GOLD STANDARD · QUEUEING THEORY · HOW X WORKS
Find the decision hidden inside the model
A busy period ends when a queue that starts with one customer first returns to empty. Its customer count is a branching total, but its duration is an extinction time measured on the clock. A Laplace-transform fixed point keeps the entire distribution visible.
This guide connects the Gold Standard of Queueing Theory, Gold Standard of Education, and How X Works hub.
Chapter contents and reading routes
Choose the route closest to your present task. Every chapter remains expanded, so the article also reads naturally from start to finish.
Orient the duration problem
Build and calculate the transform
Test stability and inversion
Transfer the mechanism
Use evidence and next routes
CHAPTER 1 OF 12 · ORIENT THE CLOCK
1. Duration asks a different question from total progeny
Start an empty single-server queue with one arrival. The busy period lasts from that arrival until the system next becomes empty. Counting customers served tells us the size of the family tree. Duration also keeps every service time and the calendar placement of descendant work.
A busy period of three short services can finish before one long service. The count distribution and duration distribution are linked but not interchangeable. The clock-time problem needs a joint branching argument or its Laplace-transform summary.
CHAPTER 2 OF 12 · ORIENT THE CLOCK
2. State the M/G/1 contract
Assume Poisson arrivals at rate λ, independent identically distributed service times S with distribution B, one work-conserving server, unlimited waiting room and no abandonment. Let T be the duration of the busy period begun by one customer.
During the initiating service, arrivals form independent child busy periods that must all finish before the original period ends. Poisson independent increments and independent service times make those descendant periods statistically repeatable. The traffic intensity is ρ=λE[S].
CHAPTER 3 OF 12 · BUILD THE TRANSFORM
3. Derive the transform fixed point
Let β(s)=E[e^(−sT)] and let B*(u)=E[e^(−uS)] be the service-time Laplace–Stieltjes transform. Conditional on S=x, the initiating service contributes e^(−sx), while the Poisson number of arrivals during x contributes the factor exp{λx[β(s)−1]}.
Averaging over S gives β(s)=B*(s+λ−λβ(s)). This is the duration analogue of a branching fixed point. It retains the service-time law instead of replacing each service by its mean.
CHAPTER 4 OF 12 · BUILD THE TRANSFORM
4. Check the transform at zero and at infinity
At s=0, β(0)=P(T<∞) when e^(−s∞) is interpreted as zero for s>0. In a stable M/G/1 queue, β(0)=1. When ρ>1, the smallest fixed point can be below one, recording a positive probability that the branching workload never becomes extinct.
As s grows, β(s) emphasises very short busy periods. If service times are strictly positive, β(s) tends toward zero as s tends to infinity. These boundary checks catch the wrong numerical root and sign errors.
| Check | Stable expectation | Meaning |
|---|---|---|
| β(0) | 1 | Busy period finite almost surely |
| −β′(0) | E[T] | Mean duration |
| β(s) for s>0 | Between 0 and 1 | Valid transform |
| s→∞ | 0 for positive S | No zero-duration period |
CHAPTER 5 OF 12 · BUILD THE TRANSFORM
5. Solve the M/M/1 case explicitly
Let service be exponential with rate μ. Then B*(u)=μ/(μ+u), so β satisfies β=μ/(μ+s+λ−λβ). Rearrangement gives λβ²−(λ+μ+s)β+μ=0.
Choose the smaller root compatible with 0≤β≤1: β(s)=[λ+μ+s−√((λ+μ+s)²−4λμ)]/(2λ). With λ=2, μ=3 per hour and s=1 per hour, β(1)=[6−√12]/4≈0.634.
CHAPTER 6 OF 12 · TEST THE RESULT
6. Recover the mean without inverting the density
Differentiate the fixed point at s=0 in the stable case. The result is E[T]=E[S]/(1−ρ). For λ=2 and μ=3, E[S]=1/3 hour and ρ=2/3, so E[T]=1 hour.
This mean equals mean service multiplied by expected total progeny, but that equality is an expectation identity. It does not make T equal to N times a fixed service duration path by path.
| Quantity | Worked value | Role |
|---|---|---|
| λ | 2 per hour | Arrival rate |
| μ | 3 per hour | Service rate |
| ρ | 2/3 | Subcritical offspring mean |
| E[T] | 1 hour | Mean busy-period duration |
| β(1) | 0.634 | Transform at 1 per hour |
CHAPTER 7 OF 12 · TEST THE RESULT
7. Choose the correct fixed point near instability
The nonlinear equation can have more than one algebraic root. Probability theory selects the smallest admissible transform root. Iteration must be initialised and monitored accordingly; a solver that returns a larger root may violate β(s)≤1 or the extinction boundary.
Near ρ=1, the mean and tail grow sharply. Finite simulation runs then under-sample rare long periods, and numerical inversion becomes sensitive to transform accuracy. Report convergence checks rather than only a plotted curve.
CHAPTER 8 OF 12 · TRANSFER AND AUDIT
8. Transfer the duration logic to cascading repairs
Imagine one repair begins while failures arrive during active repair work. Each new failure adds its own repair demand, and the episode ends only when the backlog clears. The stable structure is an initiating job, Poisson offspring during service and independent repair durations.
The changed-context check is whether failure intensity depends on backlog, technician actions or common shocks. If arrivals accelerate during a major incident or one repair changes later durations, the iid branching fixed point no longer describes the episode.
CHAPTER 9 OF 12 · TRANSFER AND AUDIT
9. Estimate duration from complete episodes
Define the empty-to-empty boundary before extracting data. If observation begins in the middle of a busy period, that episode is left censored. If it ends before the queue clears, the duration is right censored. Dropping either type biases the empirical tail.
Record arrival, service-start and service-end times on one clock. Compare empirical means and survival probabilities with model predictions, and inspect whether arrival rates and service distributions stay stable inside long episodes.
CHAPTER 10 OF 12 · TRANSFER AND AUDIT
10. Invert transforms with independent checks
Numerical inversion methods convert β(s) into a density or survival function. They need accurate complex transform evaluations, a stated precision target and safeguards against cancellation. A stable fixed-point solver at real s is not automatically stable on the complex contour used by an inversion algorithm.
Check the recovered distribution against β(s) at several positive s, total probability, the analytic mean and direct simulation. Preserve the transform equation and solver tolerances so another reader can reproduce the curve.
CHAPTER 11 OF 12 · CHECK AND CONTINUE
11. Questions readers often ask
Is the busy-period duration exponential in an M/M/1 queue?
No. Exponential service does not make the whole branching extinction time exponential. Its transform has a square-root term and the duration distribution has a heavier structured tail.
Does ρ below one guarantee every observed busy period is short?
No. It guarantees finite mean and almost-sure extinction under the model, but long periods remain possible, especially near ρ=1 or with variable service times.
CHAPTER 12 OF 12 · CHECK AND CONTINUE
12. Evidence, connected routes and the next check
The 1971 branching analysis of M/G/1 busy periods derives transform results for infinite and finite waiting rooms. Abate, Choudhury and Whitt’s numerical inversion study examines practical recovery of the busy-period density.
Continue through Takács recursion and total progeny, Pollaczek–Khinchine mean waiting, and Lindley recursion and stability. The next check is to compare the empirical transform E[e^(−sT)] with the fixed point at several s values before attempting density inversion.
Contents · Previous chapter · Return to the Queueing Theory root