Performance data, not adding up

[soleblazer]

Curious to see if anyone else has seen this, maybe I'm just missing something stupid.

I am simply trying to look at network data for a one day, one week, one month period.

When I look at an esx servers network usage counter for the past day the max value (and highest point of the graph) is 29,593KB/s

When I change the same graph to show the past week the max value is 12,307

When I change the same graph to show the past month the max value is 8,013.

Now, last I checked the current day is in the past week, and the current day is in the past month.

The spike happened early this morning so its not like it just happened and the past week/month is simply not picking it up.

Anyone else see strange stuff like this?

[hicksj · Accepted solution]

That's correct. The data decays over time.

Your daily numbers are 5 minute averages.

Weekly stats are 15 minute samples, monthly stats hourly.

So you'll find that peak values are quickly leveled out.

Example

Time x, value 0

Time x+5, value 30

Time x+10, value 0

You'll see the 30 on your daily graph, but it quickly drops to 10 on the weekly. Monthy needs more points, but lets say the remainder are 0 too... then the 30 becomes 2.5.

The purpose of the long term graphs is to display trending, not actual performance. This also keeps the database sizes more manageable.

Regards,

J

[soleblazer]

Ah ok, that makes sense then, thanks for the explanation.

[kyfbof]

I am not replying to your question, but I know this is the right people to ask the related issue. What do we need to get the monthly weekly stat/charts?

When trying to customize the chart, the "Last ...." and "From: ... To: ..." fields are all greyed out.

[hicksj]

The Last & From/To fields are only available when you select "Custom."

[dsmedtp]

I could accept averages being leveled out in time, but not peaks.

There is no use in keeping statistics at al, if all numbers are going to be inaccurate after some time.

I'm currently trying to make a performance report on a VI3 installation i did for a client, but these statistics are driving me nuts.

Many of the graphs are either not available, or not making any sense.

The total memory usage for the cluster (2 servers with 12GB each) is currently showing 55%. The other statistics are showing 'not available', so i cannot get a real value. Now i know this number is wrong, because i can see that only about 7GB (about 30% of the total 24) is in use.

Now when i look at the Hosts separately, i see one uses 24% and the other 31% of its memory, hence the 55, they just get added up instead of averaged.

And there are many more weird thing going on with the statistics, enough to make me decide the graphs look nice, but apart from that are totally useless.

[FredPeterson]

I concur. Long term graphing is not meant to showcase performance so much as highlight if upgrades are actually necessary. If long term graphing is showing constant high usage, then you may need to look into an upgrade.

Maybe. You cannot just use long term or short term, they have to be used in conjunction along with an understanding of what a system is actually doing.

If I saw sustained high water marks in long term graphing, I'd be concerned about the system - after determining possible causes.