Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

170687

October 24th, 2013 14:00

Inaccurate representations of vCenter metrics in Foglight

So after waiting for more than half a year for 7.0, it appears that it does not fix inaccurate metric representations.  Do you think the following CPU metrics depict an accurate picture?

vCenter CPU Utilization graph for a sample VM:

Capture1.JPG

vFoglight CPU Utilization graph for the exact same VM:

Capture2.JPG

I had another discussion open on this topic titled "CPU Ready Time metric off after upgrade from 6.7 to 6.8", but decided to open a new one because the title may have been misleading.  It is the same issue and has not been resolved now since 6.7.  Whatever they changed in 6.8 has made Foglight almost useless to me (unreliable at best) and I am not sure why I am even paying maintenance on it now that I have to validate everything against vCenter or ESXTOP.

2 Posts

November 14th, 2013 13:00

Please go to the Data Browser and take screenshot of the raw metrics for both 7.0 and 6.7.  We can then determine if it is a data collection issue or a charting issue. I think you will see that the data is far better in 7.0.

First set the time zonar to the last 1 hour to be consitent with what we will see in the VI Client.

Dashboards – Configuration – Data

Then in the center pane VMware – Virtual Machines – (VM Name) – Host – CPUs – Utilization

The data will look like this.

Also post a concurrent screenshot of that VM’s CPU utilization as shown in the VI Client.

data.png

viclient.png

171 Posts

October 24th, 2013 15:00

What metric value is that graph using for CPU ready? Min/Max or Average?

171 Posts

October 24th, 2013 15:00

We collect data every 5 minutes (by default) and that is comprised of the 20 second samples for the last 5 minutes.  We use those values to calculate min/max/average values during the collection cycle.  The historical views in vSphere are pulling 1 value for the metric every 5 minutes, so we should never match because the data sets are always different and even if they were the same it's virtually impossible for the polling to overlap

Even on the realtime vs. foglight, because of how we poll for the data it will never match exactly, because we don't store those 20 second chunks.

October 24th, 2013 15:00

I understand that it will not match exactly.  It you look at my previous discussion post you can see how drastically it changed from 6.7 to 6.8.

October 24th, 2013 17:00

In 6.7 it doesn't seem to change the graph whether I select min/max, or average.  That post focused on CPU Ready time because that is where I first noticed it.  It was later that I realized that all of the counters seemed to be presented different in 6.8 (and 7.0), than they were in 6.7 and earlier.  I had conference calls and webex sessions with support and R & D and I hope I don't have to start over again.  I guess my main question, and frustration, is why did it have to change?  Either it was not properly presented in 5.x and 6.x (up to 6.7) and I was making misguided resource judgement calls all along, or it is not presented correctly now.  Perhaps that is a bit too simplistic, but that is where I am now.

October 29th, 2013 12:00

I upgraded my 6.8 production instance to 7.0.  I am still not sure of what I am seeing for CPU metrics.  vCenter is understandably more detailed for the last hour view, but what I am getting from 7.0 (and 6.8 previously) is still no good.  6.7 rolls it up much better than 7.0, as illustrated below.  Please explain what I am seeing with 7.0.  Both Foglight graphs are configured with the same metric (CPU Utilization) and both are set to Min/Max.

Capture3.JPG

6.7 with the same VM

Capture5.JPG

7.0 with the same VM

Capture4.JPG

November 5th, 2013 18:00

That is CPU Utilization, not CPU ready.

171 Posts

November 5th, 2013 18:00

Or I can't read today...  Not sure why that would change so much then, but we did change the way we collect so something is most likely happening in there.  Perhaps John will chime in when he gets a moment.

171 Posts

November 5th, 2013 18:00

Is that cpuReady or the cpuReadySummation on the 7.0?  Have you tried the summation one?

November 14th, 2013 00:00

We are doing additional investigation and I will follow up with more detail.

November 14th, 2013 18:00

The 7.0 graph now looks more accurate than the 6.7 graph.  Maybe it is sampling between spikes, or the 7.0 graph that was posted above was using bad data from 6.8?  Those screenshots were taken shortly after the 6.8 to 7.0 upgrade.

Here is what I am seeing today:

vCenter

vCenter.jpg

7.0

7.0.JPG

6.7

6.7.JPG

7.0 Data

7.0.2.JPG

6.7 Data

6.7.2.JPG

No Events found!

Top