VCenter Advanced performance data charts and Sexi Graph charts do not match

[torck82]

Hi, I'm comparing the charts for all the statuses; the data does not match. For example, CPU usage in Percent and CPU utilization in Sexigraf, at 8.50 am Vcenter, shows a CPU spike at 90 Percent, but Sexigraf said it's only 61.5 percent.

[rschitz]

Hi, what's the time range you're looking at?

[torck82]

last 1 hour. 

[rschitz]

On the last 1h preview of the vCenter, you got 20 seconds of data granularity, in SexiGraf we collect those data every 5min and store the median of those 15 metrics (15*20sec=5min) to optimize data usage. After 24h we start reducing the granularity for data usage and performances like so : 5m:24h,10m:48h,60m:7d,240m:30d,720m:90d,2880m:1y,5760m:2y,17280m:5y

[torck82]

Thanks, so it won't see the spikes since it will just do the median based on when it collects the data? Would there be a way to see the data more granular for performance troubleshooting on a VM?

…

On Tue, Jul 18, 2023 at 12:53 PM Raphaël SCHITZ ***@***.***> wrote: On the last 1h preview of the vCenter, you got 20 seconds of data granularity, in SexiGraf we collect those data every 5min and store the median of those 15 metrics (15*20sec=5min) to optimize data usage. After 24h we start reducing the granularity for data usage and performances like so : 5m:24h,10m:48h,60m:7d,240m:30d,720m:90d,2880m:1y,5760m:2y,17280m:5y — Reply to this email directly, view it on GitHub <#349 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AM67EQRAIF7QEX7WCAR2IU3XQ25PDANCNFSM6AAAAAA2OMIGZ4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[rschitz]

Only for few metrics only because we also collect quickstats (which is basicaly the same thing as median) for even faster processing, so that wouldnt be really helpful in you situation. for troubleshooting you should use esxtop really IMHO

[torck82]

Thanks for the help and information. 

[rschitz]

your welcome. you have a specific issue with a vm, maybe i can help you ?

[torck82]

I do a lot of VDI troubleshooting and I look at cpu ready time and cpu usage percentage, but since vcenter doesnt hold on the data that long. I thought sexigrapf would but I would need it to be more real time and granular. 

[rschitz]

it cant be more realtime than 20 sec anyway since that's the minimum the api can provide, below that you need esxtop and having worked in many big vdi env, even with 5min of cpu usage, ready and demand avg, you should have more than enough to tell if something is wrong (vcpu/pcpu ratio also help), otherwise it's more tricky than simply over-commitment.
I'd be interested in example to prove me wrong :)

[torck82]

If I could have the same realtime vcenter charts in 20 seconds and have longer history that would be helpful for me. If not I totally understand. Thanks for the support. 

[rschitz]

that would require a lot of changes and of course wouldnt be possible over night but you're not the first one to ask me so i'm really starting to consider it.
What would be a minimum retention for those 20sec before we start to reduce the granularity? 2 days, 1 week?
also, what would be the number of vms the appliance would collect data from?
thanks for your feedback

[torck82]

One month would be nice. All Vm’s in Vcenter if possible my current environment has about 2300 VM's. Thanks

[rschitz]

according to https://m30m.github.io/whisper-calculator/ that would be 46.89 Gigabytes only vms instead of 600.81 Megabytes. For reference, we deployed a single sexigraf appliance in a 220K+ VMs env with 50+ vCenter and the total storage consumption was ~100GB :)
i'll try on cpu and latency related metrics and let you know.

[torck82]

Thank you.