Troubleshoot memory consumption between identical Windows Server environments

This article helps you measure and compare memory consumption between identical server environments. The goal is to reach Data-Based Conclusions (DBC) by establishing a baseline, collecting data over the same period, and ensuring that systems have the same workload.

Prerequisites

• Create Performance Monitor data collector sets on both servers.
• Ensure that you can collect data traces from each server environment.
• Understand different user workloads and their impacts on resource usage.

Quick workaround: Restart the servers

1. Restart both servers to clear any memory usage.
2. Monitor memory consumption for a predefined period after the reboot.

Measure the memory performance

Establish a baseline with a vanilla installation

Perform a vanilla installation on one server. Collect an eight-hour Performance Monitor trace during a normal working day.

Collect data from systems

During the same period as the vanilla installation, collect an eight-hour Performance Monitor trace from two additional systems, designated as client A and client B.

Ideally, we have the following three data sets from the same period for comparison:

• Vanilla
• Client A
• Client B

Analyze user workload impact

1. Identify the workload type (Light, Medium, Heavy, or Power) for each environment.
2. Use the examples provided in the session host virtual machine sizing guidelines to categorize user workloads.
   • Light: Basic data entry tasks (for example, database entry applications).
   • Medium: Tasks such as market research using static web pages and Microsoft Word.
   • Heavy: Software development involving dynamic web pages, Microsoft Outlook, and PowerPoint.
   • Power: Graphic design or 3D modeling with photo or video editing tools.
3. Use the following performance counter objects to compare the activity on the system:
   • Memory
   • CPU
   • Disk Activity
   • Number of Processes
   • Number of Sessions

Compare key performance counters for memory

• Available MBytes: Remaining free RAM
• % Committed Bytes In Use: Allocated physical resources
• Pool Paged Bytes: Kernel memory that can be paged out
• Pool Nonpaged Bytes: Kernel memory that can't be paged out
• Cache Bytes: Disk usage kernel memory
• Working Set: Active process memory in RAM

Different workloads affecting resource usage

Different user workloads lead to varying requirements for physical resources like CPU, RAM, and storage.

Solution: Standardize user behavior based on data collection

Identify similar usage patterns across all traces collected from the vanilla installation, client A, and client B.

Ensure workloads are comparable by reviewing performance counters under similar conditions:

Counter	Vanilla installation	Client A	Client B
% Committed Bytes In Use	X%	Y%	Z%
Available MBytes	X	Y	Z
Process\Working Set (_Total)	X	Y	Z

Address discrepancies by adjusting resource allocations or optimizing applications accordingly.

Scenario guide: Compare systems using the performance counters

Use tables to compare the three systems (vanilla, client A, and client B) by using the counters mentioned in the previous sections.

Counter	Vanilla installation	Client A	Client B
% Committed Bytes In Use	28.5%	42.7%	83%
Available MBytes (Min)	10.8 GB	7.5 GB	2.6 GB
Working Set (_Total) (Max)	5.2 GB	8.7 GB	15.8 GB

The result seems to confirm a different user load behavior. The result can also be confirmed by checking the number of processes:

Counter	Vanilla installation	Client A	Client B
System\Processes (Max)	199	268	338

For further confirmation, you can use the CPU usage counter, Processor\% Idle Time (_Total), to identify a different user load pattern. Here are screenshots of some samples (the first is for the vanilla installation, the second is for client A, and the third is for client B).

Vanilla:

Client A:

Client B:

If you want to drill further down on user load behavior, try to check the following processes:

• Outlook
• Office
• Teams (and with the Teams process, there are msedgewebview2 processes)
• Microsoft Edge
• Non-Microsoft browsers

Data collection

If you still need help from Microsoft Support for further analysis and guidance, contact us with comprehensive logs.

To prepare detailed logs of all performance counters across extended periods, follow these steps:

• Make sure to record the same time and duration if you wish to compare.

• Run the following command to create an eight-hour performance data collector set that overwrites itself:

  Logman.exe create counter PerfLog-30Sec-%ComputerName% -o "c:\perflogs\%computername%_PerfLog-30sec.blg" -f bincirc -v mmddhhmm -max 800 -c "Hyper-V Dynamic Memory Balancer (*)\*" "Hyper-V Hypervisor Virtual Processor(*)\*" "Hyper-V Hypervisor Logical Processor(*)\*" "\LogicalDisk(*)\*" "\Memory\*" "\Cache\*" "\Network Interface(*)\*" "\Paging File(*)\*" "\PhysicalDisk(*)\*" "\Processor(*)\*" "\Processor Information(*)\*" "\Processor Performance(*)\*" "\Process(*)\*" "\Redirector\*" "\Server\*" "\System\*" "\Server Work Queues(*)\*" "\Terminal Services\*" -si 00:00:30
  

  You still need to start the data collector set manually.