Queueing Systems:
Basic Definitions &
Littles Formula
Professor Izhak Rubin
Electrical Engineering Department
University of California, Los Angeles
UCLA
[email protected] 2014-2015 by Izhak Rubin
Basic Queueing System Model
Server 1
Server 2
Message
Arrivals Buffer; where
Server m
Message
Departures
Message queue
forms
Service channels
Key elements of the basic queueing system model:
Message arrival process (traffic model)
Message service time requirements
Waiting room facility (buffer)
Service facility
Joining mechanism and admission policy
Service discipline (scheduling policy)
Prof. Izhak Rubin
Message Arrival Processes
An denotes the arrival time of the nth message.
Arrival process:
A An , n 0 Renewal Point Process
Inter-arrival time:
Tn An An 1 , where Tn , n 1 i.i.d random variables
Inter-arrival time distribution:
A t P T t
E (T ) Mean inter arrival time; = 1/ = arrival rate [mess/sec]
Prof. Izhak Rubin
Service Time
S n denotes the service time requested by the nth message;
It expresses the time used by a service channel to serve / process (e.g.,
transmit or switch or perform other requested task) a message
We assume:
Sn , n 1 i.i.d random variables
Service time distribution:
B t P S t
E ( S ) Mean service time; =1/ = service rate [mess/sec]
Offered traffic rate (offered service load rate):
sec
msg
sec
fo
Erlangs
sec
sec
msg
1 Erlang = 1 sec service per sec = 1 minute service per minute.
Prof. Izhak Rubin
Service Discipline
The service discipline dictates the rule in accordance with which
arriving or queued messages are selected for service, once a service
channel becomes available.
Common service policies (scheduling schemes):
1. FIFO (FCFS): First-In First-Out (First-Come First-Served)
2. LIFO (LCFS): Last-In First-Out (Last-Come First-Served)
3. Priority class oriented (Preemptive vs. Non-preemptive)
4. Round-Robin
5. Shorter-Message-First (SMF)
Prof. Izhak Rubin
Basic Queuing Model: Notation
A/ B/ m/ N / K
Inter-arrival
distribution
time
Number of
sources
Service
distribution Number of
time
service
channels
Storage
capacity
For finite source
message arrival model
= total Number of
messages that can be
held (incl. those in
service)
M = Exponential
D = Deterministic
G = General
GI = General independent
Geom = Geometric distribution
Prof. Izhak Rubin
Queueing Processes and
Performance Metrics
Queue-Size Processes:
X t = System size at time t.
Qt = Queue size (wait size) at time t.
Z n X Rn Embedded queue size after the nth departure
Yn X An Embedded queue size prior to the nth arrival.
Waiting-Time Processes:
Vt = Virtual Waiting time at t.
Lt = Unfinished load at t.
Dn = VA = Delay time (system time) of the nth message; measured from
n
message instant of arrival to instant of departure.
Wn = VA = Waiting time of the nth message; measuring the total time spent
n
by a message in the queue, waiting to be served.
Prof. Izhak Rubin
Message Blocking
Assume the storage capacity of a QS is equal to N .
Call / Message blocking:
PB lim P Yn N
n
Time blocking:
PtB lim P X t N
t
o 1 PB
QS
o PB
System departing rate:
D o 1 PB [mess/sec]
System throughput load:
f D f AD f o 1 PB o 1 PB D
Prof. Izhak Rubin
[Erlangs]
8
Busy Period Properties
Xt
D2
Busy cycle
Yn X A , Y Yn , n 0
W2
S2
Busy period
S1
Z n X R , Z Z n , n 0
n
Gn Duration of nth busy period.
A1 A2 R1 R2
Idle period
I n Duration of nth idle period.
Cn Duration of nth busy cycle.
Vt
Cn I n Gn
N n Number of messages served
S2
in the nth busy period.
S1
W4
W2
W1
W5
W3
Prof. Izhak Rubin
Busy Period Properties
Non-blocking Queuing System:
N
C ~ Ti
i 1
Assume: finite means; renewal arrival process.
G
Wald's Lemma:
C
C NT
1
1
T
CN
N C
T1 T2
Prof. Izhak Rubin
T3
T4
10
Busy Period Properties
Blocking Queuing System:
N
C ~ Ti AD
i 1
1 k AD
lim E Tn T AD
k k
n 1
N AD
E Tn
E C
1
n 1
AD
EN
E N
AD C N D C
N 1 PB C
N=2
T1AD
T2AD
Prof. Izhak Rubin
T3AD
T4AD
11
Busy Period Properties
For a single-server QS with Poisson message arrivals at intensity ,
I t P I t 1 e t u t , and E I 1.
From Wald's Lemma,
E G E N E S E N AD E C
Since E C E G E I ,
E C E I AD E C E C
E I
1
E C
, 1
1 1
E N E C
E G E N
1
, 1
1
, 1
Prof. Izhak Rubin
12
Littles Formula
C
Xt
D1
D2
D3
X
dt
t
n
0
n 1
Little's Formula:
E
X
dt
For a non-blocking system,
t
0
DE N
1 T
X lim X t dt
D D
X
T T 0
E C
E C
X D
For a blocking system,
N
E Dn
k
1
X 1 PB D
D lim Dn n 1
k k
EN
n 1
Prof. Izhak Rubin
13
Littles Formula (Cont.)
Qt dt Wn
1
Q lim
T T
n 1
Qt dt
Q dt
C
E C
E
W
n
1 k
W lim Wn n 1
k k
EN
n 1
DE N
D W
Q
E
C
Define N s t as the number of msg. in service at t.
X t Qt N s t
X Q Ns ,
N s X Q D D W D S =D f D
Note: The departure load rate f D for single server 1.
Prof. Izhak Rubin
14
