95% bar and calculation in graphs of some devices

[acceleratebiz]

I noticed on some devices, the 95% bar and calculation isn't showing up for their graphs. Why is this and how do I get that again?

[rony]

Depends on the graph... Can you post screen shot of the graph items for the graph in question?

[acceleratebiz]

attached is the screenshot. seems like all the graphs of our "Internap Gigabit Edge Router" don't have the 95% bar. So I tried to edit one of the individual graphs under "Graph Management" and there was a graph template I could choose called "Interface - Traffic (bits/sec, 95th Percentile)." The one already selected was without the "95th Percentile." So I was going to change it and I noticed "Please note that graph data may be lost if you change the graph template after one is already applied." I can't have my graph data lost! What do I do?

[rony]

You shouldn't lose your data, those graphs, because of the data query, should use the same data sources.

[acceleratebiz]

I finally had the nerve to change the graphs to bits/sec with 95 percentile and it worked great. Didn't lose any history. But the problem is now it's been about a month and customers all see 0mbps 95percentile usage. The 95percentile bar just sits flat at the bottom of the graphs even though there are, in some cases, considerable bandwidth usage all month long. What's the deal? How can I fix this?

Please help, thanks!

[rony]

If you are using it for billing purposes, I would suggest altering your rrd's to store 5 minute averages for 1 year of time. This should resolve that issue.

[knobdy]

How do you do that?

We're not using it for billing purposes, but this sounds like a good idea.

[gandalf]

See my posts at Howto define very BIG rra "without data loss" and Howto RESIZE existing RRAs of existing RRDs
Reinhard

[acceleratebiz]

I followed the instructions to increase the size of the RRDs, but that had no effect on the 95% calculations... still zero always . Here's what I did:
[root@manager rra]# ./resize.sh -f "internap_gigabit_edge_router_traffic_in_169.rrd" -r "0 1 2 3 4 5 6 7" -o ../upsized_rra/ -g 8000 -i
-- RRDTOOL INFO internap_gigabit_edge_router_traffic_in_169.rrd ...
ds[traffic_in].type = "COUNTER"
ds[traffic_out].type = "COUNTER"
rra[0].cf = "AVERAGE"
rra[0].rows = 600
rra[1].cf = "AVERAGE"
rra[1].rows = 700
rra[2].cf = "AVERAGE"
rra[2].rows = 775
rra[3].cf = "AVERAGE"
rra[3].rows = 797
rra[4].cf = "MAX"
rra[4].rows = 600
rra[5].cf = "MAX"
rra[5].rows = 700
rra[6].cf = "MAX"
rra[6].rows = 775
rra[7].cf = "MAX"
rra[7].rows = 797
[root@manager rra]# ./resize.sh -f "internap_gigabit_edge_router_traffic_in_169.rrd" -r "0 1 2 3 4 5 6 7" -o ../upsized_rra/ -g 8000
-- RRDTOOL RESIZE internap_gigabit_edge_router_traffic_in_169.rrd RRA (0 1 2 3 4 5 6 7) growing 8000.. (94664).. RRA#0#1#2#3#4#5#6#7.. (1118664).. Done.
[root@manager rra]#


and then, of course, I moved the upsided_rra folder contents into rra/. Did I not do this right? Do I also need to go through that whole long process to "define very BIG rra 'without data loss'?"

[rony]

Btw, these changes only affect data going forward.

The data you had in the past has been consolidated away.

[acceleratebiz]

It hasn't fixed my 95% calculation in the past or since resizing the RRD file though.

[gandalf]

[root@manager rra]# ./resize.sh -f "internap_gigabit_edge_router_traffic_in_169.rrd" -r "0 1 2 3 4 5 6 7" -o ../upsized_rra/ -g 8000
-- RRDTOOL RESIZE internap_gigabit_edge_router_traffic_in_169.rrd RRA (0 1 2 3 4 5 6 7) growing 8000.. (94664).. RRA#0#1#2#3#4#5#6#7.. (1118664).. Done.
[root@manager rra]#

resizing _all_ rra's doesn't seem that lucky in my optinion. There is no need to increase the number of data points for the weekly/monthly/yearly rra's. It would have been ok to simply increase rra 0 and 4 (these store the data for 5 min intervals). Maybe I should point this out more clearly in my howto.
You may downsize those rra's 1,2,3,5,6,7 using rrdtool tune to avoid messing storage and loosing update time. Contact me if in doubt.
As to the 95th percentile problem, I can't help, sorry
Reinhard

[rony]

Let me clarify what I was saying earlier.

Ok, it's good that you have resized your RRD files. I do agree with LVM, you only need to alter the 5 minute averages. The others are just wasting space and slowing down you RRD update times.

Concerning 95th, the functions are very particular about getting the correct resolution. Which is basically the interval, 5 minute, 30 minute, 2 hours, 1 day, which are related to the rrd definitions. You will not see accurate data for a month, until a month from now. This is because, although you have altered the rrd files and the number of rows, the data you had in the past can not be pulled back into the 5 minute intervals. It has been consolidated into the 30 minute, 2 hours, and 1 day rra storage.

[acceleratebiz]

ok, so guys if I understand you properly, resizing all 8 rra's isn't necessary... but it doesn't hurt either (except for minimal extra storage required and tiny performance decrease) right?

the 95th percentile I'm very concerned about. I need to get that working. Can I query the RRD database like a sql database to get all the values so that I can write a little script to compute the 95th percentile?

What do you guys suggest?

[rony]

Performance can be a little to a lot.... Think of it this way, if you have 20 megs rrd files, this can be an issue.

Yes, you can export the data from the rrdfile. Use "rrdtool fetch" or "rrdtool export", read the man pages for syntax.