In queuing theory-based analysis, we always assume arrivals to be according to a Poisson process. "Why we do so, why not it's uniform or some other distribution?" This was a question raised by my supervisor & a colleague while I was preparing for my Qualifying Exam. As I was not able to answer well, my supervisor gave some hints. I was worried that the same question may come up during the exam so searched for an answer on web. Unfortunately, no specific answer was found. So I'm trying to lay down an answer by adding bits & pieces from here & there & my supervisor's answer.
It can be answered by looking at the properties of a Poisson Process. Recall that Poisson Processes are used to model statistically rare events.
A counting process {Nt, t ≥ 0} is a Poisson process if:
It can be answered by looking at the properties of a Poisson Process. Recall that Poisson Processes are used to model statistically rare events.
A counting process {Nt, t ≥ 0} is a Poisson process if:
- N0 = 0
- Nt has stationary independent increments
- Nt1 - Ns1 is independent from Nt2 - Ns2
- Memoryless
- Inter arrival times are independently & identically distributed set of exponentially distributed random variables
- P{NΔt = 1} = λ(Δt) + o(Δt)
- P{NΔt = 2} = o(Δt)
Property 2 (memoryless) is the key. With uniform distribution if we say, "probability of an arrival of a packet is 0.1," if a packet doesn't arrive between [0 - 9] it has to arrive at t = 10. In a natural system, such arrivals are not necessary. A packet may or may not arrive or if an Earthquake doesn't happen in Yellowstone National Park for last 9 years it doesn't necessary mean it must occur in the coming year. Hence this reflect the natural behavior & makes the analysis easier.
Property 3 & 4 says that there will only be one arrival at a time (no batch arrivals). This holds for a small time interval Δt → 0. Hence, we only need to worry about 1 arrival at a time. Even if 2 Earthquakes occur during 5 minutes, it will happen one after another. This further simplify the analysis.
Given these properties Poisson is a better & simple approximation.
Please feel free to suggest any corrections or additions....
Property 3 & 4 says that there will only be one arrival at a time (no batch arrivals). This holds for a small time interval Δt → 0. Hence, we only need to worry about 1 arrival at a time. Even if 2 Earthquakes occur during 5 minutes, it will happen one after another. This further simplify the analysis.
Given these properties Poisson is a better & simple approximation.
Please feel free to suggest any corrections or additions....
Comments
λ = lim Δt → 0 P{N_(t + Δt) = N_t + 1}/Δt
therefore it make sense to define the 3rd property like that.