The following is a summary of the statistics gathered during the deployment of an Operations Manager 2012 agent to a Windows Server 2008 R2 computer system both during the agent deployment and for five minutes after the agent deployment.  This summary also provides the statistics gathered monitoring the system agentless in OpsMgr 2012.

Testing details:

  • Statistics were gathered using a variety of methods including performance monitor, and network usage statistics.
  • The agent was deployed from an OpsMgr 2012 environment running in Hyper-V which was assigned two processors (Intel 3.07 GHz) and 1 gb of memory.
  • The agent was pushed from the Operations Manager 2012 console.

OpsMgr 2012 Windows Agent Monitoring resource requirements:

Processor: 1% average increase in processor utilization

Disk: 150 average increase in pages per second

Disk: 348 MB data (%programfiles%\System Center Operations Manager)

Network: 4.5 MB data sent and received to the system during installation

Memory: 31-48 MB less available memory (Addition of the Health Service Host and Monitoring Host account for 31 MB)

Time to Deploy to Not Monitored state: 2.5 minutes

Time to Deploy to Monitored state: 6.5 minutes (including the time to deploy to not monitored state)

 

After the installation was complete, the performance counters gathered prior to the installation were compared to those gathered after installation. The results indicate additional overhead associated with the Operations Manager 2012 agent after installation was complete.

Processor:  1% average increase in processor utilization

Disk: 9 average increase in pages per second

Disk: 351 MB data (%programfiles%\System Center Operations Manager)

Network:  .002 MB/min additional traffic

Memory:  19-32 MB less available memory (Addition of the Health Service Host and Monitoring Host account for 19 MB)

 

OpsMgr 2012 Windows Agent Resource Requirements Summary: With current computer equipment levels, the statistical impacts of the OpsMgr overhead is minor. Key Performance Indicator (KPI) information below:

  • Processor: Processor utilization impacts of about 1% (this can also be checked using the agent processor utilization counter)
  • Disk: Disk space is the only significant impact of the OpsMgr agent – plan for at least 350 MB of space for the OpsMgr agent. This will increase based upon the number (and type) of management packs deployed in the environment.
  • Memory: Memory overhead in most systems will not be significant but plan for at least 50 MB of memory for the OpsMgr agent.
  • Network: Network impacts appear to be insignificant except when deploying a large number of agents – those should be staggered to decrease impact.

Note: This is an interesting shift since the assessment I performed on OpsMgr 2007 agent deployment requirements and which I also integrated into the OpsMgr Unleashed book. For agent based monitoring processor utilization has decreased, disk space requirements have increased, memory requirements have increased, but network requirements have stayed roughly the same.  Most of this I attribute to three factors:

1) Virtualization platform evolution: The difference in virtualization platforms which were used for testing (Virtual Server 2005 R2 versus Hyper-V in Windows Server 2008 R2).

2) Hardware evolution: The difference in hardware which was used for testing (From a dual processor 1Ghz system to a quad processor 3Ghz system, from a slower IDE drive configuration to a multiple drive SATA configuration).

3) Operations Manager evolution: Original tests were done on OpsMgr 2007 RTM compared with OpsMgr 2012 RTM.

 

OpsMgr 2012 Agentless Monitoring resource requirements:

Processor: < 1% average increase in processor utilization

Disk: < 1 average increase in pages per second

Disk: < 1 MB data (as there is no %programfiles%\System Center Operations Manager folder created)

Network: < 1 MB data sent and received to the system during installation

Memory: 14 MB less available memory

Time to Deploy to Monitored state: 2.5 minutes

 

After the agent was appearing as monitored the performance counters gathered prior to the installation were compared to those gathered after installation. The results indicate additional overhead associated with the Operations Manager 2012 agentless monitoring after the agent was appearing as monitored.

Processor:  < 1% average increase in processor utilization

Disk: < 1 average increase in pages per second

Disk: < 10 MB

Network:  < 1 MB/min additional traffic

Memory:  < 1  MB less available memory

OpsMgr 2012 Agentless Monitoring Resource Requirements Summary: With current computer equipment levels, the statistical impacts of the OpsMgr overhead is minor. Key Performance Indicator (KPI) information below:

  • Processor: Processor utilization impacts of about 1% (this cannot be checked using the agent processor utilization counter since the counter is not available when agentless monitoring a system)
  • Disk: Disk space impact was insignificant as as there is no %programfiles%\System Center Operations Manager folder created.
  • Memory: Memory overhead was insignificant.
  • Network: Network impacts appear to be insignificant.

Note: This is an interesting shift since the assessment I performed on OpsMgr 2007 agent deployment requirements and which I also integrated into the OpsMgr Unleashed book. For agentless monitoring when compared with previous tests processor utilization has decreased and network traffic has  decreased. Most of this I attribute to three factors:

1) Virtualization platform evolution: The difference in virtualization platforms which were used for testing (Virtual Server 2005 R2 versus Hyper-V in Windows Server 2008 R2).

2) Hardware evolution: The difference in hardware which was used for testing (From a dual processor 1Ghz system to a quad processor 3Ghz system, from a slower IDE drive configuration to a multiple drive SATA configuration).

3) Operations Manager evolution: Original tests were done on OpsMgr 2007 RTM compared with OpsMgr 2012 RTM.

 

Windows Agent-based monitoring versus Agentless monitoring in OpsMgr 2012:

It appears that with current server hardware the resource impacts of agentless monitoring are not significant, so why not use agentless monitoring instead of agent monitoring in OpsMgr? There are several reasons:

  • OpsMgr 2012 supports up to 10 agentless managed systems reporting to a single management server, and 60 in a management group.
  • Agentless monitored systems do not provide full functionality. Most management packs available for Operations Manager do not support agentless monitoring.

However, Agentless monitoring is commonly used in these types of situations:

  • Clustered servers: If your environment has Windows clusters where Operations Manager is monitoring the nodes and proxy is enabled on the nodes of the cluster, the cluster appears as agentless monitored in OpsMgr.
  • Validated systems: These are systems where specific testing processes must occur for any changes to the configuration, including deploying agents. (A validated system is one that has a strict process to validate the functionality of the system anytime software is installed on it.) In these cases, the Information Technology (IT) group responsible for server management wants to avoid even the appearance of affecting an application server, thus avoiding potential liability. If you are in this situation, you can use the agentless management mode to manage servers that might otherwise be unmanaged.
  • Unsupported Windows operating systems: Legacy systems such as NT 4.0 or Windows 2000 that do not support an installed agent are common reasons for using an agentless configuration.
  • Service Manager: System Center 2012 Service Manager currently does not support installation of an Operations Manager 2012 agent. Monitoring Service Manager requires using agentless monitored systems.
  • Minimally monitored systems: For some systems, the goal is only to provide up/down information and minimal additional information – to avoid digging deeply into what is occurring on the system. For these systems agentless monitoring may be a good solution.
  • Licenseless Monitoring: Marnix Wolf discusses this on the OpsMgr 2007 R2 platform at: http://thoughtsonopsmgr.blogspot.com/2009/08/to-agentless-monitor-or-not-that-is.html. While the licensing model has changed in System Center 2012 this may also still be relevant to consider.