9.

Sharing systems

lect09.ppt S-38.145 – Introduction to Teletraffic Theory – Spring 2005 1

9. Sharing systems

Contents

• Refresher: Simple teletraffic model

• M/M/1-PS (∞ customers, 1 server, ∞ customer places)
• M/M/n-PS (∞ customers, n servers, ∞ customer places)
• Application to flow level modelling of elastic data traffic
• M/M/1/k/k-PS (k customers, 1 server, k customer places)

2
9. Sharing systems

Simple teletraffic model

• Customers arrive at rate λ (customers per time unit)

– 1/λ = average inter-arrival time
• Customers are served by n parallel servers
• When busy, a server serves at rate µ (customers per time unit)
– 1/µ = average service time of a customer
• There are n + m customer places in the system
– at least n service places and at most m waiting places
• It is assumed that blocked customers (arriving in a full system) are lost

µ1
λ µ
n+m µ
µ
n 3
9. Sharing systems

Pure sharing system

• Finite number of servers (n < ∞), infinite number of service places

(n + m = ∞), no waiting places
– If there are at most n customers in the system (x ≤ n), each customer has
its own server. Otherwise (x > n), the total service rate (nµ) is shared fairly
among all customers.
– Thus, the rate at which a customer is served equals min{µ,nµ/x}
– No customers are lost, and no one needs to wait before the service.
– But the delay is the greater, the more there are customers in the system.
Thus, delay is an interesing measure from the customer’s point of view.

∞ µ1
λ µ
µ
µ
n 4
9. Sharing systems

Contents

• Refresher: Simple teletraffic model

• M/M/1-PS (∞ customers, 1 server, ∞ customer places)
• M/M/n-PS (∞ customers, n servers, ∞ customer places)
• Application to flow level modelling of elastic data traffic
• M/M/1/k/k-PS (k customers, 1 server, k customer places)

5
9. Sharing systems

M/M/1-PS queue

• Consider the following simple teletraffic model:

– Infinite number of independent customers (k = ∞)
– Interarrival times are IID and exponentially distributed with mean 1/λ
• so, customers arrive according to a Poisson process with intensity λ
– One server (n = 1)
– Service requirements are IID and exponentially distributed with mean 1/µ
– Infinite number of customer places (p = ∞)
– Queueing discipline: PS. All customers are served simultaneously in a fair
way with equal shares of the service capacity µ.
• Using Kendall’s notation, this is an M/M/1-PS queue
• Notation:
– ρ = λ/µ = traffic load

6
9. Sharing systems

State transition diagram

• Let X(t) denote the number of customers in the system at time t

– Assume that X(t) = i at some time t, and
consider what happens during a short time interval (t, t+h]:
• with prob. λh + o(h),
a new customer arrives (state transition i → i+1)
• if i > 0, then, with prob. i(µ/i)h + o(h) = µh + o(h),
a customer leaves the system (state transition i → i−1)
• Process X(t) is clearly a Markov process with state transition diagram
λ λ λ
0 1 2
µ µ µ

• Note that this is the same irreducible birth-death process with an infinite
state space S = {0,1,2,...} as for the M/M/1-FIFO queue.
7
9. Sharing systems

Equilibrium distribution (1)

• Local balance equations (LBE):

π i λ = π i +1µ (LBE)
⇒ π i +1 = λ π i = ρπ i
µ

⇒ π i = ρ iπ 0 , i = 0,1,2, K
• Normalizing condition (N):
∞ ∞
∑π i = π 0 ∑ ρ i = 1 (N)
i =0 i =0
−1
 ∞
i
⇒ π 0 =  ∑ ρ 
 i =0 

= ( )
1 −1
1− ρ
= 1 − ρ , if ρ < 1
8
9. Sharing systems

Equilibrium distribution (2)

• Thus, for a stable system (ρ < 1), the equilibrium distribution exists
and is a geometric distribution:
ρ < 1 ⇒ X ∼ Geom( ρ )
P{ X = i} = π i = (1 − ρ ) ρ i , i = 0,1,2, K
ρ 2 ρ
E[ X ] = 1− ρ , D [X ] =
(1− ρ ) 2
• Remark: Insensitivity with respect to service time distribution
– The result for the PS discipline is insensitive to the service time
distribution, that is: it is valid for any service time distribution with mean 1/µ
– So, instead of the M/M/1-PS model, we can consider, as well, the more
general M/G/1-PS model

9
9. Sharing systems

Mean delay

• Let D denote the total time (delay) in the system of a (typical) customer
• Since the mean number of customers in the system, E[X], is the same
for all work-conserving queueing disciplines, also the mean delay is the
same, by Little’s result.
• Thus, we may apply the result derived for the FIFO discipline in Lect. 8:

E[D ] = µ1 ⋅ 1−1ρ

10
9. Sharing systems

Mean delay E[D] vs. traffic load ρ

– Note that the time unit is the average service requirement E[S]
6

E[D] 3

0.2 0.4 0.6 0.8 1

traffic load ρ

11
9. Sharing systems

Relative throughput

• A quality of service measure is the relative throughput E[S]/E[D]:

E[ S ] 1
= ⋅ µ (1 − ρ ) = 1 − ρ
E[ D ] µ

12
9. Sharing systems

Relative throughput E[S]/E[D] vs. traffic load ρ

0.8

0.6
E[S]/E[D]
0.4

0.2

0.2 0.4 0.6 0.8 1

traffic load ρ

13
9. Sharing systems

Contents

• Refresher: Simple teletraffic model

• M/M/1-PS (∞ customers, 1 server, ∞ customer places)
• M/M/n-PS (∞ customers, n servers, ∞ customer places)
• Application to flow level modelling of elastic data traffic
• M/M/1/k/k-PS (k customers, 1 server, k customer places)

14
9. Sharing systems

M/M/n-PS queue

• Consider the following simple teletraffic model:

– Infinite number of independent customers (k = ∞)
– Interarrival times are IID and exponentially distributed with mean 1/λ
• so, customers arrive according to a Poisson process with intensity λ
– Finite number of servers (n < ∞)
– Service requirements are IID and exponentially distributed with mean 1/µ
– Infinite number of customer places (p = ∞)
– Queueing discipline: PS. If there are at most n customers in the system
(i ≤ n), each customer has its own server. Otherwise (i > n), the total
service rate (nµ) is shared fairly among all customers.
• Using Kendall’s notation, this is an M/M/n-PS queue
• Notation:
– ρ = λ/(nµ) = traffic load
15
9. Sharing systems

State transition diagram

• Let X(t) denote the number of customers in the system at time t

– Assume that X(t) = i at some time t, and
consider what happens during a short time interval (t, t+h]:
• with prob. λh + o(h),
a new customer arrives (state transition i → i+1)
• if i > 0, then, with prob. i⋅min{µ,nµ/i}⋅h + o(h) = min{i,n}⋅µh + o(h),
a customer leaves the system (state transition i → i−1)
• Process X(t) is clearly a Markov process with state transition diagram
λ λ λ λ λ
0 1 n n+1
µ 2µ nµ nµ nµ

• Note that this is the same irreducible birth-death process with an infinite
state space S = {0,1,2,...} as for the M/M/n-FIFO queue.
16
9. Sharing systems

Equilibrium distribution (1)

• Local balance equations (LBE) for i < n:

π i λ = π i +1 (i + 1) µ (LBE)
nρ
⇒ π i +1 = ( +λ1) µ π i = i +1π i
i
( nρ )i
⇒ π i = i! π 0 , i = 0,1,K , n
• Local balance equations (LBE) for i ≥ n:

π i λ = π i +1nµ (LBE)
⇒ π i +1 = nλµ π i = ρπ i
n nn ρ i
i−n i − n ( nρ )
⇒ πi = ρ πn = ρ n!
π0 = n!
π 0, i = n, n + 1, K 17
9. Sharing systems

Equilibrium distribution (2)

• Normalizing condition (N):

∞  n −1 ( nρ )i ∞ n n ρ i 
∑ π i = π 0  ∑ i! + ∑ n!  = 1 (N)
i =0  i =0 i =n 
−1
 n −1 (nρ ) i ( nρ ) n ∞ i − n 
⇒ π 0 =  ∑ i! + n! ∑ ρ 
 i =0 i =n 
−1
 n −1 ( nρ )i 
( nρ ) n 1
=  ∑ i! + n!(1− ρ )  = , if ρ < 1
 i =0  α +β
n −1 ( nρ ) i ( nρ ) n
Notation : α = ∑ i!
, β= n!(1− ρ )
i =0 18
9. Sharing systems

Equilibrium distribution (3)

• Thus, for a stable system (ρ < 1, that is: λ < nµ), the equilibrium
distribution exists and is as follows:

ρ <1 ⇒
 ( nρ )i 1
 i! ⋅ α + β , i = 0,1, K , n
P{ X = i} = π i =  n i
 n ρ ⋅ 1 , i = n, n + 1, K
 n! α + β
• Remark: Insensitivity with respect to service time distribution
– The result for the PS discipline is insensitive to the service time
distribution, that is: it is valid for any service time distribution with mean 1/µ
– So, instead of the M/M/n-PS model, we can consider, as well, the more
general M/G/n-PS model 19
9. Sharing systems

Mean delay

• Let D denote the total time (delay) in the system of a (typical) customer
• Since the mean number of customers in the system, E[X], is the same
for all work-conserving queueing disciplines, also the mean delay is the
same, by Little’s result.
• Thus, we may apply the result derived for the FIFO discipline in Lect. 8:

(
E[ D ] = µ1 ⋅ n(1W
−ρ )
p
+1 )
– where pw refers to the probability
∞ ∞ ( nρ ) n
nn ρ i β
pW = P{ X * ≥ n} = ∑ π i = ∑ π 0 ⋅ n! = π 0 ⋅ n!(1− ρ ) = α +β
i =n i =n

20
9. Sharing systems

Mean delay E[D] vs. traffic load ρ

– Note that the time unit is the average service requirement E[S]
6

4
n=1
E[D] 3
2
3
2
10
100
1

0.2 0.4 0.6 0.8 1

traffic load ρ

21
9. Sharing systems

Relative throughput

• A quality of service measure is the relative throughput E[S]/E[D]:

E[ S ] 1 n (1− ρ ) n (1− ρ )
=
E[ D ] µ
⋅ µ ⋅ =
pW ( n ) + n(1− ρ ) pW ( n ) + n (1− ρ )

E[ S ] 1− ρ
n = 1 : E[ D ] = p (1) +1− ρ = 1 − ρ
W
E[ S ] 2(1− ρ )
n = 2: =
E[ D ] pW ( 2) + 2(1− ρ )
= 1− ρ 2

22
9. Sharing systems

Relative throughput E[S]/E[D] vs. traffic load ρ

0.8
10 100
2 3
0.6
n=1
E[S]/E[D]
0.4

0.2

0.2 0.4 0.6 0.8 1

traffic load ρ

23
9. Sharing systems

Contents

• Refresher: Simple teletraffic model

• M/M/1-PS (∞ customers, 1 server, ∞ customer places)
• M/M/n-PS (∞ customers, n servers, ∞ customer places)
• Application to flow level modelling of elastic data traffic
• M/M/1/k/k-PS (k customers, 1 server, k customer places)

24
9. Sharing systems

Application to flow level modelling of elastic data traffic

• M/G/n-PS model is applicable to flow level modelling of elastic data

traffic
– customer = TCP flow
– λ = flow arrival rate (flows per time unit)
– r = access link speed for a flow (data units per time unit)
– C = nr = speed of the shared link (data units per time unit)
– E[L] = average flow size (data units)
– E[S] = 1/µ = E[L]/r = average flow transfer time with access link rate
– ρ = λ/(nµ) = traffic load
• A quality of service measure is the throughput
E[ L ] r ⋅E[ S ] r ⋅n(1− ρ ) (1− ρ )
θ = E[ D ] = E[ D ] = p ( n) + n(1− ρ ) = C ⋅ p ( n) + n(1− ρ )
W W

25
9. Sharing systems

Throughput θ vs. traffic load ρ

– Note that the rate unit is the link rate C

1

r=C
0.8

0.6
r = C/2
throughput θ
0.4 r = C/3

0.2
r = C/10
r = C/100
0.2 0.4 0.6 0.8 1
traffic load ρ

26
9. Sharing systems

Contents

• Refresher: Simple teletraffic model

• M/M/1-PS (∞ customers, 1 server, ∞ customer places)
• M/M/n-PS (∞ customers, n servers, ∞ customer places)
• Application to flow level modelling of elastic data traffic
• M/M/1/k/k-PS (k customers, 1 server, k customer places)

27
9. Sharing systems

M/M/1/k/k-PS queue

• Consider the following simple teletraffic model:

– Finite number of independent customers (k < ∞)
• on-off type customers (alternating between idleness and activity)
– Idle times are IID and exponentially distributed with mean 1/ν
– One server (n = 1)
– Service requirements are IID and exponentially distributed with mean 1/µ
– As many customer places as customers (p = k)
– Queueing discipline: PS.
• Using Kendall’s notation, this is an M/M/1/k/k-PS queue
• On-off type customer:
service
1
idleness
0

28
9. Sharing systems

State transition diagram

• Let X(t) denote the number of customers in the system at time t

– Assume that X(t) = i at some time t, and
consider what happens during a short time interval (t, t+h]:
• if i < k, then, with prob. (k−i)νh + o(h),
an idle customer becomes active (state transition i → i+1)
• if i > 0, then, with prob. i(µ/i)h + o(h) = µ + o(h),
an active customer becomes idle (state transition i → i−1)
• Process X(t) is clearly a Markov process with state transition diagram
kν (k−1)ν 2ν ν
0 1 k−1 k
µ µ µ µ

• Note that process X(t) is an irreducible birth-death process

with a finite state space S = {0,1,…,k}
29
9. Sharing systems

Equilibrium distribution (1)

• Local balance equations (LBE):

π i (k − i )ν = π i +1µ (LBE)
µ
⇒ π i = ( k −i )ν π i +1
1 µ k −i
⇒ π i = ( k −i )! (ν ) π k , i = 0,1,K, k

30
9. Sharing systems

Equilibrium distribution (2)

• Normalizing condition (N):

k k
µ
∑ π i = π k ∑ ( k −1 i )! (ν ) k −i = 1 (N)
i =0 i =0
−1
 k 1 µ k −i  1
⇒ π k =  ∑ ( k −i )! (ν )  = k
 i =0  1 ( µ )i
∑ i! ν
i =0
1 ( µ ) k −i
µ ( k −i )! ν
⇒ π i = π k ⋅ ( k −1 i )! (ν )i = k
1 ( µ )i
∑ i! ν
i =0 31
9. Sharing systems

THE END

32