Monitor Server High-CPU Utilization

[TristanForward1]

We have a pretty standard ArcGIS deployment

- Portal
   - Host Server
   - Image Server
   - Map (3)

- UnFederated Server
- Monitor Server

- DB
   - DataStore (Relational/Tile Cache)
   - Spatial-Temporal DataStore

   - Oracle DB

In Monitor with the centralized recommended specs of 4 CPU cores and 16 GB RAM. We have a total of 62 counters for Production all set to the longer time intervals of recommended settings.

We have run the stagger collection times script and tried to reduce collection times for certain less critical aspects of the system such as RDP, however we are still getting High-CPU spikes with Monitor Server.

Any suggestions?

[Anonymous User]

Try adding 2 more CPU cores to this VM.

[FrankPizzi]

Spikes are common.  You should see the spikes as the monitor runs on different intervals. 

[Nryan]

Hello! I just had the pleasure of debugging this oddity in our server.
This happens because of the scheduled observers. The observer that grabs the system metrics is "Metrics - Host Windows". This (for us) is scheduled to occur every 1 minute. Other observers were also scheduled to occur every 1 minute.

In other words, the CPU Utilization is the exact CPU Usage % of when the "Metrics - Host Windows" observer runs. Since other observers run at this exact same time, it causes a brief CPU spike.

If you measure the CPU Spike from Performance Monitor, it exactly matches what Monitor Displays. 

Setting "Metrics - Host Windows" to check every 1 minute, and others at other times, creates a much more accurate chart. An example is below. 

In below example, system metrics are recorded every 1 minute while all other observers run every 5 minutes. You can see the dramatic spike that happens every 5 minutes.

To be honest, it's a pretty serious configuration flaw with observer scheduling. ESRI should include an option to where we can stagger the observer run times. Example, start running at "00:00:30" with a repeat interval of 1 minute. This would perfectly fix this problem!

Please vote for this idea here!!! https://community.esri.com/t5/arcgis-monitor-ideas/add-start-time-setting-for-observers/idi-p/171287...