349. Linking Equation
A simple but powerful approximation for how variability propagates through tandem queues. The departure-process variability from one station is the arrival variability for the next.
349.1. Formula
For a single station with utilization , arrival CoV , and service CoV , the departure CoV is:
— a weighted average of arrival and service variability, weighted by and .
349.2. Two extreme regimes
Low utilization (): . The station is mostly idle; departures inherit the arrival pattern.
High utilization (): . The station is always busy; departures inherit the service-time pattern.
In between: blend.
349.3. Tandem queues: variance flows downstream
Tandem layout:
Station 1 → Station 2 → Station 3 → ... → Station N
Arrival CoV at station 2 = departure CoV from station 1 = .
Applying the linking equation repeatedly:
So variance propagates downstream. Stations near the bottleneck (high ) propagate their service variability strongly; idle stations pass through arrival variability.
349.4. Why it matters: the Hopp-Spearman corollary
In a balanced line (all stations have similar and ), variability builds up but stays bounded. In an unbalanced line, one bad station injects variance that downstream stations can’t fully absorb — leading to queue blowups far from the original problem.
This is part of why line balancing (here) and variance reduction (eliminating setups, breakdowns) are foundational manufacturing improvements.
349.5. Bottleneck analysis
The linking equation lets you predict where queues will build before operations start:
- Compute at each station
- Apply the linking equation from upstream to downstream
- Combine with VUT / Sakasegawa to estimate queue lengths
Identifies bottleneck location and severity quantitatively.
349.6. Limitations
- Approximation — exact only for special distributions
- Independence assumed — auto-correlation in service or arrivals breaks the formula
- Open queues — doesn’t handle blocking / starvation directly (need finite-buffer extensions)
For finite buffers: see two-machine finite-buffer models (Buzacott-Shanthikumar) — more complex but capture blocking effects.
349.7. See also
- Kingman VUT
- Sakasegawa
- Best/Worst/PWC — Hopp-Spearman line performance
- Line Balancing