Queuing Theory: Models, Systems & Waiting Line Analysis

TAKEAWAYS:

• Queuing theory analyzes waiting lines to optimize service efficiency using Poisson arrivals and exponential service times.
• Kendall’s notation (A/S/c/K) classifies systems by arrival type, service discipline, and queue capacity.
• Key metrics (L = avg. queue length, W = avg. waiting time) are derived from Little’s Law: $L=\lambda W$.
• Single-channel vs. multi-channel systems differ in cost, complexity, and customer experience.
• Real-world tools: Banks use multi-server queues for ATMs; eSewa employs priority queues for high-value transactions.
• Exam focus: Solve numerical problems (e.g., calculate $L_q$, $W$) and interpret case studies (e.g., hospital ERs, call centers).

1. Introduction to Queuing Theory

Queuing theory studies waiting lines to balance service efficiency and customer satisfaction. It’s critical for industries where delays cost money or reputation—like Ncell’s customer service or Khalti’s payment processing.

Why It Matters

• Reduces costs: Idle resources (e.g., Daraz delivery agents) or frustrated customers (e.g., NTC call centers) hurt profitability.
• Improves experience: Airlines (e.g., Yeti Airlines) use priority queues for VIP passengers.
• Designs systems: Hospitals (e.g., CIMS Hospital) optimize ER triage queues to save lives.

Key Terms

2. Queuing System Components

Every queue has four core elements:

1. Input Source: Where customers arrive (e.g., Pathao’s ride requests).
2. Queue Discipline: How customers are served (e.g., first-come-first-served (FCFS)).
3. Service Mechanism: How service is provided (e.g., single teller vs. multiple ATMs).
4. Output: Departure of served customers (e.g., Nepal Rastra Bank’s loan approvals).

Visual: Queuing System Structure

flowchart TD
  A["Input Source\n(e.g., Customers)"] --> B["Arrival Process\n(λ: Poisson)"]
  B --> C["Queue\n(Waiting Line)"]
  C --> D["Service Facility\n(e.g., Bank Teller, μ)"]
  D --> E["Departure Process\n(Served Customers)"]
  C -->|"Feedback"| F["Rejected/Abandoned\n(e.g., Impatient customers)"]

3. Kendall’s Notation: Classifying Queues

Queues are labeled as A/S/c/K, where:

• A: Arrival distribution (e.g., M = Markovian/Poisson).
• S: Service time distribution (e.g., M = Exponential).
• c: Number of servers (e.g., 1 = single teller, 3 = 3 ATMs).
• K: Queue capacity (e.g., ∞ = unlimited, 10 = max 10 customers).

Example Systems

4. Assumptions of Single-Channel Queuing (M/M/1)

For simplicity, queuing models assume:

1. Poisson arrivals: Arrivals are random and independent (e.g., WhatsApp messages).
2. Exponential service times: Time to serve a customer is unpredictable (e.g., bank loan processing).
3. FIFO discipline: First-in, first-out (e.g., Daraz order fulfillment).
4. Infinite queue length: No limit on waiting customers (unless specified).
5. Stationary system: $\lambda <\mu$ (queue doesn’t grow infinitely).

5. Key Performance Metrics

Worked Example: Ncell Call Center

• λ (arrivals/hour): 48 calls
• μ (service rate): 60 calls/hour (per agent)
• ρ: $48/60=0.8$

Calculations:

1. L (avg. calls in system): $L=\frac{0.8}{1-0.8}=4$ calls.
2. W (avg. call duration): $W=\frac{1}{60-48}=0.05$ hours = 3 minutes.

Interpretation:

• Customers wait ~2.4 minutes in queue ($W_q=W-1/\mu$).
• Hiring a second agent reduces $\rho$ to 0.4, cutting wait times by 75%.

6. Multi-Channel Queues (M/M/c)

When multiple servers exist (e.g., Daraz’s 5 delivery agents), use:

• Erlang’s C formula for $L_q$ and $W_q$.
• Key trade-off: More servers reduce wait times but increase costs.

Example: eSewa Payment Counters

• λ: 30 transactions/hour
• μ: 15 transactions/hour/agent
• c: 2 agents

Calculations:

1. ρ: $30/(2\times 15)=1$ → Unstable system (queue grows infinitely). Fix: Add a third agent ($c=3$), so $\rho =0.5$.

7. Queue Discipline Strategies

8. Real-World Applications in Nepal

Case 1: NTC Customer Care

• Problem: Single queue for all complaints (λ = 120 calls/hour, μ = 80 calls/hour).
• Solution: Added priority queues for urgent repairs (e.g., power outages).
• Result: $W_q$ dropped from 1.5 hours to 20 minutes for high-priority calls.

Case 2: Kathmandu Traffic (Single-Lane Roads)

• Model: M/M/1 with abandonments (drivers leave if wait > 5 mins).
• Data: λ = 30 cars/hour, μ = 40 cars/hour, but 20% abandon.
• Fix: Added multi-lane signals (M/M/2), reducing $W_q$ by 60%.

Case 3: Daraz Order Fulfillment

• Process: Multi-channel queue (M/M/5) for 5 delivery agents.
• Challenge: Peak hours (λ = 100 orders/hour) vs. μ = 80 orders/hour/agent.
• Solution: Dynamic routing (assign orders to least busy agent).

9. Special Cases and Extensions

Finite Queue (M/M/1/K)

• Example: CIMS Hospital ER with max 10 patients ($K=10$).
• Impact: If $K$ is reached, new arrivals are blocked or rejected.
• Formula: Use Erlang’s B or C formulas for blocked/rejected systems.

Bulk Arrival Queues (M^k/M/1)

• Example: Group bookings at Himalayan Java (e.g., 5 customers arrive together).
• Notation: $M^k$ where $k$ = average group size.

Non-Poisson Arrivals

• Example: Rush-hour traffic (arrivals follow a deterministic pattern).
• Model: Use G/G/c notation (General arrival/service times).

10. Queuing Theory in Decision Making

Example: Bank ATM Placement

Scenario: A bank wants to place ATMs near a busy street.

• Option 1: 1 ATM ($\mu =20$ transactions/hour), $\lambda =40$.
  • $\rho =2$ → Unstable (queue grows infinitely).
• Option 2: 2 ATMs ($\mu =40$), $\lambda =40$.
  • $\rho =1$ → Stable but long waits ($W_q\to \infty$).
• Option 3: 3 ATMs ($\mu =60$), $\lambda =40$.
  • $\rho =0.67$, $W_q=0.33$ hours = 20 minutes.

Decision: Choose Option 3 for acceptable wait times.

11. Exam Tip: How to Score Full Marks

1. Understand Kendall’s Notation:

   • Always label systems correctly (e.g., M/M/3/∞ for 3 ATMs).
   • Common mistake: Forgetting to check $\rho <1$ (invalid if $\rho \ge 1$).

2. Memorize Key Formulas:

   • Little’s Law: $L=\lambda W$ (always true, even for complex queues).
   • M/M/1 metrics: Derive $L$, $L_q$, $W$, $W_q$ from $\rho$.

3. Case Study Approach:

   • Step 1: Identify $\lambda$, $\mu$, $c$, and queue discipline.
   • Step 2: Calculate $\rho$. If $\rho \ge 1$, state the system is unstable.
   • Step 3: Compute metrics and interpret results (e.g., "Customers wait 15 mins on average").

4. Compare Systems:

   • Use tables to contrast single vs. multi-channel (e.g., Ncell vs. Ncell VIP).

5. Real-World Links:

   • Relate to Nepali examples (e.g., traffic signals, bank queues).
   • Avoid generic answers: Examiners reward contextual applications.

12. Practice Problem (Exam-Style)

Case: A Khalti payment center has:

• 1 teller ($\mu =12$ transactions/hour).
• Average arrival rate $\lambda =8$ transactions/hour.
• Question:
  1. Calculate $L$, $L_q$, $W$, and $W_q$.
  2. If Khalti adds a second teller, how does $W_q$ change?
  3. Suggest one improvement to reduce wait times further.

Solution:

1. $\rho =8/12=0.667$.

   • $L=\frac{0.667}{1-0.667}=2$ customers.
   • $L_q=\frac{0.667^2}{0.333}=1.33$ customers.
   • $W=\frac{1}{12-8}=0.25$ hours = 15 mins.
   • $W_q=0.25-\frac{1}{12}=0.167$ hours = 10 mins.

2. With 2 tellers ($\mu =24$):

   • New $\rho =8/24=0.333$.
   • New $W_q=\frac{0.333}{24-8}=0.025$ hours = 1.5 mins.

3. Improvement:

   • Priority queues for high-value transactions (e.g., > Rs. 50,000).
   • Self-service kiosks to handle 30% of transactions (reduces $\lambda$).

13. Summary Table: Queuing Models

14. Common Pitfalls in Exams

1. Ignoring $\rho$:

   • If $\rho \ge 1$, the queue is unstable (infinite wait times).
   • Fix: Always check $\lambda <\mu$.

2. Mixing L and Lq:

   • $L$ = total customers in system (waiting + served).
   • $L_q$ = only waiting customers.

3. Assuming FIFO:

   • Some systems use priority (e.g., airport security lanes).
   • Always clarify the discipline in the question.

4. Unit Mismatch:

   • Ensure $\lambda$ and $\mu$ are in the same time units (e.g., both per hour).

15. Final Visual: Queuing Theory in Action

mindmap
  root((Queuing Theory in Nepal))
    Ncell["Ncell Customer Care"]
      Ncell --> Problem1["Single queue, high abandonment"]
      Ncell --> Solution1["Priority queues + multi-channel routing"]
    Daraz["Daraz Delivery"]
      Daraz --> Process1["M/M/5 queue for 5 agents"]
      Daraz --> Metric1["Dynamic assignment reduces Wq by 40%"]
    Traffic["Kathmandu Traffic"]
      Traffic --> Model1["M/M/1 with abandonments"]
      Traffic --> Fix1["Multi-lane signals (M/M/2)"]
    Banks["Nabil Bank ATMs"]
      Banks --> Strategy1["M/M/3 for peak hours (λ=60, μ=75)"]
      Banks --> Result1["Wq < 5 mins"]