DELAY MANAGEMENT 75
vides the requirements into quality levels, and during the application runtime, it selects the desired
delay quality level, and passes it as the reference value ref
delay
to the priority adaptor. The priority
adaptor takes the delay reference value ref
delay
from the Requirement Adaptor, the measured QoS delay
value d
i
(t) from the delay monitor, and calculates the desired packet priority q which will be then
given to the Classifier. (Note that the priority calculation details will be presented in Section 4.2.3.)
The Classifier maps the calculated desired application packet priority q into the network packet
priority of the network service class to be used by the network WTP differentiation scheduler.
4.2.2 Proportional Delay Differentiation Scheduler
As discussed above, the proportional QoS service differentiation model was first introduced as a
PHB (Per-Hop-Behavior) for DiffServ in the context of wireline networks [7]. The proportional
differentiation relation is defined as a quality ratio of average delays
( , )
i
d t t τ+
and
( , )
j
d t t τ+
from
two different network service classes i and j, as shown in (Equation 4.1.) [7]. The relation indicates
that even though the actual quality (e.g., delay) of each service class might vary with traffic load, the
quality ratio between classes will remain constant in various-sized timescales.
In this section, we consider the proportional delay differentiation model in the domain of
wireless single hop networks such as IEEE 802.11. In IEEE 802.11 network, which consists of a set
of nodes N, the traffic is sent among different pairs of nodes and is divided into K network service
classes. The proportional differentiation relation in Equation 4.1 holds for traffic between any pairs
of nodes. It is important to stress that in wireline networks, the proportional differentiation model
only applies for traffic originated from the same router in wireline networks, it applies from differ-
ent source nodes.
d ( , )
, , {1, 2,..., }
d ( , )
j
i
i
j
t t
i j i j K
t t
δ
τ
δ
τ
+
= " ¹ Ù Î
+
(4.1)
where d
i
is the service differentiation parameter for class i, and
d ( , )
i
t t τ+
is the average delay for
class i, (i = 1,2,…, K) in the time interval (t, t + t) with t as the monitoring time scale.
One of the packet scheduling algorithms that can enforce the proportional delay differen-
tiation in short timescales is the waiting time priority (WTP) scheduler [23]. In this algorithm, a
network packet in the network queue is assigned a weight which increases proportionally to the
packets waiting time. Network service classes with higher differentiation parameter have larger
weight-increase factors. The packet with the largest weight value is served first in nonpreemptive
order. Formally, let w
p
(t) be the waiting time of a network packet p of network class i at time t, we
define its normalized waiting time
( , )
p
w t i
at time t as follows:
( , ) ( )
p p i
w t i w t δ= ×
(4.2)

Get Quality of Service in Wireless Networks Over Unlicensed Spectrum now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.