Unit II Advanced Queueing

Models
Non-Markovian Queues

1. Definition of Non-Markovian Queues

A Non-Markovian queue is a queueing system where either the arrival process or the
service process (or both) do not follow the Markov property, meaning:
1. Interarrival times or service times are not memoryless.
2. The system’s future state depends on more than just the current state, unlike
Markovian queues.
In Kendall’s notation A /S /c:
 Markovian queues: M/M/1 or M/M/c → Poisson arrivals (M), exponential service (M)
 Non-Markovian queues: G/G/1 or G/G/c → General distributions (G) for arrivals or
service times
Key idea: Non-Markovian queues handle more realistic situations where arrivals or service
times may be deterministic, normally distributed, Erlang, or any general distribution.

2. Characteristics of Non-Markovian Queues

1. General arrival or service patterns: Interarrival or service times may follow any
distribution, not necessarily exponential.
2. No memoryless property: Future behavior depends on the history, not just the current
number of customers.
3. More realistic modeling: Can model situations like scheduled appointments, bulk
service, or variable service times.
4. Analysis is more complex: Exact solutions often do not exist; simulations or
approximations are used.

3. Examples of Non-Markovian Queues

Example 1: Hospital Appointment System
 Patients have scheduled appointment times, so interarrival times are not exponential.
 Service times may vary depending on patient complexity (general distribution).
Example 2: Manufacturing System
 Machines process items with variable service times (e.g., normally distributed).
 Arrival of raw materials may be periodic or batch-based (non-Poisson).
Example 3: Supermarket Checkout
 Customers arrive randomly but service times vary with number of items.
 Service times may follow an Erlang or uniform distribution rather than exponential.

4. Types of Non-Markovian Queues

A. G/G/1 Queue
 G = General arrival distribution
 G = General service time distribution
 1 = Single server
Characteristics:
 Can model any realistic arrival and service pattern.
 Exact analytical formulas for performance measures (like waiting time) are usually not
available, often solved using approximation formulas or simulation.
Example: A single-machine production line with variable processing times and batch arrivals.

B. G/G/c Queue
 c servers, general arrival and service distributions.
Characteristics:
 More complex than single-server systems.
 Approximation methods (like Kingman’s formula) are used for average waiting time.
Kingman’s Approximation for average waiting time in G/G/1:
2 2
ρ ca + cs 1
W q≈ ⋅ ⋅
1−ρ 2 μ
Where:
 ρ=λ/ μ = traffic intensity
 c a=¿ coefficient of variation of interarrival times
 c s =¿ coefficient of variation of service times

✅ This formula helps estimate waiting time without exact solution.

5. Performance Measures
For Non-Markovian queues, we usually focus on:
1. Average number in system (L)
2. Average number in queue (Lq)
3. Average waiting time in system (W)
4. Average waiting time in queue (Wq)
Since exact solutions may not exist, simulation or approximations are commonly used.

6. Real-World Applications
1. Healthcare systems – appointments, surgeries with variable durations.
2. Manufacturing systems – processing jobs with different times, bulk arrivals.
3. Banking and customer service – variable teller service times, scheduled customers.
4. Transportation – buses/trains with scheduled arrivals and variable service times at
stations.

7. Comparison with Markovian Queues

Feature Markovian Queue Non-Markovian Queue
Arrival process Poisson General (any)
Service times Exponential General (any)
Memoryless Yes No
property
Analysis Closed-form formulas exist Often requires simulation or
approximation
Applications Simple banks, call centers, M/M/1 or Hospitals, manufacturing,
M/M/c systems scheduled services

8. Summary
 Non-Markovian queues generalize Markovian queues by allowing general
distributions for arrivals and service times.
 They are more realistic for real-world systems where memoryless assumptions do not
hold.
 Analysis is more complex; often requires approximations or simulations.
 Examples: Hospital appointments, manufacturing lines, supermarkets with varying
service times, and transport scheduling.
. The Pollaczek–Khinchine (P-K) Formula

1. Background: M/G/1 Queue

 M: Poisson arrivals (Markovian interarrival times, rate λ ).
 G: General service time distribution (not necessarily exponential).
 1: Single server.
Why is this important? 👉 In the real world, service times are rarely exponential. For example,
some customers take longer at a bank counter, or patients may take different consultation times
with a doctor. The M/G/1 model captures this reality.

2. The Pollaczek–Khinchine (P-K) Formula

The P-K formula gives the average waiting time in the queue (W q) and is one of the few exact
results available for non-Markovian queues.

Formula:
2
λ E[S ]
W q=
2(1−ρ)
where:
 λ = arrival rate (customers per unit time)
 S = service time random variable
 E [S] = mean service time
2
 E [S ] = second moment of service time
 ρ=λE[S ] = traffic intensity (must be ρ<1)

3. Interpretation
 W q increases if arrival rate λ is high.
 W q increases if service times are highly variable (since E [S2 ] captures variance).
 When service times are exponential, the formula reduces to the M/M/1 waiting time
result.
4. Related Results
Using Little’s Law ( Lq= λ W q):
 Average number in queue (Lq):
2 2
λ E [S ]
Lq =
2(1−ρ)
 Average waiting time in system (W):
W =W q + E [S ]

 Average number in system (L):

L=λW =Lq + λE [S ]

5. Example: Bank Teller System

 Customers arrive at a rate λ=5 per hour.

 Service time mean = 8 minutes = E [S]=8/60=0.133 hours.

 Service time variance = 0.01 hr❑2.

 Then,
2
E [S ]=¿
 Traffic intensity:
ρ=λE[S ]=5× 0.133=0.665
 Waiting time in queue:
5 × 0.0277 0.1385
W q= = ≈ 0.207 hours(12.4 minutes)
2(1−0.665) 0.67

✅ Interpretation: On average, a customer waits about 12.4 minutes in queue, in addition to the 8
minutes of service time.

6. Applications of P-K Formula

The P-K formula is widely applied in real-world systems where service times are not
exponential:
 5. Banking and customer service → Different customers take different amounts of time.
6. Healthcare systems → Patient consultation times vary greatly.
7. Manufacturing systems → Processing times depend on job complexity.
8. Computer/communication networks → Packets may have variable lengths, affecting
transmission/service times.
9. Call centers → Call handling times vary depending on issue type.

7. Summary
 Pollaczek–Khinchine formula applies to the M/G/1 queue.
 It gives exact results for average waiting time and queue length, even when service
times are not exponential.
 The formula shows the effect of variability in service times through the second moment
2
E [S ].
 Applications are everywhere: banks, hospitals, telecom, computing, and production
lines.

👉
Below is a fully detailed, step-by-step derivation of the Pollaczek–Khinchine (P–K) formula
for an M/G/1 queue. I’ll state assumptions, define every symbol, show every manipulation, and
include the key integral identities used.

Assumptions & notation

 Arrivals: Poisson process with rate λ .
 Service times: i.i.d. random variable S with pdf f S (s), cdf F S (s), survival
F S (s)=P( S >s ).
 Moments: E [S] exists and E [S2 ]< ∞.
 Traffic intensity (utilization): ρ=λE[S ], and we assume stability ρ<1.
 W q: steady–state waiting time in queue for an arriving customer (does not include that
customer's own service).
 N : number of customers waiting (excluding any in service) seen by an arriving customer.
 R : residual (remaining) service time of the job in service at an arbitrary time (or at an
arrival epoch, same by PASTA). If server idle, residual ¿ 0.
Goal: derive
2
λ E[S ]
W q= .
2(1−ρ)

Step 1 — Decompose the waiting time of an arriving

customer
An arriving customer sees either an idle server or a busy server. If server is busy, the arrival must
wait:
 the residual remaining time R of the job in service, plus
 the service times of the N customers already waiting.
Thus the waiting time (random) for the arrival is
N
W q=R+ ∑ Si ,
i=1

where the Si are i.i.d. copies of S (service times of those ahead).

Take expectations (linearity of expectation):
 [ ]
N
E [W q ]=E [R ]+ E ∑ Si .
i=1

Because service times are i.i.d. and independent of N (the number waiting),

[∑ ]
N
E Si =E[ N ] E[ S].
i=1

So (1) becomes
W q=E[ R]+ E [N ]E [S ].

Step 2 — Relate E [N ] to W by Little’s law q

Little’s law applied to the queue (customers waiting) gives

E [N ]=λ E[W q ].

(Justification: PASTA — Poisson arrivals see time averages — makes this applicable at arrival
epochs.)
Substitute (3) into (2):
W q=E[ R]+(λ W q) E[ S].

Rearrange:
W q−λE [S]W q=E [R].

Factor:
(1− ρ)W q =E[ R].

Hence
E[ R]
W q= .
1−ρ
So the remaining task is to compute the mean residual time E [R].

Step 3 — Compute the mean residual life E [R]

We compute E [R] by conditioning on whether the server is busy (probability ρ ) or idle
(probability 1− ρ).
 If server idle: residual ¿ 0.
  If server busy: we must compute the expected residual remaining time of the currently
ongoing service.
Thus
E [R]=P(busy)⋅ E [R ∣ busy ]= ρ⋅ E [R ∣ busy ].
So compute E [R ∣ busy ]. I give two equivalent, fully detailed ways — size-biasing (intuitive)
and an integral/tail identity (direct).

Method A — Size-biasing (intuitive / standard)

When you pick a random time and find the server busy, the probability that the service in
progress has total length in [s , s+ ds] is proportional to s f S (s) (longer services occupy more time
and are more likely to be the one you catch). This gives the size-biased density

(size) s f S (s)
fS (s)= , s ≥0 .
E[S ]
Given that the total service length is s, the remaining time if observation occurs uniformly within
the service is the expected residual
s
E [residual∣ S=s]= .
2
(Reason: if the start of the service is uniformly distributed in [0 , s ], the remaining time is s−U
with U ∼Uniform (0 , s), mean s/2.)
Average over the size-biased density:
∞ ∞ ∞
s (size) s s f S (s) 1
E [R ∣ busy ]=∫ ⋅ f S (s)ds=∫ ⋅ ∫ 2
ds= s f S (s) ds .
0 2 0 2 E [S ] 2 E [S] 0
∞

But ∫ s f S (s)ds=E[ S ]. So
2 2

2
E [S ]
E [R ∣ busy ]= .
2 E[S ]
Substitute (6) in (5):
2
E[S ]
E [R]=ρ ⋅ .
2 E [S]
Method B — Tail integral identity (alternative rigorous
route)
We can derive the same E [R ∣ busy ] from the residual-life density and a standard identity for
moments.
10. The equilibrium residual life (residual at a random time, conditioned on busy) has density
proportional to the remaining tail: its density g(x ) (for residual x ≥ 0 ) satisfies
∞

∫ f S (s )ds F S (x)
x
g(x )= = .
E [S] E[S ]
Hence
∞ ∞
1
E [R ∣ busy ]=∫ x g(x )dx= ∫ x F S (x ) dx .
0 E [S] 0

2. Use the standard identity for nonnegative random variables:

∞
E [S ]=2∫ x F S (x)dx .
2

(Proof sketch of the identity: integrate by parts ∫ x f S (x)dx=¿ . Under E [S2 ]< ∞ the boundary
2

0
term vanishes, giving the identity.)
Thus
2 2
1 E [S ] E[S ]
E [R ∣ busy ]= ⋅ = ,
E [S] 2 2 E[ S]
the same as (6). Then (7) follows.

Step 4 — Combine to obtain the P–K formula

Insert (7) into (4):
2
E[ R] ρ E [S ]
W q= = ⋅ .
1−ρ 1−ρ 2 E[S ]
ρ
Now use λ= (since ρ=λE[S ]), so
E [S ]
ρ
=λ .
E[S ]
Therefore
2
λ E[S ]
W q= .
2(1−ρ)
This is the Pollaczek–Khinchine formula:
2
λ E[S ]
W q= .
2(1−ρ)
Using Little’s law Lq= λ W q gives:
2 2
λ E [S ]
Lq = .
2(1−ρ)
And total time in system:
W =W q + E [S ], L= λW .

Step 5 — Check special case: exponential service

(M/M/1)
If S ∼Exp(μ), then E [S]=1/ μ and E [S2 ]=2/μ 2. Plug into P–K:
2
λ (2/ μ ) λ
W q= = 2 .
2(1−ρ) μ (1−ρ)

Since ρ=λ/ μ,
λ ρ
= ,
μ (1−ρ) μ (1−ρ)
2

which is the standard M/M/1 waiting-time formula. So P–K reduces correctly.

Step 6 — Important assumptions & remarks

11. PASTA (Poisson Arrivals See Time Averages) was used implicitly to equate arrival-
epoch probabilities to time-average probabilities (so probability the server is busy at
arrival = ρ , and residual life seen at arrival has same law as at a random time).
12. The derivation requires E [S2 ]< ∞ so the integrals and boundary terms behave.
13. The formula highlights variability: E [S2 ] (or equivalently Var (S)) enters — larger
variance → larger mean waiting time.
14. The formula is exact for M/G/1, not an approximation.
Optional: compact sequence of the algebraic steps (for
quick reference)
15. W q=E[ R]+ E [N ]E [S ].
16. E [N ]=λ W q ⇒ W q−λE [S]W q=E [R] .
17. (1− ρ)W q =E[ R] ⇒ W q=E[ R]/(1−ρ).
18. E [R]=ρ ⋅ E[ R ∣ busy ].
2
E [S ]
19. E [R ∣ busy ]= .
2 E[S ]
2 2
ρ E [S ] λE [S ]
20. Combine: W q= ⋅ = .
1−ρ 2 E[S ] 2(1−ρ)

If you want, I can now:

 (A) produce a fully worked numeric example (with numbers for λ , E [S], Var (S)) and
compute each intermediate quantity step by step, or
 (B) show the alternative derivation using transforms / embedded Markov chain (this is
longer but gives the full queue-length generating function), or
 (C) illustrate how waiting time grows as variance increases (plot / table).
Which of (A),(B),(C) would you like next?
Queues in Series
1. What are Queues in Series?
 A queue in series (or tandem queue) is a system where a
customer/job passes through two or more service stations
sequentially.

 Each station has:

o its own queue,

o its own server(s),
o possibly different service time distributions.
 After completing service at one station, the customer immediately
joins the queue for the next station.

👉 In short: The output of one queue becomes the input of the next queue.

2. Structure of Queues in Series

For two stations:
Q1 Q2
Arrivals → Server 1 →→ Server 2→ Departures .
General case:
Q1 → Q2 →⋯ →Q k

where a customer must go through k service points.

3. Key Properties
21. Arrival Process to the First Queue

o Usually modeled as a Poisson process.

22. Departure Process

o For M/M/1 queues: the departure process is also Poisson with the same rate
(Burke’s theorem).
o This makes analysis of tandem M/M/1 queues tractable.
23. Waiting Time
 o The total waiting time for a customer = sum of waiting + service times across all
queues in the series.

4. Example 1: ATM + Cash Counter

 Step 1: A customer inserts their ATM card and enters PIN (Server 1).
 Step 2: Then waits while machine counts and dispenses cash (Server 2).
Here, the queue in front of ATM (Q1) plus queue for cash dispenser (Q2) form a two-stage
tandem queue.
 Total time in system = waiting time at ATM + service at ATM + waiting at dispenser +
service at dispenser.

5. Example 2: Hospital System

 Patients first go to registration counter (Queue 1).
 Then they wait to see the doctor (Queue 2).
 After consultation, they may need to go to the pharmacy (Queue 3).
Each step is a queue in series. The total waiting time is the sum across all counters.

6. Example 3: Manufacturing Line

 A job passes through:

a. Cutting machine (Queue 1)

b. Polishing machine (Queue 2)
c. Quality inspection (Queue 3)
Jobs form tandem queues in the production line. Bottlenecks occur if one stage is slower.

7. Analysis (Simple Case: M/M/1 → M/M/1)

Suppose:
 Arrivals: Poisson with rate λ .
 Each station: M/M/1 queue with exponential service rates μ1 and μ2.
At Station 1:
 Utilization: ρ1= λ/ μ1.
ρ1
 Mean waiting time in queue: W q 1= .
μ1 (1−ρ1)

At Station 2:
 Departure from Station 1 is Poisson (Burke’s theorem) → input rate to Station 2 is also λ .
 Utilization: ρ2= λ/ μ2.
ρ2
 Waiting time in queue: W q 2= .
μ2 (1−ρ2 )

Total waiting time in system:

W q=W q 1 +W q 2 .

Total sojourn (system) time:

1 1
W =W q + + .
μ1 μ2

8. Why Important?
 Computer systems: a web request goes through firewall → application server →
database server.
 Airports: passengers queue at check-in → security → boarding.
 Telecom networks: packets go through routers sequentially.
 Manufacturing: multi-stage assembly lines.
In all cases, the slowest stage (the bottleneck) dominates performance.

9. Extension: General Queues in Series (M/G/1 →

M/G/1)
 If service times are general (not exponential), analysis becomes harder because departure
process is no longer Poisson.
 Approximation methods (like Pollaczek–Khinchine, Jackson networks, or simulation) are
used.

✅ Summary
  Queues in series (tandem queues): Customers/jobs pass through multiple service
stations one after another.
 Analysis is easy for M/M/1 tandem queues due to Burke’s theorem.
 Applications: manufacturing, hospitals, IT networks, transport, banking.
 Performance depends heavily on the slowest (bottleneck) station.

1. What is an Open Queueing Network?

An open queueing network is a system of multiple interconnected service stations (queues)
where:
 Jobs (customers, packets, products, patients, etc.) can enter from outside the system,
move between stations, and eventually leave the system.
 Each station may have one or more servers, its own queue, and its own service time
distribution.
 Routing between stations follows given probabilities.
👉 “Open” means jobs can enter and exit the network (unlike closed networks, where the
number of jobs is fixed).

2. Structure of an Open Queueing Network

General components:
 External arrivals: Jobs arrive from outside (often modeled as Poisson
arrivals with rate λ 0).

 Routing: After service at one station, a job can

o move to another station (with probability pij ), or

o leave the system (with probability 1−∑ p ij).
j

 Departures: Jobs eventually exit.

3. Jackson Network (Special Case)
The most famous open queueing network is the Jackson Network (when arrivals are Poisson
and service times are exponential).

Jackson’s Theorem
If:
24. External arrivals are Poisson,
25. Service times are exponential (M/M/1 or M/M/c),
26. Routing is Markovian (probabilistic, independent of past),
then:
 The steady-state distribution of jobs at each queue is the same as if each queue were
independent M/M/1 (or M/M/c) with an effective arrival rate λ i.
 This makes analysis tractable.

4. Effective Arrival Rates in Open Networks

Let there be k stations.
 External arrival rate to station i : γ i.
 Routing probability from station i to station j : pij.

Then the effective arrival rate at station i, λ i, satisfies:

k
λ i=γ i+ ∑ λ j p ji , i=1 ,2 , … , k .
j =1

This is a system of linear equations. Solve for all λ i.

Once we know λ i, each station can be analyzed like an independent M/M/1 queue.

5. Example 1: Two-Station Jackson Network

Suppose:
 Jobs arrive externally to Station 1 at rate γ 1=5 per hour.
 No external arrivals to Station 2 (γ 2=0 ).
 After service at Station 1, a job goes to Station 2 with probability p12=0.3 , or exits with
probability 0.7 .
 After service at Station 2, jobs always leave ( p21=0 ).
  Service rates: μ1=10 /hr, μ2=8/hr.

Step 1: Solve for effective arrival rates

 For Station 1: λ 1=γ 1+ λ2 p 21=5+ 0=5 .

 For Station 2: λ 2=γ 2+ λ1 p 12=0+5(0.3)=1.5 .

Step 2: Utilizations
 Station 1: ρ1= λ1 / μ1=5/10=0.5 .
 Station 2: ρ2= λ2 / μ2=1.5 /8=0.1875 .

Step 3: Average queue length & waiting time

2
ρ1 0.25
 For Station 1 (M/M/1): Lq 1= = =0.5 . W q 1=Lq 1 / λ1=0.5 /5=0.1 hr (6
1−ρ1 0.5
min).
2
ρ
 For Station 2: Lq 2= 2 ≈ 0.043 . W q 2=Lq 2 / λ2 ≈ 0.028 hr (1.7 min).
1−ρ 2

👉 Total expected waiting in network = 6 + 1.7 = 7.7 minutes.

6. Real-Life Examples of Open Queueing Networks

27. Computer systems

o Requests arrive from users → processed by web server → application server →

database → leave system.
28. Banking system

o Customers enter → registration counter → teller → sometimes to loan officer →

exit after completion.
29. Hospital workflow

o Patient enters → registration → lab test → doctor consultation → pharmacy →

leaves hospital.
30. Manufacturing/assembly line

o Raw materials arrive → cutting → welding → painting → packaging → product

exits system.
31. Telecommunication networks

o Packets enter router → forwarded to next router(s) → eventually reach

destination.
7. Why Important?
 They capture complex, realistic systems better than single queues.
 Jackson Networks provide a powerful framework to analyze them.
 Help in capacity planning, performance evaluation, and bottleneck identification.

✅ Summary
 An open queueing network allows jobs to enter from outside, move between service
stations, and exit.
 The arrival rate at each station is solved from flow balance equations.
 If exponential service and Poisson arrivals are assumed (Jackson Network), each station
behaves like an independent M/M/1 queue.
 Widely applicable in computer systems, banks, hospitals, factories, and telecoms.

Would you like me to also explain the difference between open, closed, and mixed queueing
networks (with examples), so you can see the bigger picture?

1. What is an Open Queueing Network?

An open queueing network is a system of multiple interconnected service stations (queues)
where:
 Jobs (customers, packets, products, patients, etc.) can enter from outside the system,
move between stations, and eventually leave the system.
 Each station may have one or more servers, its own queue, and its own service time
distribution.
 Routing between stations follows given probabilities.
👉 “Open” means jobs can enter and exit the network (unlike closed networks, where the
number of jobs is fixed).

2. Structure of an Open Queueing Network

General components:
 External arrivals: Jobs arrive from outside (often modeled as Poisson
arrivals with rate λ 0).

 Routing: After service at one station, a job can

 o move to another station (with probability pij ), or
o leave the system (with probability 1−∑ p ij).
j

 Departures: Jobs eventually exit.

3. Jackson Network (Special Case)

The most famous open queueing network is the Jackson Network (when arrivals are Poisson
and service times are exponential).

Jackson’s Theorem
If:
32. External arrivals are Poisson,
33. Service times are exponential (M/M/1 or M/M/c),
34. Routing is Markovian (probabilistic, independent of past),
then:
 The steady-state distribution of jobs at each queue is the same as if each queue were
independent M/M/1 (or M/M/c) with an effective arrival rate λ i.
 This makes analysis tractable.

4. Effective Arrival Rates in Open Networks

Let there be k stations.
 External arrival rate to station i : γ i.
 Routing probability from station i to station j : pij.

Then the effective arrival rate at station i, λ i, satisfies:

k
λ i=γ i+ ∑ λ j p ji , i=1 ,2 , … , k .
j =1

This is a system of linear equations. Solve for all λ i.

Once we know λ i, each station can be analyzed like an independent M/M/1 queue.
5. Example 1: Two-Station Jackson Network
Suppose:
 Jobs arrive externally to Station 1 at rate γ 1=5 per hour.
 No external arrivals to Station 2 (γ 2=0 ).
 After service at Station 1, a job goes to Station 2 with probability p12=0.3 , or exits with
probability 0.7 .
 After service at Station 2, jobs always leave ( p21=0 ).
 Service rates: μ1=10 /hr, μ2=8/hr.

Step 1: Solve for effective arrival rates

 For Station 1: λ 1=γ 1+ λ2 p 21=5+ 0=5 .

 For Station 2: λ 2=γ 2+ λ1 p 12=0+5(0.3)=1.5 .

Step 2: Utilizations
 Station 1: ρ1= λ1 / μ1=5/10=0.5 .
 Station 2: ρ2= λ2 / μ2=1.5 /8=0.1875 .

Step 3: Average queue length & waiting time

2
ρ1 0.25
 For Station 1 (M/M/1): Lq 1= = =0.5 . W q 1=Lq 1 / λ1=0.5 /5=0.1 hr (6
1−ρ1 0.5
min).
2
ρ
 For Station 2: Lq 2= 2 ≈ 0.043 . W q 2=Lq 2 / λ2 ≈ 0.028 hr (1.7 min).
1−ρ 2

👉 Total expected waiting in network = 6 + 1.7 = 7.7 minutes.

6. Real-Life Examples of Open Queueing Networks

35. Computer systems

o Requests arrive from users → processed by web server → application server →

database → leave system.
36. Banking system

o Customers enter → registration counter → teller → sometimes to loan officer →

exit after completion.
37. Hospital workflow
 o Patient enters → registration → lab test → doctor consultation → pharmacy →
leaves hospital.
38. Manufacturing/assembly line

o Raw materials arrive → cutting → welding → painting → packaging → product

exits system.
39. Telecommunication networks

o Packets enter router → forwarded to next router(s) → eventually reach

destination.

7. Why Important?
 They capture complex, realistic systems better than single queues.
 Jackson Networks provide a powerful framework to analyze them.
 Help in capacity planning, performance evaluation, and bottleneck identification.

✅ Summary
 An open queueing network allows jobs to enter from outside, move between service
stations, and exit.
 The arrival rate at each station is solved from flow balance equations.
 If exponential service and Poisson arrivals are assumed (Jackson Network), each station
behaves like an independent M/M/1 queue.
 Widely applicable in computer systems, banks, hospitals, factories, and telecoms.

Would you like me to also explain the difference between open, closed, and mixed queueing
networks (with examples), so you can see the bigger picture?