Queueing disciplines

 Nonpreemptive
 Firstt come first
Fi fi t served
d (FCFS)
 Head-of-the-line (HOL) priority
 Shortest Processing Time (SPT) first

 Round-robin
Round robin (RR) and Processor-sharing
Processor sharing (PS)

1/31/2017 Queueing disciplines (Simon S. Lam) 1

1
M/G/1 queue (FCFS discipline)
M/G/ d sc pl ne)
 Poisson arrivals at λ customers per second
 Service times with a general probability
distribution
mean value x and second moment x 2
 Define ρ = λx

 Another derivation
der vat on of P
P-K
K formula us
using
ng
mean residual life and Little’s Law

Queueing disciplines (Simon S. Lam) 2

2
M/G/1 queue (FCFS)

Now apply Little’s law

Queueing disciplines (Simon S. Lam) 3

3
M/G/1 queue (FCFS)

Queueing disciplines (Simon S. Lam) 4

4
M/G/1 Head-of-the-Line ((HOL)) nonpreemptive
p p

Queueing disciplines (Simon S. Lam) 5

5
M/G/1 (HOL) nonpreemptive

Queueing disciplines (Simon S. Lam) 6

6
 M/G/1 (HOL) nonpreemptive
apply
Little’s
law

Queueing disciplines (Simon S. Lam) 7

7
M/G/1 (HOL) nonpreemptive

Queueing disciplines (Simon S. Lam) 8

8
M/G/1 (HOL) nonpreemptive

Queueing disciplines (Simon S. Lam) 9

9
M/G/1 Shortest processing
ti
time fi
firstt (SPT) nonpreemptive
ti

Queueing disciplines (Simon S. Lam) 10

10
M/G/1 SPT nonpreemptive
(analogous to case 1 of discrete case)

Queueing disciplines (Simon S. Lam) 11

11
M/G/1 SPT nonpreemptive

Queueing disciplines (Simon S. Lam) 12

12
 Queueing disciplines

 Nonpreemptive
 Firstt come fi
Fi firstt served
d (FCFS)
 Head-of-the-line (HOL) priority
 Shortest Processing Time (SPT) first

 Round-robin
Round robin (RR) and Processor-sharing
Processor sharing (PS)

1/31/2017 Queueing disciplines (Simon S. Lam) 13

13
Round-robin (RR) scheduling

 The job first in queue gets a quantum q of service.

Then if it needs more service, it is returned to the end
of the queue.
 Good for CPU scheduling because job size is
unknown a priori
priori.

 In packet switching, a packet’s size is known

 But
B t si
size of
f application
li ti d data
t unit
it may nott b
be k
known
Queueing disciplines (Simon S. Lam) 14

14
Processor-Sharing (PS) discipline for
M/G/1
 What is the average delay and wait of a job
with service time x in the limit as q  0 ?
 T(x) = ave. delay of a job with service time x
x
=
1− ρ
 W(x)
W( ) = ave. wait
it of
f a jjob
b with
ith service
i titime x
ρx
=
1− ρ

[From
F Kleinrock,
l k Vol.
l 2, page 168]
1
Queueing disciplines (Simon S. Lam) 15

15
Packet scheduling in networks
 What if the app. user is willing to pay more money
for pr
priority
or ty ser
service?
ce?
Network neutrality advocates do not like this
 RR and PS scheduling – Are they more fair?
 How to implement
l PS scheduling?
h d l
 A packet can be thought of as a quantum in RR for an
application data unit.
 Delay of an application data unit is more important than
packet delay.
 We will return to these issues when we study
deterministic delay guarantees for a packet-
switching network.

Queueing disciplines (Simon S. Lam) 16

16
The end

Queueing disciplines (Simon S. Lam) 17

17