How do I solve performance issues with the Windows Machine Agent?

[Joe.Catera]

How do I solve performance issues when collecting metrics with the Windows Machine Agent?

Issue

Some users have experienced slow performance and/or high resource consumption when collecting server metrics on Windows with Server Visibility enabled. 

Solution

The ServerMonitoring extension uses theWindows Management Instrumentation (WMI) to collect metrics on Windows machines. If you experience problems using WMI, you can configure the ServerMonitoring extension to use SIGAR instead.

The trade-off is that SIGAR cannot collect all the metrics that WMI can. See What metrics are not available when using SIGAR?, below for the list. In addition, this solution is available only on version 4.3.8 (or higher) of the Machine Agent.

To use SIGAR, specify the following System Property on the agent (the default is false)

-Dappdynamics.machine.agent.extensions.sim.sigarCollectorEnabled = true

 

What metrics are not available when using SIGAR?

 If you use SIGAR (not WMI) with the ServerMonitoring extension, the following metrics will be unavailable.

Server Dashboard

• Server Dashboard > CPU chart > CPUs (number)
• Server Dashboard > Properties > all CPU metrics

Metric Browser

• Hardware Resources|Disks|<partition>|Avg Service Time (ms)
• Hardware Resources|Disks|<partition>|Avg Queue Time (ms)
• Hardware Resources|Disks|<partition>|% CPU Time
• Hardware Resources|Disks|<partition>|Avg IO Utilization (%)
• Hardware Resources|Disks|<partition>|Avg read time (ms)
• Hardware Resources|Disks|<partition>|Avg write time (ms)
• Hardware Resources|Disks|<partition>|Queue Time (ms) 95th Percentile
• Hardware Resources|Disks|Avg IO Utilization (%)
• Hardware Resources|Process|<processName>|Faults|Major Faults/sec
• Hardware Resources|Process|<processName>|Faults|Minor Faults/sec
• Hardware Resources|CPU|%IOWait 95th Percentile
• Hardware Resources|Memory|Swap Total (MB)
• Hardware Resources|Memory|Swap Used (MB)
• Hardware Resources|Memory|Swap Free (MB)
• Hardware Resources|Memory|Swap Used %
• Hardware Resources|Memory|Swap Free %
• Hardware Resources|Memory|Pages Paged In 95th Percentile
• Hardware Resources|Memory|Pages Paged Out 95th Percentile
• Hardware Resources|Memory|Pages Swapped In 95th Percentile
• Hardware Resources|Memory|Pages Swapped Out 95th Percentile
• Hardware Resources|Load|Last 1 minute
• Hardware Resources|Load|Last 5 minutes
• Hardware Resources|Load|Last 15 minutes
• Hardware Resources|Service Availability|<service ID>|Response Time (ms)
• Hardware Resources|Service Availability|<service ID>|Response Size (bytes)
• Hardware Resources|Service Availability|<service ID>|Success Rate (%)
• Hardware Resources|Service Availability|<service ID>|Response Size (bytes)
• Hardware Resources|Service Availability|<service ID>|Success Rate (%)
• Hardware Resources|Network|<iface>|Avg Utilization (%)
• Hardware Resources|Network|Avg Utilization (%)

[Karen.Isaakyan]

Bumped into WMI failing every 20 minutes on Windows Server 2012 R2 due to https://docs.microsoft.com/en-us/troubleshoot/windows-server/system-management-components/high-cpu-u...

which provides a decent description of the reasoning behind the WMI timeouts occurring on a regular basis.

WMI stands for Windows Management Instrumentation not Monitoring Interface as above.

[Claudia.Landivar]

@Karen.Isaakyan , Thank you for sharing the Windows documentation, as well as pointing out the incorrect acronym. We appreciate it!

We'll get that revised in the article.

The acronym now reads as standing for Windows Management Instrumentation.

Claudia Landivar

Community Manager & Editor

[Karen.Isaakyan]

Continuing on investigating the timeouts I've figured out they're due to appd collector.vbs script using WMI to query the following classes:

Win32_PerfRawData_PerfDisk_LogicalDisk
Win32_PerfRawData_Tcpip_NetworkInterface
Win32_PerfRawData_PerfOS_System

Is there any chance Appd SIM Extension will move to PDH (Performance Data Helper) instead of WMI, following the OSHI path: https://github.com/oshi/oshi/issues/505

[Biswajit.Nanda]

Karen,

The newer version of AppDynamics Machine Agent (with SIM) has already moved away from using WMI to OSHI libraries already for Memory, CPU and Process metrics starting from v4.5.17.

For the network, disk and OS metrics, we still continue to use WMI, which we agree that has performance issues. AT this point, there is no plan for the machine-agent to use PDH.