Calculating % Proc Time and % Core Util Time

[WilsF]

Hi,

I'm trying to write a PowerShell script to collect % Processor Time to % Core Util Time ratio of multiple vSphere hosts using the Get-ESXtop command in PowerCLI.  I've been using LucD's guide to Get-ESXtop (http://www.lucd.info/2011/04/03/hitchhikers-guide-to-get-esxtop-part-2-the-wrapper/) , but can't get results that even come close to matching actual ESXtop data. 

In order to get % Proc and % Core Util Time, I use the UsedTimeInUsec and CoreHaltTimeInUsec LCPU counters, respectively.  Using the derivation method listed for PCPUUtil, I convert these raw values to the percentages from ESXtop: (n – n-1) / ESX-interval / 1E6 * 100

These percentages more-or-less match up with ESXtop for % Proc Time, but the values for % Core Util Time do not make sense.  The differences in the raw CoreHaltTimeInUsec data are roughly two orders of magnitude higher than those of the UsedTimeInUsec numbers leading to percentages that don't make sense (>100%).

Does anyone here have any ideas why this discrepancy is present?

Here's an example of some of my raw numbers:

Differences in UsedTimeInUsec	% Processor Time
323544 81555 746247 298459 500901 499624 664155 545064 626532 207229	6.47088 1.6311 14.92494 5.96918 10.01802 9.99248 13.2831 10.90128 12.53064 4.14458

Differences in CoreHaltTimeInUsec	% Core Util Time
5124259 5124259 4560600 4560548 4651946 4651946 4424210 4424210 4773042 4773042	102.48518 102.48518 91.212 91.21096 93.03892 93.03892 88.4842 88.4842 95.46084 95.46084

Appreciate any ideas!

Wilson

[LucD]

Would you mind sharing the script you have so far?
And which PowerCLI and vSphere version are you using?

There might be some differences with my original post in later versions.

[WilsF]

I've attached the function that does the calculations.

I'm interested in 2 * % Proc Time/ % Core Util Time ratio, so I end up canceling out factors (ESX_interval, 1E6, and 100) in both the numerator and denominator.  Also, since the counters are returned per logical CPU, I sum the differences for each LCPU before taking the ratio.  To assist in debugging, I've added the respective percentages back into the returned object.

[WilsF]

Does anyone have any ideas based off of the cmdlet I provided?  Assuming I'm calculating this correctly, either the raw data returned for the counters are wrong or the counters I'm using are wrong.

[LucD]

Hold on, I'm preparing a blog post on the subject.

[WilsF]

Thanks LucD!  Out of curiosity, do you have an ETA on the blog post?

[LucD]

Just finishing the module now, hope to publish tomorrow.

A quick preview, don't mind the negative values

[WilsF]

The preview looks promising!  Looking forward to the post.

[aseymour4]

Hi Luc,

Have you had a chance to upload that module to your blog? I'm very interested to read your thoughts on the topic!

[LucD]

I know, I'm slacking.

Discovered a bug in my code, think I got it fixed now.

[LucD · Accepted solution]

The post is up, see Get-EsxTop – Another Look

It only contains limited metrics for now.

I hope to expand soon.

[WilsF]

This is exactly what I was looking for times one hundred.  Thank you for writing the post with a module to go along with it!

[aseymour4]

Thanks, Luc! WilsF and I greatly appreciate the thoroughness of your analysis and documentation.