Calculate why equal average workloads can produce different waits

Establish the model before calculating

Bristol’s notes specify a single FIFO server, Poisson arrivals, independent identically distributed service times and unlimited waiting space. Here service requirements are independent of arrivals; the server works whenever jobs wait. Require finite service moments and utilization below one. This is a stationary M/G/1 teaching model, not a description automatically established by a busy production board.

The fictional operation is an automated asset-processing station. No worker is being assessed. Choose a clearly defined submission and a completed-processing event before considering any actual evidence. If a real report combines processing with client approval, its duration would not represent the same task as the illustration.

Ask an analyst which assumptions the available records could support. A handful of observations or a familiar-looking histogram is not sufficient justification for adopting the model. Keep the owner’s operational account beside the statistical interpretation, and record unresolved contradictions rather than choosing whichever account produces the more convenient answer.

Calculate the service second moment

Let service duration S be measured in hours. Arrivals occur at rate λ = 3 per hour. Constant service of 0.25 gives E[S] = 0.25 and E[S²] = 0.0625 hours squared. Alternatively, equal probabilities of 0.1 and 0.4 give mean 0.5 × 0.1 + 0.5 × 0.4 = 0.25, but second moment 0.5 × 0.01 + 0.5 × 0.16 = 0.085. Square each duration before averaging; squaring the mean would erase the difference.

Treat those probabilities as supplied teaching inputs, not estimates from a production sample. Label the two scenarios clearly so a reader cannot confuse a hypothetical redesign with an observed operating pattern. Keep the invented input record separate from any future evidence collected under an authorized review.

Bristol explains that an arrival encountering ongoing work is more likely to encounter a long service than a short one. That remaining work helps explain why the distribution matters, not just its mean.

This explanation should prompt an operational question rather than an instruction to remove difficult work. Perhaps an owner needs to understand whether the observed task definitions are consistent. Do not infer that a longer legitimate job is wasteful or that rejecting it would improve the service that the organization actually intends to provide.

Compute the delay and enforce the boundary

Virtamo gives mean queue delay Wq = λE[S²] / (2(1 − ρ)), where ρ = λE[S]. Both scenarios have ρ = 0.75. The denominator is 0.5. Constant service gives 3 × 0.0625 / 0.5 = 0.375 hours, or 22.5 minutes; mixed service gives 3 × 0.085 / 0.5 = 0.51 hours, or 30.6 minutes. These exclude the job’s own processing.

At 4 arrivals per hour, utilization becomes 1. The stable-model requirement fails and the denominator is zero. Report the comparison as outside this calculation’s domain; do not return a finite wait or evaluate above the boundary as a negative delay.

Have another reader reproduce the two results from the written inputs. Ask them to identify what the result actually describes before discussing its magnitude. An accurate calculation paired with an inaccurate label could still mislead a campaign owner preparing a customer-facing statement.

Preserve both the accepted demonstration and the rejected boundary case. A worksheet that produces a plausible number for every input can be less trustworthy than one that explicitly stops. The stopping decision is part of the deliverable, not a formatting defect to be hidden before a presentation.

Use the contrast to ask a better operational question

Present the result as a controlled thought experiment. It shows why the original summary was insufficient for this particular inquiry. It does not show that the real service has either invented pattern, or that a proposed change would recreate the comparison.

Agree what additional evidence would justify further analysis. The process owner might need to clarify what constitutes a submission, which events are included in a duration or whether an apparent change follows a system release. Choose a focused question rather than requesting an unrestricted export of personal work histories.

Keep recommendations proportional to that evidence. A real campaign might require a different analytical model and commercial review before anyone changes intake or makes a service promise. Preserve the assumptions and the unanswered questions with the result so a future reviewer can distinguish a useful illustration from an operational commitment.

Make the reasoning concrete.

An asset-operations lead receives two proposals with matching headline averages and wants to know whether that settles their comparison.

The lead requests a bounded review without changing staffing, intake or customer promises.

Take the questions into your next brief.

Use these prompts in your own notes. Nothing is submitted from this page.

1. Object

   Which actual processing event would this illustration represent?

2. Evidence

   What would justify the model rather than merely make it convenient?

3. Calculation

   Can a second reader reproduce the fictional contrast?

4. Boundary

   Which input makes the worksheet stop?

5. Decision

   What additional evidence would support an operational recommendation?

Watch for the wrong conclusion.

• Calling a stipulated teaching probability an observed frequency.
• Removing legitimate complex work solely to improve a report.
• Publishing the illustration as a customer guarantee.

Sources & scope

Original operational reasoning and invented inputs. Mathematical applications and repeated method claims remain attributed. Teaching documents establish the stated model, not its fit to a real organization. Pure Python arithmetic was executed; no simulation, real service prediction or operational action occurred.

• 6.262 Lecture 13: Little, M/G/1, ensemble averages →Primary teaching material · Supports: Supports the variability distinction.
• The M/G/1 queue →Primary teaching material · Supports: Supports the mean-delay calculation.
• Simple queueing models →Primary teaching material · Supports: Supports the model assumptions and remaining-service interpretation.