95th percentile for last 30 days

[doh!nut]

I'm sure this is a common question/operation but I don't seem to be able to find the answers via search (or in docs) so apologies if it's out there and I missed it!

My understanding so far is that you must increase the default RRA size for "5 minute average" (or whatever highest resolution you define) to 3x the period in question. E.g. if you want 95th percentile of 1 month you should set to 26280 rows (according to http://forums.cacti.net/download.php?id=11555)

So I do that (and resize my existing RRD files - when I get around to it - as detailed here: http://forums.cacti.net/viewtopic.php?t=11425).

Then how do I go about displaying the 95th percentile value for the last 30 days? Or is there some way I can show the 95th percentile for the last 30 days on a 24/48/72 hour graph?

Do I have to add a graph showing the 30 day period, then the figure shown in the legend is the accurate one (graph itself may not be exactly correct due to consolidation required to display that much data in small number of pixels)?

A related(?) question: 95th defaults to in+out, but I want to consider 95th on this basis:

Take all outbound traffic samples for last 30 days, order highest/lowest and disguard top 5%. Repeat for inbound. 95th = highest of the inbound and outbound values (not the two added together).

I presume that one of the variables used to select which 95th calculation type will control this, but I don't understand the documentation for this very well

Whilst I'm asking questions, can someone confirm what the "timespan" setting is all about for the RRAs? Does that need changing to something different as well?

[rony]

Well, you got step on done. Updating the rra and hopefully resizing your rrdtool files as well.

You need to select the time frame from the time span selector at the top, clicking on the graph will only give you the graphs for those consolidation functions.

Graph variables, Nth Percentile are explained in the documentation. If you have questions about a certain type of Nth percentile, let me know.

[doh!nut]

Well, you got step on done. Updating the rra and hopefully resizing your rrdtool files as well.

Yep - as it happened I made a n00b mistake when upgrading to 0.8.7b (from 0.8.7a) in that I didn't keep a backup of the RRAs, so that made cacti create some new ones of the appropriate size anyway. (I checked this via rrdtool info on the command line)

You need to select the time frame from the time span selector at the top, clicking on the graph will only give you the graphs for those consolidation functions.

I see... so how would I export a graph (i.e. via the 'Graph Export' feature) showing the 95th value for last 30 days?

Graph variables, Nth Percentile are explained in the documentation. If you have questions about a certain type of Nth percentile, let me know.

Thanks, I've had another look. From what I understand, the max "Nth Percentile Type" is the one that behaves as I described - is that correct?

[rony]

Export for billing purposes? Check Addons.

[doh!nut]

No. Actually your billing script isn't suitable for my needs, and it also looks like cacti might not be

Is there someone willing to answer my questions and help out a little? Or is there somewhere else I could pose these questions for a little help?

[TheWitness]

I believe Mr. Rony attempted to answer your questions. You should RTM. The 95th percentile is very flexible.

You however, are looking to discard some data, which RRDtool does not typically do. I would suggest a customization. I have written several functions that do exactly what you want. You would simply read the data into an array, discard the data you don't want, take the Nth percentile and display it as an HRULE or something.

There is not much Cacti can't do, you simply have to put some effort into it.

TheWitness

[doh!nut]

Thanks TheWitness. I apologise - I know everyone is busy with their own things, and cacti is probably very obvious once you understand it, but I've only just started using it so I'm still getting to grips with it.

Also I don't believe I'm asking cacti to discard any data? I want to be able to determine the 95th value for the last 30 days, or maybe for a given 30 day range. The problem is that cacti wants to display a pre-determined range such as 24 hours, a week etc. and uses certain consolidation operations to achieve this.

I do not mind the consolidation for graph display - I've read enough threads on here to understand why that's necessary - but I need the accurate 95th percentile value.

I should explain that I'm trying to create a graph from a switch connecting to our transit provider so that we can see our own 95th percentile measured at our end compared to their measurement - just for information.

I'm thinking that the timespan setting for RRA is maybe relevant? No?

Your help (including Mr. Rony btw!) is very much appreciated, and those functions sound interesting - any chance you could share?

[TheWitness]

Then, you need to maintain your 5 minute RRA for 30 days. It's that simple. It's a simple adjustment to the RRA settings. Keep in mind, that when you import templates, any changes to the RRA's are global. So, in 087 we have provided an option to ignore them.

TheWitness

[doh!nut]

You need to select the time frame from the time span selector at the top, clicking on the graph will only give you the graphs for those consolidation functions.

With regards to the time span selector, this is fine, but the graph previews only show the graphs themselves - not the actual numbers (i.e. I'm most interested in the figure for 95th rather than a line that allows me to take a reasonable guess at it).

Then, you need to maintain your 5 minute RRA for 30 days.

Ok... so far I've changed my 5 minute RRA to 90 days (see my first post in this thread) - but my understanding is/was that the 95th value shown on any graph only applies to the visible time.

For example, if my graph is showing me data for the last 24 hours (even with 90 days of 5 minute RRA, which this 24 hour graph is based on) it will draw the HRULE and 95th value (number) at the bottom corresponding to the last 24 hours? Is this wrong... will it really be the last 90 days in this configuration?

[rony]

Nth Percentile only show for the selected timespan on the graph. If you want to see the 5 Minute Average Nth Percentile for the last 30 days you need to selected the the appropriate timespan at the top of the graph page. This is the only way to view a certain time frame in a non-consolidated manor.

When you do this, rrdtool will use the highest resolution data for the give time frame. Because you have 90 days of 5 Minute Average data, it will use that if you select a time frame with in that stored data.

[doh!nut]

Nth Percentile only show for the selected timespan on the graph. If you want to see the 5 Minute Average Nth Percentile for the last 30 days you need to selected the the appropriate timespan at the top of the graph page. This is the only way to view a certain time frame in a non-consolidated manor.

When you do this, rrdtool will use the highest resolution data for the give time frame. Because you have 90 days of 5 Minute Average data, it will use that if you select a time frame with in that stored data.

Great.

So is there any way I can have Cacti print the value (e.g. 47.52 Mbps) as well as draw the line on the graph (for the 30 day 5 minute avg. Nth percentile)?

[doh!nut]

Answer = yes

I've been concentrating on the "Preview View", but actually if you switch to "Tree View" (with appropriate graphs added) you can use the timespan controls just fine.

More precisely, any view where you're not using thumbnail previews will show this properly - and of course because you can set the timespan selector to precise dates/times you require there's no need to worry about trying to export at the specific time.

Thanks for all the help - I got there in the end

[kdmillerii]

I would like to plot the 95th percentile ingress (how we are billed) by month as a line on the standard interface graph. So if you have the graph time setup over 4 months, the 95th percentile line would be be consistent through each month, not 1 line for the entire timespan.