Object { $_.Name -like "*Fleet-*" }

foreach ($Volume in $FleetVolumes) {
 # Remove the volume (WARNING: destroys data)
 # Remove-VirtualDisk -FriendlyName $Volume.SharedVolumeInfo.FriendlyVolumeName -Confirm:$false
 
 Write-Host "Would remove: $($Volume.Name)" -ForegroundColor Yellow
}

# Uncomment to actually remove:
# $FleetVolumes | ForEach-Object { Remove-VirtualDisk -FriendlyName $_.SharedVolumeInfo.FriendlyVolumeName -Confirm:$false }

Part 8: Document Baseline for Customer Handover

8.1 Create Baseline Statement

$Baseline = @"

================================================================================
STORAGE PERFORMANCE BASELINE STATEMENT
================================================================================

Cluster: $((Get-Cluster).Name)
Test Date: $(Get-Date -Format "yyyy-MM-dd")
Test Duration: Approximately 4 hours
Test Tool: VMFleet with DiskSpd
VMs Deployed: $TotalVMs VMs across $TotalNodes nodes

BASELINE PERFORMANCE METRICS:

| Workload Profile | IOPS | Throughput | Avg Latency |
|------------------|------|------------|-------------|
| Peak (4K Random) | XXX,XXX | X,XXX MB/s | X.X ms |
| General (8K Mixed) | XXX,XXX | X,XXX MB/s | X.X ms |
| VDI (8K 80/20) | XXX,XXX | X,XXX MB/s | X.X ms |
| SQL (64K Seq) | XXX,XXX | X,XXX MB/s | X.X ms |

HEALTH DURING TEST:
 Storage Pool Status: Healthy
 Virtual Disks: All Healthy
 Cluster Nodes: All Online
 No errors or faults observed

NOTES:
 - Performance metrics establish baseline for this specific cluster
 - Actual workload performance depends on application characteristics
 - Recommend monitoring during production workload onboarding
 - Contact Azure Local Cloud for performance optimization if needed

================================================================================
Validated By: _________________________ Date: ____________
================================================================================

"@

$Baseline | Add-Content $SummaryFile
Write-Host "Baseline statement added to report" -ForegroundColor Green
Write-Host "Report location: $SummaryFile" -ForegroundColor Cyan

8.2 Finalize Report

# Add report footer
$Footer = @"

================================================================================
END OF VMFLEET STORAGE PERFORMANCE REPORT
================================================================================
Report Files:
 Summary: $SummaryFile
 Raw Data: $CSVFile
================================================================================

"@

$Footer | Add-Content $SummaryFile

# Display report location
Write-Host "`n`nVMFleet testing complete!" -ForegroundColor Green
Write-Host "Summary Report: $SummaryFile" -ForegroundColor Cyan
Write-Host "CSV Data: $CSVFile" -ForegroundColor Cyan

Validation Checklist

Category	Requirement	Status
Setup	VMFleet module installed	☐
Setup	Collect volume created	☐
Setup	Fleet volumes created (per node)	☐
Setup	Template VHD copied	☐
Fleet	Fleet VMs deployed	☐
Fleet	All VMs running	☐
Testing	Core workload profiles run	☐
Testing	Results collected to CSV	☐
Performance	IOPS within expected range	☐
Performance	Latency < 10ms average	☐
Performance	No storage faults during test	☐
Cleanup	Fleet VMs removed	☐
Report	Baseline statement documented	☐

Troubleshooting

Fleet VMs Won't Start

# Check for resource issues
Get-ClusterResource | Where-Object State -ne "Online"

# Verify VHD is accessible
Test-Path "C:\ClusterStorage\Collect\FleetImage.vhdx"

# Check event log
Get-WinEvent -LogName "Microsoft-Windows-Hyper-V-VMMS-Admin" -MaxEvents 20

Poor Performance Results

# Check storage health
Get-StorageSubSystem | Get-StorageHealthReport

# Verify RDMA is working
Get-SmbMultichannelConnection

# Check for throttling
Get-StorageQoSFlow | Where-Object { $_.Status -ne "Ok" }

Results Not Collected

# Verify result path
Test-Path "C:\ClusterStorage\Collect\result"

# Check for result files
Get-ChildItem "C:\ClusterStorage\Collect\result" -Recurse

Next Step

Proceed to Task 3: Network & RDMA Validation once VMFleet testing is complete.

Previous	Up	Next
← Task 1: Infrastructure Health Validation	Testing & Validation	Task 3: Network & RDMA Validation →

Version Control

Version	Date	Author	Changes
1.0.0	2026-03-24	Azure Local Cloudnology Team	Initial release

Overview​

Purpose of This Step​

Prerequisites​

Variables from variables.yml​

Report Output​

Part 1: Install VMFleet Module​

1.1 Install from PowerShell Gallery​

1.2 Verify DiskSpd Availability​

Part 2: Prepare Storage Volumes​

2.1 Calculate Recommended Volume Sizes​

2.2 Create Collect Volume​

2.3 Create Per-Node Fleet Volumes​

2.4 Prepare Windows Server Core VHD​

Part 3: Install and Deploy VMFleet​

3.1 Install Fleet Infrastructure​

3.2 Create Fleet VMs​

3.3 Verify Fleet Deployment​

Part 4: Start Fleet and Monitor Dashboard​

4.1 Start the Fleet​

4.2 Launch Monitoring Dashboard​

4.3 Azure Monitor Integration​

Available Azure Workbooks During Testing​

Real-Time Performance Queries (Log Analytics)​

Part 5: Run Core Workload Tests​

5.1 Run Measure-FleetCoreWorkload​

5.2 Workload Profile Details​

5.3 Run Individual Workload Profiles​

Part 6: Collect and Analyze Results​

6.1 Parse Results to CSV​

6.2 Generate Performance Summary​

6.3 Expected Performance Ranges​

6.4 Understanding Result Columns​

Part 7: Stop and Cleanup Fleet​

7.1 Stop Fleet VMs​

7.2 Remove Fleet (Cleanup)​

7.3 Cleanup Fleet Volumes (Optional)​

Part 8: Document Baseline for Customer Handover​

8.1 Create Baseline Statement​

8.2 Finalize Report​

Validation Checklist​

Troubleshooting​

Fleet VMs Won't Start​

Poor Performance Results​

Results Not Collected​

Next Step​

Navigation​

Version Control​