Acceptable ranking error for approximate quantiles calculation.
Consider we have 1000 measurements and calculating 2-nd quartile (50% or 0.5).
Exact quantile calculation must return 1000 * 0.5 = 500 item
of ordered list of measurement but we need to keep all 1000 measurements
in memory.
Approximate quantiles calculation guaranted to return an item between
1000 * (0,5 - Epsilon) and 1000 * (0.5 + Epsilon). So for default value of 0.01 value between490and510` position will be returned,
greatly relaxing memory requirements.
Lesser values means greater precision and greater memory and cpu requirements,
while greater values means lesser precision but lesser memory and cpu penalty.
Quantiles window size in seconds. NOC maintains 2 quantile windows -
temporary and active one, purging active and swapping windows
every default_quantiles_window seconds.