I'm in the process of configuring a performance monitor(s) against all servers in my domain. I'm using WhatsUp Gold 16.2 to send an alert when we reach certain thresholds. I recently read an article by Microsoft that up to 150 Pages/sec was an acceptable threshold. The problem with WUG is that the smallest unit I can measure is one minute. I thought about multiplying 150*60 and sending an alert on 9000 plus, but that doesn't seem like that'd be a good indication of memory pressure
One of my sysadmins created a monitor to send an alert once we had over 4000 page faults in 30mins. I don't think this any better than my proposed monitor above.
An example of the 4000 page fault monitor:
Physical, not a VM Server 2003 R2 w/SP2 Functions as a Domain Controller in a remote office 32 bit 4 GB RAM 4 GB pagefile 2 procs 2.8GHz
RAM 30 day graph: Average: 18.5% Max: 26.6%
Virtual Memory 30 day graph: Average: 5.38% Max: 6.66%
The above system fired an alert with over 4400 page faults in 30 minutes. It is obvious by the stats that there is no memory pressure to speak of, nor is the 4000 page fault monitor a good indicator of a problem.
Most of my environment is virtual (VMWare & Hyper-V), mostly VMWare. Most all servers are a variation of Server 2008, very few 2003, and several 2012. Have very few that are physical.
My question really boils down to:
What should I monitor and what threshold to fire an alert on with a sample rate as high as one minute?
