Operations Research & Capacity Lab

Queuing Theory (M/M/s) Calculator

Model M/M/1 and M/M/s multi-server queuing systems, customer wait times, queue lengths, server utilization, and optimal staffing costs.

Queue Engine Staffing Matrix Queuing Equations FAQ

Service System Presets

Load calibrated queuing systems.

Step 1: Arrival Rate (λ), Service Speed (μ), Server Count (s) & Hourly Costs

System Demand & Capacity Inputs

⏱️ Arrival & Service Flow

Mean incoming customers.

Processing speed per server.

Total active service stations / staff.

💰 Economic Cost Tradeoffs

Loaded wages + station overhead.

Value of customer wait time / churn.

Server Utilization (ρ)
90.0%
2 servers handling 45 cust/hr (Total cap: 50/hr)
Average Wait Time in Line (Wq)
4.0 min
Total time in system (wait + service): 6.4 min
Average Queue Length (Lq)
3.03 cust
P(Delay): 85.3% | P(Idle): 5.3% | System: 4.8 cust
Total Hourly System Cost
$150.05 / hr
Labor: $44.00/hr + Customer Wait Cost: $106.05/hr

Capacity Staffing Matrix: Servers (s) vs. Arrival Demand (λ)

Simulates customer wait time (Wq minutes), total system cost ($/hr), and server utilization (ρ) across staffing levels.

Server Staffing -40% Demand -20% Demand Baseline Demand +20% Demand +40% Demand

Operations Research Principles

Understanding M/M/s Queuing Dynamics

Key operations management, capacity planning, and waiting line principles:

  • The Exponential Wait Curve: As utilization exceeds 80–85%, arrival surges create severe queue accumulation because servers lack buffer time to clear backlogs.
  • Erlang C Delay Formula: Calculates the exact probability that an arriving customer must wait ($P_w$) in a multi-server pool before being served.
  • Single Queue vs. Multiple Lines: Pooling multiple servers into a single serpentine line (M/M/s) drastically reduces average wait time compared to separate isolated lines ($s imes ext{M/M/1}$).
  • Economic Cost Tradeoff: Staffing costs increase linearly with more servers, while customer waiting costs drop exponentially, creating a distinct U-shaped total cost minimum.

Connect queue flow with WIP in the Little's Law Lab.

Mathematical Formulation

M/M/s Erlang C equations

ho = rac{lambda}{s cdot mu} quad ( ext{Stable when } ho < 1.0)

P_0 = left[ sum_{n=0}^{s-1} rac{(lambda/mu)^n}{n!} + rac{(lambda/mu)^s}{s!(1 - ho)} ight]^{-1}

P_w = rac{(lambda/mu)^s}{s!(1 - ho)} cdot P_0 quad ( ext{Erlang C Delay Probability})

L_q = rac{P_w cdot ho}{1 - ho} quad ext{and} quad W_q = rac{L_q}{lambda}

W = W_q + rac{1}{mu} quad ext{and} quad L = L_q + rac{lambda}{mu}

ext{Total Hourly Cost} = (s cdot C_{ ext{server}}) + (L_q cdot C_{ ext{wait}})

Model call center staffing in the Erlang C Lab.

FAQ

Queuing theory & waiting line questions

What is the M/M/s Queuing Model?

The M/M/s model represents a waiting line system with Poisson arrival rates (Markovian), exponential service times (Markovian), and 's' parallel, identical service channels.

What happens when server utilization (rho) approaches 100%?

As server utilization approaches 100%, queue length (Lq) and wait time (Wq) explode non-linearly towards infinity due to random arrival clustering and service time variance.

What is the difference between Wq and W?

Wq measures the average time a customer spends waiting in line before service begins. W measures the total time spent in the system (waiting time Wq plus service duration 1/mu).

How do you find the economically optimal number of servers?

The optimal server count minimizes total hourly cost, balancing the cost of adding staff against the customer goodwill cost of waiting in line.

Can I export queuing performance schedules to CSV?

Yes. You can export complete wait time metrics, Erlang probabilities, traffic intensities, and 6x5 sensitivity matrices as a UTF-8 CSV spreadsheet with formula defense.

Continue Exploring Operations & Capacity Tools

Explore our Operations Hub, analyze call center queues in the Erlang C Lab, model cycle time and WIP in the Little's Law Lab, optimize assembly stations in the Line Balancing Lab, or calculate line pace in the Takt Time Lab.