forzenfire
Contributor
Contributor

Performance Thresholds : Verification Required

I have gone through couple of forums and articles to identify these thresholds, could someone please advise if this can be used as a general best practice, leaving out the business requirement whether it is business critical workloads or not.

Please view the attached spreadsheet for the metrics.

Thank you.

Kind regards,

Karthik

0 Kudos
6 Replies
jklick
Enthusiast
Enthusiast

Feedback:

  • CPU Ready - I would not use a ms value for the threshold. The ms thresholds you have listed should get multiplied by the number of vCPUs allocated to the virtual machine. Using 5% and 10% is safer, but requires coverting from ms.
  • Memory usage is synonomous with Consumed or Active, depending on the object type you're looking at. Memory usage for a VM is the percentage of allocated memory resources that are "active". Memory usage for a host or cluster is the percerntage of total configured memory that is "consumed".
  • If you're measure disk latency from a VM, I think you're limited to a single metric that is the total roundtrip latency for a disk command. However, at the host level you can break it down to device latency, kernel latency, and queue latency. Device latency is your more typical form of latency and it measures the round trip between the host and the storage. However, kernel latency and queue latency are typically 0 in healthy environments. Anything greater than 0 can indicate a problem. At the VM level, you're only getting the total latency and can be missing part of the story if there is any kernel or queue latency mixed in.

Hopefully this helps.

@JonathanKlick | www.vkernel.com
forzenfire
Contributor
Contributor

Hi,

I am still wondering are there official VMware documentation to advise the thresholds for these parameters.

I have gone across performance troubleshooting, performance best pratices documentation but finidng it hard to find the thresholds as a good advise from vmware.

Cheers

0 Kudos
jklick
Enthusiast
Enthusiast

The best I've seen is the Performance Troubleshooting Guide. Its last iteration was for vSphere 4.1, but it is largely still accurate, particularly the mentioned thresholds. Anything you do not find in that guide (or that deviates from the guide) is found by personal experience or amongst the blogs of the many VMware experts out there.

@JonathanKlick | www.vkernel.com
forzenfire
Contributor
Contributor

Thank you, i was doing that only and have found all these thresholds. It would be good to have a documentation from VMware outlining these thresholds as best practices.

Cheers and i think this is going to be a dead thread after this.....was really keen to get some pointers from the community.....

0 Kudos
sidbryd
Enthusiast
Enthusiast

From a storage perspective I would also consider monitoring the queue length as if this is consistently high can indicate a performance bottlkneck.

Regards

Andrew

0 Kudos
mcowger
Immortal
Immortal

All your network and disk values are worthless without an understanding of the application's requirements.

Most many of the DBs I work with, anything over 2ms disk response time is catastrophic.  For some of the big data ones, anything under 30ms is basically unachieveable.  I might as well suggest that every speed limit in the country should be 35 MPH, without regard to the nature of the road.

Also, it doesn't make sense to measure IO in KBps with a critical value of '75%'. 

--Matt VCDX #52 blog.cowger.us
0 Kudos