Could someone have a look at: HdrHistogram/HdrHistogram#111

@obourgain One of the use of this column (perhaps only use?) is to serve as an X value to plot percentile values with a logarithmic X axis. Example, from the Javascript code of this html chart: http://hdrhistogram.github.io/HdrHistogram/plotFiles.html

If you look at the code you'll see the X tick labels as:

var ticks =

[{v:1,f:'0%'},

{v:10,f:'90%'},

{v:100,f:'99%'},

{v:1000,f:'99.9%'},

{v:10000,f:'99.99%'},

{v:100000,f:'99.999%'},

{v:1000000,f:'99.9999%'},

{v:10000000,f:'99.99999%'},

{v:100000000,f:'99.999999%'}];

If you use a log scale, the log value actually represents the number of significant digits in the percentile. E.g.

And 1/(1-0.9)=10, 1/(1-0.99)=100 etc... and log(1/(1-p))=number of significant digits in the percentile.

log(1/(1-0.9))=1, log(1/(1-0.99))=2 etc... So that log value is particularly useful for spreading out the X axis evenly based on the number of digits in the percentile.

To be short, if you want to plot all percentile values and have a nice view of the details at higher percentile precision, you would then use a series with X = (1/(1-p)) on a log scale and Y=latency.

If you look at the code you'll see the X tick labels as:

var ticks =

[{v:1,f:'0%'},

{v:10,f:'90%'},

{v:100,f:'99%'},

{v:1000,f:'99.9%'},

{v:10000,f:'99.99%'},

{v:100000,f:'99.999%'},

{v:1000000,f:'99.9999%'},

{v:10000000,f:'99.99999%'},

{v:100000000,f:'99.999999%'}];

If you use a log scale, the log value actually represents the number of significant digits in the percentile. E.g.

And 1/(1-0.9)=10, 1/(1-0.99)=100 etc... and log(1/(1-p))=number of significant digits in the percentile.

log(1/(1-0.9))=1, log(1/(1-0.99))=2 etc... So that log value is particularly useful for spreading out the X axis evenly based on the number of digits in the percentile.

To be short, if you want to plot all percentile values and have a nice view of the details at higher percentile precision, you would then use a series with X = (1/(1-p)) on a log scale and Y=latency.