counter wrap and data source "maximum value"

[jlg]

hi folks,

i have a snmp stat that reports "seconds of use" as a counter. this stat internally wraps at 43,000 seconds. when it wraps, the graph has a huge spike since this is nowhere near the 32 & 64 bit counter wrap limits that rrdtool can automatically handle.

so researching the problem, it looks like a possible solution is to set the data source's "maximum value". research also says that i should be setting "maximum value" to the "maximum value per second" for my stat. well, the "maximum seconds of use per second" can only be "1"...so i am tempted to use this value.

however, i hesitate because i know that there is the issue of "imperfect sampling intervals". ie, the poller cannot sample at exact 300 second intervals and rrdtool will interpolate to compensate. i am concerned that if i set the maximum to "1" and the interpolated value is 1.0000001, that i will end up with holes in my graph.

so is it wise to set the "maximum value" to a slightly higher number like "1.5" to allow for fudge in the system?

or should i be solving this problem a different way?

thanks,
jlg

[gandalf]

Chopping of values that exceed maximum is done after building the difference to the preceding value (for COUNTERS) and after computing the rate (division by interval length). So you should be correct. But this is from theory ...
Reinhard

[jlg]

for the benefit of others, i will answer my own question here. ***YES*** you should allow for some fudge factor due to interpolation.

in my case, there is no way to get more than 1.000000 seconds out of a time period of 1.000000 seconds. so i had set my data source maximum to "1". and most of the time, this worked just fine. however, reviewing the data after 3 days, i found one unknown (NaN) in my rrd which i assume is due to going over "1".

Code: Select all

<!-- 2007-01-09 19:20:00 PST / 1168399200 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-09 19:25:00 PST / 1168399500 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-09 19:30:00 PST / 1168399800 --> <row><v> 9.9667777778e-01 </v></row>
<!-- 2007-01-09 19:35:00 PST / 1168400100 --> <row><v> 9.9998888889e-01 </v></row>
<!-- 2007-01-09 19:40:00 PST / 1168400400 --> <row><v> 9.9668881506e-01 </v></row>
<!-- 2007-01-09 19:45:00 PST / 1168400700 --> <row><v> 9.9667774086e-01 </v></row>
<!-- 2007-01-09 19:50:00 PST / 1168401000 --> <row><v> NaN </v></row>
<!-- 2007-01-09 19:55:00 PST / 1168401300 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-09 20:00:00 PST / 1168401600 --> <row><v> 9.9668881506e-01 </v></row>
<!-- 2007-01-09 20:05:00 PST / 1168401900 --> <row><v> 9.9997785161e-01 </v></row>
<!-- 2007-01-09 20:10:00 PST / 1168402200 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-09 20:15:00 PST / 1168402500 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-09 20:20:00 PST / 1168402800 --> <row><v> 1.0000000000e+00 </v></row>

jlg

[gandalf]

Thanks for posting this. IMHO, this should work as you said, but it is better to SEE it actually
Reinhard

[jlg]

more updates:

1. cacti complains if i try to set the data source "maximum value" to any number with a decimal point. at least...in the version that i'm running which is debian sarge's version which i understand is a probably 1 or 2 years old. so i had to set it to "2".

btw, it would probably be helpful to list somewhere that "maximum value" is the "maximum rate per second" when using COUNTER. otherwise, i would naturally assume that i should put in 43,000 for my case.

2. note to #1 above: i can use "rrdtool tune <file.rrd> --maximum <ds>:1.1" to set the value to "1.1". so the limitation is in cacti.

3. the other possibility for NaN in my data is a wrap since my data source wraps at 43,000 instead of 4GB. however, i have been snapshotting the raw data and found no wrap at the date listed in my above data. but there is an inconsistency that i need to uncover before i can be certain of the above conclusion. i'll update here when i know more.

jlg

[gandalf]

2. note to #1 above: i can use "rrdtool tune <file.rrd> --maximum <ds>:1.1" to set the value to "1.1". so the limitation is in cacti.

jlg

Please create a bug report at http://bugs.cacti.net for that issue
Reinhard

[jlg]

2. note to #1 above: i can use "rrdtool tune <file.rrd> --maximum <ds>:1.1" to set the value to "1.1". so the limitation is in cacti.

jlg

Please create a bug report at http://bugs.cacti.net for that issue
Reinhard

will do...

jlg

[jlg]

update:

after having set the maximum to 1.1, i now see values larger than 1.00000

Code: Select all

<!-- 2007-01-11 13:00:00 PST / 1168549200 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-11 13:05:00 PST / 1168549500 --> <row><v> 9.9337763012e-01 </v></row>
<!-- 2007-01-11 13:10:00 PST / 1168549800 --> <row><v> 1.0032778883e+00 </v></row>
<!-- 2007-01-11 13:15:00 PST / 1168550100 --> <row><v> 1.0000111483e+00 </v></row>
<!-- 2007-01-11 13:20:00 PST / 1168550400 --> <row><v> 1.0033222222e+00 </v></row>
<!-- 2007-01-11 13:25:00 PST / 1168550700 --> <row><v> 1.0000111111e+00 </v></row>
<!-- 2007-01-11 13:30:00 PST / 1168551000 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-11 13:35:00 PST / 1168551300 --> <row><v> 1.0000000000e+00 </v></row>
<!-- 2007-01-11 13:40:00 PST / 1168551600 --> <row><v> 9.9337763012e-01 </v></row>
<!-- 2007-01-11 13:45:00 PST / 1168551900 --> <row><v> 1.0032778883e+00 </v></row>

jlg