Showing posts with label diskspd. Show all posts
Showing posts with label diskspd. Show all posts

Friday, September 6, 2024

Revisiting Storage Performance Benchmarking

Few years ago, I had the opportunity to explore the intricacies of storage performance benchmarking using tools like FIO, DISKSPD, and Iometer. Those studies provided valuable insights into the performance characteristics of various storage solutions, shaping my understanding and approach to storage performance analysis. As I prepare for an upcoming project in this domain, I find it essential to revisit my previous work, reflect on the lessons learned, and share my experiences. This blog post aims to provide a comprehensive overview of my benchmarking journey and the evolving landscape of storage performance studies.


Recent advancements 

The field of storage technology has seen significant advancements in recent years. The rise of NVMe and storage-class memory technologies has also redefined high-end storage performance, offering unprecedented speed and efficiency. These advancements highlight the dynamic nature of storage performance benchmarking and underscore the importance of staying updated with the latest tools and methodologies.

Challenges

Benchmarking storage performance is not without its challenges. One of the primary difficulties is ensuring a consistent and controlled testing environment, as variations in hardware, software, and network conditions can significantly impact results. Another challenge is the selection of appropriate benchmarks that accurately reflect real-world workloads, which requires a deep understanding of the specific use cases and performance metrics. Additionally, interpreting the results can be complex, as it involves analyzing multiple metrics such as IOPS, throughput, and latency, and understanding their interplay. These challenges necessitate meticulous planning and a thorough understanding of both the benchmarking tools and the storage systems being tested.

Prior works

Following are some of the articles on storage benchmarking that I’ve published in the past:

Custom storage benchmarking framework

While there are numerous storage benchmarking tools available, such as VMFleet and HCIBench, I wanted to highlight a custom framework I developed a few years ago. Here are some reasons why we created this custom tool:

  • Great learning experience: It provided valuable insights into how things work.
  • Customization: Being a custom framework, it allows you to add or remove features as needed.
  • Flexibility: You can modify multiple parameters to suit your requirements.
  • Custom test profiles: You can create tailored storage test profiles.
  • No IP assignment needed: There’s no need for IP assignment or DHCP for the stress test VMs.
  • Centralized log collection: It offers centralized log collection for detailed analysis.


You can access the scripts and readme on my GitHub repository:

https://github.com/vineethac/vsan_cluster_storage_benchmarking_with_diskspd


Here is an overview.

  • Profile Manifest: All storage test profiles are listed in profile_manifest.psd1. You can define as many profiles as you want.
  • VM Template: A Windows VM template should be present in the vCenter server.
  • Benchmarking Manifest: Details of vCenter, cluster name, VM template, number of stress test VMs per host, etc., are provided in benchmarking_manifest.psd1.
  • Deploy Test VMs: deploy_test_vms.ps1 will deploy all the test VMs with pre-configured parameters.
  • Start Stress Test: start_stress_test.ps1 will initiate the storage stress test process for all the profiles mentioned in profile_manifest.psd1 one by one.
  • Log Collection: All log files will be automatically copied to a central location on the host from where these scripts are running.
  • Cleanup: Use delete_test_vms.ps1 to clean up the stress test VMs from the cluster.


Note:
 These scripts were created about five years ago, and I haven’t had the opportunity to refactor them according to current best practices and new PowerShell scripting standards. I plan to enhance them in the coming months!

This overview should provide you with a clear understanding of the overall process and workflow involved in the storage benchmarking process. I hope it was useful. Cheers!

Saturday, April 18, 2020

vSAN performance benchmarking

In this article, I will explain briefly on performance benchmarking considerations, factors affecting performance, and some of the best practices. We do performance benchmarking to understand the capabilities and bottlenecks of a system. When I say system it could be a storage system, CPU, GPU, network switch, etc. Now let's consider a VMware vSAN cluster infrastructure. It includes multiple components and each of these contributes to the performance. In this case, the vSAN cluster is the solution under test. We will have to conduct performance benchmarking to understand the storage performance behavior of the cluster. When I say storage behavior it includes the IOPS, latency, and throughput that the cluster can produce under varying loads.

The goal of benchmarking
  • Identify bottlenecks
    • Hardware bottleneck
    • Software bottleneck
    • Application bottleneck
  • Compare tradeoffs
  • Manage expectations
  • Make decisions

Usually in a real-world scenario, benchmarking will be done once the cluster is deployed/ ready and before starting to host production workload on top of it. As these benchmark values define the performance maximums it will be helpful to decide on when to scale or upgrade the cluster before it hits a bottleneck.

Fundamental factors of vSAN performance

Server hardware
  • Compatibility as per vSAN HCL
Host
  • Number of hosts in the cluster
  • Power settings
  • CPU - number of cores and frequency 
Storage
  • Hybrid or All-flash
  • NVMe, SAS, or SATA
  • Number of disk groups per host
  • Storage controller configuration
  • Compatibility of hardware devices as per vSAN HCL
Network
  • 10/ 25/ 40 GbE
  • MTU 
  • LAG
SPBM policy
  • FTT (Failures To Tolerate)
  • FTM (Mirroring/ Erasure coding)
  • Thin or Thick provision
Security
  • Encryption
  • Checksum
Other
  • Stripe width
  • Flash read cache reservation
  • IOPS limit for object
All of the above factors will affect performance. So you should know the benefits and tradeoffs. 

Benchmarking methodology

Image credit: VMware

Storage benchmarking tools

IO load generation tools
Application-specific tools
  • HammerDB (MSSQL, Oracle)
  • Jetstress (MS Exchange)
  • SLOB (Oracle)
  • DBGen (MSSQL, Oracle)

Best practices

  • Understand the production performance metrics.
  • Test what you plan to deploy.
  • Workload modeling.
  • Plan for use case testing.
  • Choose an appropriate size for benchmarking
  • Choose the right tool.
  • Pre-allocate blocks while testing.
  • Test for a longer time duration.
  • Deploy multiple VMs with multiple VMDKs.

References

Thursday, May 11, 2017

Benchmarking Hyper-Converged Storage Spaces Direct (S2D) Cluster

Finally I managed to write some PowerShell code as I am completely inspired by my new PS geek friends. The scripts can be used to generate load and stress test your S2D as well as traditional Hyper-V 2016 cluster. These are functionally similar to VM Fleet. There are 7 scripts in total.
  1. create_clustered_testvms.ps1 : this script creates virtual machines on the cluster nodes which will be used for stress testing
  2. start_all_testvms.ps1 : start all those clustered VMs that you just created
  3. io_stress_trigger.ps1 : to trigger IO stress on all VMs using diskspd 
  4. rebalance_all_testvms.ps1 : this script is originally from winblog, I just made a small change so that it will use live migration while moving the clustered VMs back to their owner node
  5. watch_iops_live.ps1 : to view read, write and total iops of each CSV disk on the S2D cluster 
  6. stop_all_testvms.ps1 : to shutdown all the clustered VMs
  7. wipeoff_testvms.ps1 : to delete all VMs that you created using the first script
PS Version which I am using is given below.


Now I will explain briefly about how to use these scripts and a few prerequisites. Say, you have a 4 node hyper-converged S2D cluster.


As there are 4 nodes , you should have 4 cluster shared volumes (CSV). Assign one CSV to each cluster node as shown below. This means when you create VMs on NODE-01, it will be placed on CSV Volume AA, for NODE-02 VMs will be placed on Volume BB and so on. 


Volume AA is C:\ClusterStorage\Volume1


Similarly,
Volume BB is C:\ClusterStorage\Volume2
Volume CC is C:\ClusterStorage\Volume3
Volume DD is C:\ClusterStorage\Volume4


Your cluster shared volumes are ready now. Create 2 folders inside Volume1 as shown below.


Copy all the 7 PS scripts to scripts folder. Code of each script is given at the end.


You now need a template VHDX and that needs to be copied to template folder.
Note: It should be named as "template"


This template is nothing but a Windows Server 2016 VM created on a dynamically expanding disk. So you just have to create a VM with dynamically expanding VHDX, install Windows Server 2016 and set local administrator password to "Pass1234". Download diskspd from Microsoft, unzip it and just copy the diskspd.exe to C drive of the VM you just created.


Shutdown the VM. No need to sysprep it. Copy the VHDX disk of the VM to template folder and rename it to "template". The disk will be around 9.5 GB in size. Once the template is copied, you are all set to start. 

Step 1: 
Run create_clustered_testvms.ps1 
This will create clustered testvms on each of the nodes. It will be done in such a way that VMs on NODE-01 will be stored on Volume1, VMs on NODE-02 will be stored on Volume2 and so on 

Step 2:
Run start_all_testvms.ps1 
This will start all the testvms

Step 3:
Wait for a few seconds to ensure all the testvms are booted properly; then run io_stress_trigger.ps1 and provide necessary input parameters 



Step 4:
You can watch IOPS of the cluster using watch_iops_live.ps1


If you would like to live migrate some testvms while the stress test is running you can try it and observe the IO variations. But before running the io_stress_trigger.ps1 again you have to move/ migrate all those testvms back to their preferred owners. This can be done using rebalance_all_testvms.ps1 .  

If any testvms are not running on their preferred owner, then io_stress_trigger.ps1 will fail for those VMs. Here, testvms running on NODE-01 has preferred owner NODE-01, similarly for all other testvms. So you have to make sure all the testvms are running on their preferred owner before starting io stress script.

Use stop_all_testvms.ps1 to shutdown all the clustered testvms that you created on step 1. To delete all the testvms, you can use wipeoff_testvms.ps1 .

NOTE: While running scripts 2,3,6 and 7 please make sure all the testvms are running on their preferred owner. Use rebalance_all_testvms.ps1 to assign all testvms back to its preferred owner! Also please run all these scripts on PowerShell with elevated privileges after directly logging into any of the cluster nodes.

All codes given below. It might not be optimal but I am pretty sure it works! Cheers !
--------------------------------------------------------------------------------------------------------------------------

#BEGIN_create_clustered_testvms.ps1
#Get cluster info
$Cluster_name = (Get-Cluster).name
$Nodes_name = (Get-ClusterNode).name
$Node_count = (Get-ClusterNode).count

#Input VM config
$VM_count = Read-Host "Enter number of VMs/ node"
$Cluster_VM_count = $VM_count*$Node_count
[int64]$RAM = Read-Host "Enter memory for each VM in MB Eg: 4096"
$RAM = 1MB*$RAM
$CPU = Read-Host "Enter CPU for each VM"

#Creds to Enter-PSSession
$pass = convertto-securestring -asplaintext -force -string Pass1234
$cred = new-object -typename system.management.automation.pscredential -argumentlist "administrator", $pass

#Loop for each node in cluster
for($i=1; $i -le $Node_count; $i++){
    $VM_path = "C:\ClusterStorage\Volume$i"
    $Node = $Nodes_name[$i-1]

    #Remote session to each node
    $S1 = New-PSSession -ComputerName $Node -Credential $cred

    #Loop for creating new testvms on each node
    for($j=1; $j -le $VM_count; $j++){
        $VM_name = "testvm-$Node-$j"
        new-vm -name $VM_name -computername $Node -memorystartupbytes $RAM -generation 2 -Path $VM_path
        set-vm -name $VM_name -ProcessorCount $CPU -ComputerName $Node
        New-Item -path $VM_path\$VM_name -name "Virtual Hard Disks" -type directory

        #Copy template disk
        Copy-Item "C:\ClusterStorage\Volume1\template\template.vhdx" -Destination "$VM_path\$VM_name\Virtual Hard Disks" -Verbose
        Add-VMHardDiskDrive -VMName $VM_name -ComputerName $Node -path "$VM_path\$VM_name\Virtual Hard Disks\template.vhdx" -Verbose

        #Create new fixed test disk
        New-VHD -Path "$VM_path\$VM_name\Virtual Hard Disks\test_disk.vhdx" -Fixed -SizeBytes 40GB
        Add-VMHardDiskDrive -VMName $VM_name -ComputerName $Node -path "$VM_path\$VM_name\Virtual Hard Disks\test_disk.vhdx" -Verbose

        Get-VM -ComputerName $Node -VMName $VM_name | Start-VM
        Add-ClusterVirtualMachineRole -VirtualMachine $VM_name
        Set-ClusterOwnerNode -Group $VM_name -owner $Node
        Start-Sleep -S 10

        #Remote session to each testvm on the node to initialize and format test disk (drive D:)
        Invoke-Command -Session $S1 -ScriptBlock {param($VM_name2,$cred2) Invoke-Command -VMName $VM_name2 -Credential $cred2 -ScriptBlock {
            Initialize-Disk -Number 1 -PartitionStyle MBR
            New-Partition -DiskNumber 1 -UseMaximumSize -DriveLetter D
            Get-Volume | where DriveLetter -eq D | Format-Volume -FileSystem NTFS -NewFileSystemLabel Test_disk -confirm:$false
            }} -ArgumentList $VM_name,$cred

        Start-Sleep -S 5
        Get-VM -ComputerName $Node -VMName $VM_name | Stop-VM -Force
        }
    }
#END_create_clustered_testvms.ps1


--------------------------------------------------------------------------------------------------------------------------

#BEGIN_start_all_testvms.ps1
#Get cluster info
$Cluster_name = (Get-Cluster).name
$Nodes_name = (Get-ClusterNode).name
$Node_count = (Get-ClusterNode).count

for($i=1; $i -le $Node_count; $i++){
    $Node = $Nodes_name[$i-1]
    Get-VM -ComputerName $Node -VMName "testvm-$Node*" | Start-VM -AsJob

    }
#END_start_all_testvms.ps1


--------------------------------------------------------------------------------------------------------------------------

#BEGIN_io_stress_trigger.ps1
#Get cluster info
$Cluster_name = (Get-Cluster).name
$Nodes_name = (Get-ClusterNode).name
$Node_count = (Get-ClusterNode).count

#Creds to Enter-PSSession
$pass = convertto-securestring -asplaintext -force -string Pass1234
$cred = new-object -typename system.management.automation.pscredential -argumentlist "administrator", $pass

$time = Read-Host "Enter duration of stress in seconds (Eg: 300)"
$block_size = Read-Host "Enter block size (Eg: 4K)"
$writes = Read-Host "Enter write percentage (Eg: 20)"
$OIO = Read-Host "Enter number of outstanding IOs (Eg: 16)"
$threads = Read-Host "Enter number of threads (Eg: 2)"

#Loop for each node in cluster
for($i=1; $i -le $Node_count; $i++){
    $VM_path = "C:\ClusterStorage\Volume$i"
    $Node = $Nodes_name[$i-1]

    #Remote session to each node
    $S1 = New-PSSession -ComputerName $Node -Credential $cred

    $VM_count = (Get-VM -ComputerName $Node -VMName "testvm-$Node*").Count

    #Loop for creating new testvms on each node
    for($j=1; $j -le $VM_count; $j++){
        $VM_name = "testvm-$Node-$j"

        #Remote session to each testvm
        Invoke-Command -Session $S1 -ScriptBlock {param($VM_name2,$cred2,$time1,$block_size1,$writes1,$OIO1,$threads1) Invoke-Command -VMName $VM_name2 -Credential $cred2 -ScriptBlock {param($time2,$block_size2,$writes2,$OIO2,$threads2)
        C:\diskspd.exe -"b$block_size2" -"d$time2" -"t$threads2" -"o$OIO2" -h -r -"w$writes2" -L -Z500M -c38G D:\io_stress.dat
        } -AsJob -ArgumentList $time1,$block_size1,$writes1,$OIO1,$threads1 } -ArgumentList $VM_name,$cred,$time,$block_size,$writes,$OIO,$threads


        }
    }
#END_io_stress_trigger.ps1


--------------------------------------------------------------------------------------------------------------------------

#BEGIN_rebalance_all_testvms.ps1
$clustergroups = Get-ClusterGroup | Where-Object {$_.IsCoreGroup -eq $false}
 foreach ($cg in $clustergroups)
 {
     $CGName = $cg.Name
     Write-Host "`nWorking on $CGName"
     $CurrentOwner = $cg.OwnerNode.Name
     $POCount = (($cg | Get-ClusterOwnerNode).OwnerNodes).Count
     if ($POCount -eq 0)
     {
         Write-Host "Info: $CGName doesn't have a preferred owner!" -ForegroundColor Magenta
     }
     else
     {
         $PreferredOwner = ($cg | Get-ClusterOwnerNode).Ownernodes[0].Name
         if ($CurrentOwner -ne $PreferredOwner)
         {
             Write-Host "Moving resource to $PreferredOwner, please wait..."
             $cg | Move-ClusterVirtualMachineRole -MigrationType Live -Node $PreferredOwner
         }
         else
         {
             write-host "Resource is already on preferred owner! ($PreferredOwner)"
         }
     }
 }
 Write-Host "`n`nFinished. Current distribution: "

 Get-ClusterGroup | Where-Object {$_.IsCoreGroup -eq $false} 
#END_rebalance_all_testvms.ps1


--------------------------------------------------------------------------------------------------------------------------

#BEGIN_watch_iops_live.ps1
#Get cluster info
$Cluster_name = (Get-Cluster).name
$Nodes_name = (Get-ClusterNode).name

while($true)
{

    [int]$total_IO = 0
    [int]$total_readIO = 0
    [int]$total_writeIO = 0
 
    clear
     
    "{0,-15} {1,-15} {2,-15} {3,-15} {4, -15} {5, -15}" -f "Host", "Total IOPS", "Reads/Sec", "Writes/Sec", "Read Q Length", "Write Q Length"

    for($j=1; $j -le $Nodes_name.count; $j++){

        $Node = $Nodes_name[$j-1]

        $Data = Get-CimInstance -ClassName Win32_PerfFormattedData_CsvFsPerfProvider_ClusterCSVFS -ComputerName $Node | Where Name -like Volume$j

        [int]$T = $Data.ReadsPerSec+$Data.WritesPerSec
     
        "{0,-15} {1,-15} {2,-15} {3,-15} {4,-15} {5, -15}" -f "$Node", "$T", $Data.ReadsPerSec, $Data.WritesPerSec, $Data.CurrentReadQueueLength, $Data.CurrentWriteQueueLength

        $total_IO = $total_IO+$T
        $total_readIO = $total_readIO+$Data.ReadsPerSec
        $total_writeIO = $total_writeIO+$Data.WritesPerSec
        }

    echo `n
    "{0,-15} {1,-15} {2,-15} {3,-15} " -f "Cluster IOPS", "$total_IO", "$total_readIO", "$total_writeIO"

    Start-Sleep -Seconds 3

}
#END_watch_iops_live.ps1

--------------------------------------------------------------------------------------------------------------------------

#BEGIN_stop_all_testvms.ps1
#Get cluster info
$Cluster_name = (Get-Cluster).name
$Nodes_name = (Get-ClusterNode).name
$Node_count = (Get-ClusterNode).count

for($i=1; $i -le $Node_count; $i++){
    $Node = $Nodes_name[$i-1]
    Get-VM -ComputerName $Node -VMName "testvm-$Node*" | Stop-VM -Force -AsJob
    }
#END_stop_all_testvms.ps1

--------------------------------------------------------------------------------------------------------------------------

#BEGIN_wipeoff_testvms.ps1
#Get cluster info
$Cluster_name = (Get-Cluster).name
$Nodes_name = (Get-ClusterNode).name
$Node_count = (Get-ClusterNode).count

#Loop for each node in cluster
for($i=1; $i -le $Node_count; $i++){
    $VM_path = "C:\ClusterStorage\Volume$i"
    $Node = $Nodes_name[$i-1]

    $VM_count = (get-vm -ComputerName $Node -Name "testvm-$Node-*").Count

    #Loop to delete testvm on each node
    for($j=1; $j -le $VM_count; $j++){

        $VM_name = "testvm-$Node-$j"
        $full_path = "$VM_path\$VM_name"

        Get-VM -Computername $Node -VMname $VM_name | stop-vm -force
        Get-ClusterGroup $VM_name | Remove-ClusterGroup -Force -RemoveResources
        Get-VM -Computername $Node -VMname $VM_name | remove-vm -force
        Remove-Item $full_path -Force -Recurse -ErrorAction SilentlyContinue -Verbose
        }

    }
#END_wipeoff_testvms.ps1

--------------------------------------------------------------------------------------------------------------------------