DASE7034 Chap.11 Single and Multi-Server Queueing Models
Single and Multi-Server Queueing Models
M/M/1 assumes Poisson arrivals, exponential service, one server and a stable load; M/M/s extends the process to parallel servers and requires its own probability structure. Finite-capacity and general-service models change the state process again.
Capacity planning compares waiting performance with staffing or facility cost rather than minimising delay alone.
Use the single-server model only after checking load
For M/M/1, arrivals follow a Poisson process, service times are exponential, one server works at mean rate mu and the system has the assumed capacity and discipline.
Utilisation is rho equals lambda divided by mu, and steady-state formulas require rho below one. Total expected time is one divided by mu minus lambda; subtract mean service time to obtain queue waiting.
The corresponding number measures follow from Little's Law with consistent rate units.
Treat parallel service as a different probability model
In M/M/s, total nominal capacity is s times mu and stability requires lambda below that amount, but waiting is not obtained by replacing mu with s mu in the M/M/1 time formula.
The probability that all servers are busy must be calculated from the multi-server state probabilities. Utilisation is lambda divided by s mu, and queue delay then depends strongly on both utilisation and the number of servers.
State whether servers have identical rates and share one waiting line.
Match finite capacity and service variability
A maximum system size changes arrivals because a customer finding the system full is blocked. Offered demand, admitted throughput and observed queue length must be distinguished.
With non-exponential service, an M/G/1 or approximation may be more appropriate, and service-time variability becomes a driver of waiting. Priorities, abandonment, breakdowns and multiple customer classes also alter the model.
Formula choice should follow the process assumptions rather than the fact that a familiar expression is available.
Make capacity a decision with costs and targets
Additional servers reduce congestion but add staffing or facility cost. Compare alternatives using a common time and monetary basis, or impose a clear service-level condition such as a waiting-time target.
The option with the smallest delay is not automatically the best operating choice. Report utilisation, the chosen waiting measure, stability margin and cost consequence. Test demand scenarios because performance becomes highly nonlinear near saturation, and record whether the recommendation remains feasible when arrival or service rates move adversely.
What this chapter covers
- 01
M/M/1 steady state
- 02
M/M/s pooled service
- 03
Finite system capacity
- 04
Waiting-cost trade-offs
- 05
Model assumption checks
Worked example · free
Recompute a one-server benchmark
- 2Compute utilisation as ten divided by fifteen.
- 2Apply W equal to one over service minus arrival rate.
- 2Convert one fifth hour to twelve minutes.
Key terms
- M/M/1 Queue
- A one-server queue with Poisson arrivals and independent exponential service times.
- Pooled Queue
- A common waiting line feeding the next available server among a parallel group.
- Capacity Planning
- The choice of service resources by comparing waiting performance, cost and service requirements.
Single and Multi-Server Queueing Models FAQ
Can total service capacity replace mu in the one-server formula?
No. In a multi-server system, service capacity depends on how many customers are present until all servers are busy. Use the M/M/s probability structure rather than substituting s times mu into M/M/1 waiting formulas.
Why does waiting grow sharply near full utilisation?
Random variation creates temporary clusters of arrivals and long services. With little spare capacity, the system clears those clusters slowly, so expected waiting rises nonlinearly as utilisation approaches one.
Are separate lines equivalent to one pooled queue?
Not generally. Separate lines can leave one server idle while another line waits, whereas pooling sends the next customer to the next available server. The routing discipline changes waiting performance.
Exam move
For each queue, write assumptions, rates, server count, total capacity and stability before selecting a formula. Cross-check W, Wq, L and Lq using service time and Little's Law. When comparing staffing options, put waiting and staffing costs on the same time basis. Build an assumption card for M/M/1, M/M/s, finite-capacity and general-service models.
On each card, record arrival process, service distribution, server arrangement, capacity, discipline and stability condition. Given a scenario, underline the phrases that support or contradict each assumption before selecting any equation.
For a multi-server system, calculate total capacity and utilisation but resist substituting that capacity into the single-server waiting formula; instead identify the probability term required by M/M/s. Cross-check W, Wq, L and Lq using mean service time and Little's Law. Compare staffing alternatives in a table containing service margin, delay, blocking if relevant and cost on the same planning horizon.
Stress-test a higher-arrival and slower-service case because queueing performance can change sharply near saturation. End with the least-cost stable option or the smallest option meeting the stated service target, not simply the one with the shortest wait.
Working through Single and Multi-Server Queueing Models in DASE7034? Sia is AskSia’s AI Mathematics tutor — ask any DASE7034 Single and Multi-Server Queueing Models question and get a clear, step-by-step explanation grounded in how DASE7034 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.