Run performance diagnostics reports on Azure virtual machines

The following operating systems are currently supported for both on-demand and continuous diagnostics:

• Windows Server 2025
• Windows Server 2022
• Windows Server 2019
• Windows Server 2016
• Windows Server 2012 R2
• Windows Server 2012
• Windows 11
• Windows 10

The following distributions are currently supported for on-demand diagnostics.

[*] See the following notes.

• RHEL 8 doesn't have Python installed by default because both Python 2 and Python 3.6 are available. To install Python 3.6, run yum install python3.
• PCI devices information is not collected on Debian based distributions.
• LVM information is partially collected on some distributions.

1. Download PerfInsights.zip.

2. Unblock the PerfInsights.zip file. To do this, right-click the PerfInsights.zip file, and select Properties. In the General tab, select Unblock, and then select OK. This action ensures that the tool runs without any other security prompts.

   

3. Expand the compressed PerfInsights.zip file to your temporary drive.

Download PerfInsights.tar.gz to a folder on your virtual machine and extract the contents using the below commands from the terminal.

wget https://download.microsoft.com/download/9/F/8/9F80419C-D60D-45F1-8A98-718855F25722/PerfInsights.tar.gz

tar xzvf PerfInsights.tar.gz

You can run the following on-demand reports from Windows machines:

Quick analysis

This scenario collects the disk configuration and other important information, including:

• Event logs
• Network status for all incoming and outgoing connections
• Network and firewall configuration settings
• Task list for all applications that are currently running on the system
• Microsoft SQL Server database configuration settings (if VM is running SQL Server)
• Storage reliability counters
• Important Windows hotfixes
• Installed filter drivers

This is a passive collection of information that shouldn't affect the system.

Benchmarking

This scenario runs the Diskspd benchmark test (IOPS and MBPS) for all drives that are attached to the VM.

Performance analysis

This scenario runs a performance counter trace by using the counters that are specified in the RuleEngineConfig.json file. If the VM is identified as a server that is running SQL Server, a performance counter trace is run. It does so by using the counters that are found in the RuleEngineConfig.json file. This scenario also includes performance diagnostics data.

Azure Files analysis

This scenario runs a special performance counter capture together with a network trace. The capture includes all the Server Message Block (SMB) client shares counters. The following are some key SMB client share performance counters that are part of the capture:

Advanced performance analysis

When you run an advanced performance analysis, you select traces to run in parallel. If you want, you can run them all (Performance Counter, Xperf, Network, and StorPort).

You can run the following on-demand reports from Linux machines:

Quick performance analysis

This scenario collects basic information such as storage and hardware configuration of your virtual machine, various logs, including:

• Operating System information
• PCI device information
• General Guest OS logs
• Configuration files
• Storage information
• Azure Virtual Machine Configuration (collected using Azure Instance Metadata Service)
• List of running processes, Disk, Memory, and CPU usage
• Networking information

This is a passive collection of information that shouldn't affect the system.

Performance analysis

This scenario is similar to Quick performance analysis but allows capturing diagnostics information for longer duration.

HPC performance analysis

This scenario is meant for troubleshooting issues on HPC size VMs, meaning H-Series and N-Series. It checks a VMs configuration against what the Azure HPC Platform team has tested and recommends. It also collects logs and diagnostics related to the status and configuration of the special hardware that is available on those VMs, including:

• GPU Driver information
• GPU hardware diagnostics
• InfiniBand driver information and configuration
• InfiniBand device diagnostics
• Network configuration files
• Performance tuning information

Performance diagnostics trace

Runs a rule-based engine in the background to collect data and diagnose ongoing performance issues. Rules are displayed in the report under the Category -> Finding tab.

Each rule consists of the following:

• Finding: Description of the finding.
• Recommendation: Recommendation on what action could be taken for the finding. There are also reference link(s) to documentation that provides more information on the finding and/or recommendation.
• Impact Level: Represents the potential for having an impact on performance.

The following categories of rules are currently supported:

• High resource usage:
  • High CPU usage: Detects high CPU usage periods, and shows the top CPU usage consumers during those periods.
  • High memory usage: Detects high memory usage periods, and shows the top memory usage consumers during those periods.
  • High disk usage: Detects high disk usage periods on physical disks, and shows the top disk usage consumers during those periods.
• Storage: Detects specific storage configurations.
• Memory: Detects specific memory configurations.
• GPU: Detects specific GPU configurations.
• Network: Detects specific network settings.
• System: Detects specific system settings.

Open Windows command prompt as an administrator, and then run PerfInsights.exe to view the available commandline parameters and their descriptions.

cd <the path of PerfInsights folder>
PerfInsights

The basic syntax for running PerfInsights scenarios is:

PerfInsights /run <ScenarioName> [AdditionalOptions]

Use the /list command to view the list of supported scenarios:

PerfInsights /list

Following are examples of running different troubleshooting scenarios:

• Run the performance analysis scenario for 5 minutes:

PerfInsights /run vmslow /d 300 /AcceptDisclaimerAndShareDiagnostics

• Run the advanced scenario with Xperf and Performance counter traces for 5 minutes:

PerfInsights /run advanced xp /d 300 /AcceptDisclaimerAndShareDiagnostics

• Run the benchmark scenario for 5 minutes:

PerfInsights /run benchmark /d 300 /AcceptDisclaimerAndShareDiagnostics

• Run the performance analysis scenario for 5 minutes and upload the result zip file to the storage account:

PerfInsights /run vmslow /d 300 /AcceptDisclaimerAndShareDiagnostics /sa <StorageAccountName> /sk <StorageAccountKey>

Before running a scenario, PerfInsights prompts you to agree to share diagnostic information and to agree to the EULA. Use /AcceptDisclaimerAndShareDiagnostics option to skip these prompts.

If you have an active support ticket with Microsoft and running PerfInsights per the request of the support engineer you are working with, make sure to provide the support ticket number using the /sr option.

By default, PerfInsights will try updating itself to the latest version if available. Use /SkipAutoUpdate or /sau parameter to skip auto update.

If the duration switch /d is not specified, PerfInsights will prompt you to repro the issue while running vmslow, azurefiles and advanced scenarios.

When the traces or operations are completed, a new file appears in the same folder as PerfInsights. The name of the file is PerformanceDiagnostics_yyyy-MM-dd_hh-mm-ss-fff.zip. You can send this file to the support agent for analysis or open the report inside the zip file to review findings and recommendations.

Navigate to the folder that contains perfinsights.py file, and then run perfinsights.py to view the available commandline parameters and their descriptions.

cd <the path of PerfInsights folder>
sudo python perfinsights.py

The basic syntax for running PerfInsights scenarios is:

sudo python perfinsights.py -r <ScenarioName> -d [duration]<H | M | S> [AdditionalOptions]

You can use the following example to run Quick performance analysis scenario for 1 minute and create the results under /tmp/output folder:

sudo python perfinsights.py -r quick -d 1M -a -o /tmp/output

You can use the following example to run performance analysis scenario for 5 minutes and upload the result (stores in a TAR file) to the storage account:

sudo python perfinsights.py -r vmslow -d 300S -a -t <StorageAccountName> -k <StorageAccountKey> -i <full resource Uri of the current VM>

You can use the following example to run the HPC performance analysis scenario for 1 minute and upload the result TAR file to the storage account:

sudo python perfinsights.py -r hpc -d 60S -a -t <StorageAccountName> -k <StorageAccountKey> -i <full resource Uri of the current VM>

When the run is completed, a new tar file appears in the same folder as PerfInsights unless no output folder is specified. The name of the file is PerformanceDiagnostics_yyyy-MM-dd_hh-mm-ss-fff.tar.gz. You can send this file to the support agent for analysis or open the report inside the file to review findings and recommendations.