rp

Documentation

Overview

Package rp implements a random process where identifies peak load.

1. Identify peak load The current approach: Mann–Whitney U test

Consider a count process: N1= (n_1, ..., n_{p/2}), N2 = (n_{p/2}, ..., n_{p})
and our goal is to check if N2 is significant higher than N1.
Let hypothesis: H0: \mu_{N_1} \leq \mu_{N_2}, and H1: \mu_{N_1} > \mu_{N_2}
Then rejection region is:

  z = \left|\frac{\mu_{N_2} - \mu_{N_1}}{\sigma_{N_1}}\right| \geq z(c)

We reject H0 under given confidence level.
Since this approach assume the count process obey normal dist, it is
recommended to have a window size higher than 15.

2. Check if system resource can handle next peak load: Poisson + Moving average The current approach: Markov Poisson process + Moving average

Consider p windows, and each number of events obey Poisson process,
then \lambda = \frac{1}{p} \sum_{i=1}^{p} n_i, and in the next window,
the probability of having k number of events is:

  P(N=k) = \frac{\lambda^k \exp{-\lambda}}{k!}, k = 0, 1, ...

Assume a system can handle Q requests, then we have:

  P(N\leq Q) = \sum_{i=0}^{Q} P(N=i) = \sum_{i=0}^{Q} \frac{\lambda^k \exp{\left(-\lambda\right)}}{k!}

Index

• type RandomProcess
  • func NewCountProcess(maxsize, confidenceLevel float64) RandomProcess

Types

type RandomProcess

type RandomProcess interface {
	Store(nevent float64)
	Significant() bool
	Acceptable(k float64) (int64, bool)
}

RandomProcess defines two type of tests for given observations.

func NewCountProcess

func NewCountProcess(maxsize, confidenceLevel float64) RandomProcess

NewCountProcess creates a new counting random process. maxsize represents the observation window size of the count process. confidenceLevel gives the confidence level of the significant test and acceptance test, it is recommended to set below 0.05.