Curious about the Sampling - Simulation - Percentiles section from SREcon22 - Amsterdam. I'm not certain how the error % is being calculated.
In the talk you state that the formula calculates the error, "the error is 1 minus p divided by n times p." I can't seem to get the same +/- that you do in the examples.
On slide 90 there are 1000 latency values and 10% sampling, so N * p = 100 values.
(1 - .10) / (100) = 0.009
√(0.009) = 0.0948
The answer according to the slide is 36% error, what am I missing to get that result?
If I throw the numbers into your sampling calculator, I do get Relative Error of 9.49% for the request rate, which is a similar number, but the percentile relative error is calculated to be 17.2%
There is the 'divide by p' section on slide 89, but that doesn't return the expected 36% either.
0.0948 / .1 = 0.948
Slide 90 does then put p90 in the 'divide by p' section which also doesn't help me.
Thanks! :)
Curious about the
Sampling - Simulation - Percentilessection from SREcon22 - Amsterdam. I'm not certain how the error % is being calculated.In the talk you state that the formula calculates the error, "the error is 1 minus p divided by n times p." I can't seem to get the same +/- that you do in the examples.
On slide 90 there are 1000 latency values and 10% sampling, so
N * p = 100 values.The answer according to the slide is 36% error, what am I missing to get that result?
If I throw the numbers into your sampling calculator, I do get Relative Error of 9.49% for the request rate, which is a similar number, but the percentile relative error is calculated to be 17.2%
There is the 'divide by p' section on slide 89, but that doesn't return the expected 36% either.
0.0948 / .1 = 0.948Slide 90 does then put p90 in the 'divide by p' section which also doesn't help me.
Thanks! :)