Adaptive Multi-Dimensional QoS-Based Packet Scheduling Scheme for Multimedia Broadcasting Over Geostationary Satellite Networks

Abstract

Future success towards 3G and beyond systems is in supporting a variety of multimedia services with diverse quality-of-service (QoS) demands. With their inherent broadcast capabilities, the broadband satellite networks are regarded as a promising platform for delivering multimedia services. For these systems, it is highly desired that the available resources can be utilized in an optimized way. Packet scheduling schemes play a key role in providing various QoS support for provisioning multimedia services. By taking into account essential aspects of QoS provisioning whilst preserving the system power/resource constraints, the proposed Adaptive Multi-dimensional QoS-based (AMQ) packet scheduling scheme aims to effectively satisfy diverse QoS requirements and adaptively optimize the resource utilization for satellite multimedia broadcasting. Simulation results show that the AMQ achieves much better performance than those of existing schemes by satisfying multiple QoS aspects, such as delay, throughput, channel utilization and fairness.

Full text

Page 1

Adaptive Multi-dimensional QoS-based Packet Scheduling Scheme for
Multimedia Broadcasting over Geostationary Satellite Networks
Hongfei Du, Linghang Fan, Barry G. Evans
Center for Communication Systems Research, Univeristy of Surrey,Guildford, Surrey, GU2 7XH, UK
Email: {H.Du, L.Fan, B.Evans}@surrey.ac.uk
Abstract— Future success towards 3G and beyond systems is in
supporting a variety of multimedia services with diverse
quality-of-service (QoS) demands. With their inherent broadcast
capabilities, the broadband satellite networks are regarded as a
promising platform for delivering multimedia services. For these
systems, it is highly desired that the available resources can be
utilized in an optimized way. Packet scheduling schemes play a
key role in providing various QoS support for provisioning
multimedia services. By taking into account essential aspects of
QoS provisioning whilst preserving the system power/resource
constraints, the proposed Adaptive Multi-dimensional QoS-based
(AMQ) packet scheduling scheme aims to effectively satisfy
diverse QoS requirements and adaptively optimize the resource
utilization for satellite multimedia broadcasting. Simulation
results show that the AMQ achieves much better performance
than those of existing schemes by satisfying multiple QoS aspects,
such as delay, throughput, channel utilization and fairness.
Index Terms — Packet scheduling, radio resource management,
S-DMB, MBMS, quality of service.
R
I.INTRODUCTION
ECENT
broadcasting have offered the mobile and broadcast
industries a beneficial platform to deliver multimedia
services to mass-market in a spectrum-efficient and
cost-effective way. The rapid growth in high-speed and
high-quality multimedia communications entails diverse
quality-of-service (QoS) requirements to be supported for
various multimedia applications including voice, data as well
as real-time video streaming. A variety of initiatives [1-4] has
been envisaged to provide one-to-many content distribution to
mobile users. As a complementary technology to 3G mobile
networks, the Satellite Digital Multimedia Broadcasting
(S-DMB) system is attracting a lot of attention within the
satellite community [4] as a cost-effective approach for
delivering Multimedia Broadcast/Multicast Services (MBMS)
services over satellite broadcasting networks. Based on its
broadcast nature, the S-DMB system offers extensive coverage,
low transmission cost for large numbers of terminals as well as
high QoS guarantees for real time multimedia applications. By
employing the wideband code-division multiple access
(WCDMA) with frequency division duplexing (FDD), the
system can be closely integrated with existing mobile cellular
networks, and minimise potential cost impacts on both 3G
cellular terminals and network operators.
advances in the mobile multimedia
Given the unidirectional nature of the S-DMB system and the
point-to-multipoint services it provides, aimed at maximizing
spectrum efficiency and satisfying diverse QoS requirements
whilst preserving the radio resources, the design of Radio
Resource Management (RRM) functionalities, especially the
packet scheduling scheme, proves to be an challenging task.
Although numerous studies on packet scheduling schemes have
been proposed in the literature for both wire- and wireless-
network [5-7], they cannot be easily applied to S-DMB because
of its unique nature. One popular research subject foreseenin
this context is to exploit the channel quality of fast-varying
wireless link for more efficientpacketscheduling [6].However,
giventhe unidirectional nature and long propagation delay, the
S-DMB system is unable to track real time channel state
information from the mobile terminal side, which makes the
channel-state dependent scheduling not feasible. Even if such
information were available, it still has to be exploited in an
unconventional manner considering the point-to-multipoint
natureofthesupported services, i.e. increased heterogeneity of
users interested in the same content. Furthermore, future
multimedia applications feature increasingly diverse range of
capabilities and QoS requirements, hence the packet scheduling
has to take into account both the differentiation and fulfilment
of these requirements. Finally, given the limited available
power for satellite transmission, the packet scheduling has to
be designed so as to optimize the overall transmit power.
Previous work on the packet scheduling in S-DMB has been
systematically formulated and addressed via adaptation of two
well-known scheduling algorithms [7], namely weighted fair
queuing (WFQ) and multi-level priority queuing (MLPQ), both
of which prove difficult and inefficient in provisioning
QoS-differentiated multimedia services in satellite networks. In
order to achieve better packet scheduling performance in terms
of both efficiency and fairness, inherited from the proportional
delay differentiation (PDD) in the context of differentiated
service networks, a delay differentiation queuing (DDQ) was
proposed in our early work [8], offering improved performance
on delay, jitter, and channel utilization. However, DDQ
experiences unbalanced performance among multiple QoS
attributes, namely the gain achieved in one attribute leads to the
performance degradation on other attributes. Furthermore,
multimedia services feature differentiated delay constraints,
applying the delay constraints for differentiated services in an
equal way may lead to inferior QoS guarantee for high priority
queue, therefore the delay profile has to be considered against
the respective delay constraints (i.e. maximum acceptable delay)
specified by the service. Finally, rather than scheduling
competing flows in a static manner, to provide more adaptive
and flexible QoS provisioning, it is highly desired that the
scheduler is capable of adaptively selecting the best scheduling
policy according to the diverse QoS preferences of the services
and the instantaneous performance dynamics.
1930-529X/07/$25.00 © 2007 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.
Authorized licensed use limited to: University of Surrey. Downloaded on February 5, 2010 at 10:24 from IEEE Xplore. Restrictions apply.
4613

Page 2

Figure 1. S-DMB system overview.
For these reasons, we propose a novel packet scheduling
scheme, namely adaptive multi-dimensional QoS-based (AMQ)
algorithm, at the Medium Access Control (MAC) layer that
considers multiple performance criteria across layers in order
to adopt the most appropriate packet scheduling policy in
response to diverse QoS demands and traffic dynamics. The
novelties of the proposed AMQ scheme are that: 1) it satisfies
multiple essential QoS requirements at both application layer
and transport layer, 2) adaptively tracks the queuing dynamics
induced by heterogeneous traffics at the Radio Link Control
(RLC) sub-layer of the data link layer, and 3) is capable of
dynamically adapting itself to the most appropriate scheduling
policy according to service QoS preferences and instantaneous
performance variations. The proposed AMQ scheme is
mathematically formulated and evaluated in a unidirectional
geostationary satellite broadcast system (i.e. S-DMB) through
extensive analysis/simulation
proposed methodology can also be applied adaptively to any
WCDMA-based broadcast/multicast network.
studies; nevertheless, the
The paper starts with a brief review of S-DMB and packet
scheduling. The subsequent section details the proposed AMQ
algorithm and addresses its scalability and complexity.
Performance of AMQ against a variety of existing packet
scheduling schemes is then evaluated in Section IV, where the
simulation methodology is presented and performance results
are discussed. Finally, we conclude this paper in Section V.
II.S-DMB SYSTEM AND PACKET SCHEDULING
As shown in Fig. 1, the S-DMB system defines a hybrid
satellite-terrestrial communication
unidirectional geostationary satellite component that is
responsible for the delivery of the point-to-multipoint MBMS
services and provides a European coverage by multiple umbrella
cells. Being closely integrated into the 2.5G/3G baseline
architecture, the system enjoys maximum reusing of technology
and infrastructure and minimum system development cost. The
user equipment (UE) applies the standard 3G terminal enriched
with S-DMB-enablingfunctions,which, given the unidirectional
nature, are very limited. The terrestrial gap-fillers, identified as
intermediate module repeater (IMR), are co-installed physically
at the terrestrial base stations to enhance the signal reception
quality and provide adequate coverage in urban, built-up areas.
system, featuring a
Figure 2. Proposed AMQ packet scheduling framework.
It is noteworthy that no direct satellite return link is
envisaged under the baseline S-DMB infrastructure, the return
path is rather provided via the terrestrial link if needed. It is
assumed that MBMS services are intended for transmission to
UEs in either a broadcast or multicast way. In the latter case
service is only delivered to the UEs within a specific multicast
group. Packets from the BM-SC are firstly buffered at the
satellite hub (SAT-Hub) - or Node B - in a FIFO manner before
being scheduled for transmission over satellite link.
In S-DMB, the nonavailability of a return link penalizes the
system effectiveness and efficiency on short-term resource
allocation. Therefore, no fast power control mechanism is
applicable in such a system, whilst the packet scheduling
algorithm, which is the single function performing fast
resource allocation, is the focus of efficient resource allocation.
As shown in Fig. 2, the packet scheduling strategy can be
conceptualised into the following two main steps:
•
Service Prioritization: The incoming service requests are
re-ordered according to the priority criteria. In selecting the
respective criteria, the multiple performance attributes are
considered to provide dynamic scheduling task.
•
Resource Allocation: Once all the sessions are prioritized,
bit rate and transmit power are assigned to each session in
the each transmission time interval (TTI).
III.AMQ PACKET SCHEDULING
A.Overview
Advances in multimedia applications entail the packet
scheduling algorithm to support diverse QoS among
heterogeneous traffics. The proposed AMQ algorithm takes into
account several key performance criteria simultaneously for
assuring comprehensive QoS satisfaction. On one hand, rather
than differentiating the competing sessions with respect to their
inherent traffic priorities (i.e. service types), the AMQ scheme
considers the application prescribed QoS requirements as a
combination of multiple QoS attributes. On the other hand, the
queuing dynamics of the competing flows at the RLC layer are
monitored and considered in response to the fast-varying traffic
dynamics. The proposed AMQ mechanism operates at the MAC
sub-layer of the data link layer within the S-DMB RRM
functionality entity.
1930-529X/07/$25.00 © 2007 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.
Authorized licensed use limited to: University of Surrey. Downloaded on February 5, 2010 at 10:24 from IEEE Xplore. Restrictions apply.
4614

Page 3

As shown in Fig. 2, the admitted ongoing sessions comprise
multiple MBMS sessions with diverse QoS demands. In
S-DMB, each session is assumed to retain an individual
Forward Access CHannel (FACH) queue in the RLC buffer.
Packets in the FACH queues are prioritized in decreasing order,
based on the parameters abstracted from both radio resource
allocation (RRA) at the beginning of each session starts and the
RLC queuing buffer at per-TTI scale. The involved parameters
are then become subject to two formulated mechanisms:
service classification and queue differentiation. The former is
performed as the QoS classification of competing service flows
depending on their QoS requirements, which performs once
during the phase of service establish or re-negotiation. Whilst
the latter keeps tracking the queuing dynamics for competing
flows during the session transmissions, on a TTI-by-TTI basis.
To consider both QoS criteria and queuing behaviours, we
introduce an adaptive priority function (APF) for handling the
contributing parameters from aforementionedtwomodules. The
involved parameters can be effectively sub-categorized into two
main streams: static priority attribute (SPA) and dynamic
priority attribute (DPA). SPA refers to the QoS guarantees
expressed in terms of service prescribed QoS rank, required
data rate, queuing delay/buffer occupancy bound and targeted
packet loss rate (PLR)/throughput, which keep constant during
the session transmission. Whilst the DPA represents the
instantaneous queuing behaviours at current TTI in terms of
queuing delay, queue length, packet loss rate and throughput,
these performance criteria keep tracking the queuing status
dynamically and update themselves in per-TTI scale.
Upon receiving the SPAs/DPAs in either per-session or
per-TTI scale, APF carries out the ranking and priority
derivation process and comes up with a quantified priority
associated with each FACH queue for current TTI. The queue
with the highest priority is to be served ahead of the other
competitors. The objective of the AMQ problem is to provide
the highest possible level of diverse QoS satisfaction among
heterogeneous multimedia subject to the system resource and
power constraints. The prioritized queues are then passed to
“Resource Allocation” for the allocation of required resources.
B. Algorithm Description
Taking into account the parameters abstracted from the
SPA/DPA list, we define APF function ϑj(n) for FACH
transport channel j at the current TTI n as:
ϑj(n)=αj·Τ j(n) ·Λ j(n) ·Γ j(n) ·Ξ j(n) ·Η j(n),
j = 1, ...,J; n = 1,...,N.
(1)
where αjis the prescribed QoS rank for the jth session, J is the
total number of FACH queues, N is the total number of TTIs.
For each TTI n, the instantaneous queuing behaviors in queue j
can be characterized by a multi-dimensional vector (Τ j(n), Λ
j(n), Γ j(n), Ξ j(n), Η j(n)), denoting the performance coefficients
of queuing delay, buffer occupancy, data rate, packet loss rate
and throughput, which reflect the current distance between
achieved performance and its desire threshold.
The first involved profile αj, namely the QoS profile, is
essentially a time-independent parameter designated for each
queue, reflecting the relative priority level of the service
carried by the jth FACH queue. The higher αj is, the higher
priority of the session is. It is noteworthy that QoS profile is
the premier criterion in the APF, which means that in majority
of the time, the high QoS sessions will be served ahead of their
low QoS counterparts. However, this is not necessarily the
truth when one or more performance criteria are degraded to
such an extremely severe condition that the scheduler must
take immediate action to prevent the session getting undesired
loss (e.g. buffer overflow, exceptional long delay).
Due to the unidirectional nature of the envisaged S-DMB, the
end-to-end delay in the network is not obtainable at the
SAT-Hub. Queuing delay experienced in the RLC buffer is
thereby employed in defining the delay-related metric in this
paper. We define the mean queuing delay for the jth FACH
queue until the nth TTI as:
??
Θ∈∆∈
τ
)()(
)()(
)(
,,
nNnN
nn
n
q
j
l
j
k
q
jk
k
q
jk
j
+
+
=
ττ
,
N. 1,...,n
J...., 1,j
=
=
(2)
where
queue before TTI n,
queuing in the jth FACH buffer at TTI n, ?:= {1,2, …,
?:= {
)(nNl
j
+1,…,
)(nNl
j
queuing delay for kth packet arrived in the queue j, defined as:
)(nNl
j
the number of packets that have left the jth
)(nNq
j
is the number of packets that are
)(nNl
j
},
+
)(nNq
j
},
)(
, n
k
q
j τ
is the current
??
??
?
>−⋅
≤−
=
)(Nk)(
)(Nk)()(
)(
l
j
l
j
.
n ifkTTn
nif kTkT
n
avl
jtti
avl
j
lev
j
q
kj τ
,
N.1,...,n
J....,1, j
=
=
(3)
where n·Tttirepresents current timing (Ttti is the value of TTI, i.e.
80ms in our simulation),
Tavl
j
arrival time and leaving time of the kth packet in the jth queue.
)(k
and
)(kTlev
j
denote the
In S-DMB, a queuing delay threshold is assigned to each
admitted session, representing the maximum acceptable
queuing delay for the corresponding service. Let τj
maximum acceptable queuing delay for the jth FACH queue
specified by session’s QoS requirements. We associate with
each FACH queue j a queuing delay profile Τj(n) given by:
?
≤
=
j
)(nif
ττ
τ
*denote the
?
?
?
?
>
*
j
j
*
j
*
j
j
)(
)(1
)(
j
n
nif
nT
τ
ττ
,
N.1,...,n
J....,1,j
=
=
(4)
This attribute depends on the maximum queuing delay
tolerated by the corresponding service, which proportionally
adjusts itself in response to the difference between the mean
queuing delay(
)(
τ
) and its delay threshold. It is only effective
when the mean queuing delay is beyond the designated delay
threshold. It is noted that the delay threshold can be regarded as
a tuneable parameter upon balancing the system performance.
jn
Once the finite length buffer at the SAT-Hub is employed, it
is vital, especially for loss-sensitive service, to maintain the
queue length at a safe level to prevent the system from the
excessive packet loss due to buffer overflow. Let λj
*denote the
1930-529X/07/$25.00 © 2007 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.
Authorized licensed use limited to: University of Surrey. Downloaded on February 5, 2010 at 10:24 from IEEE Xplore. Restrictions apply.
4615

Page 4

maximum buffer length for the jth FACH queue. The buffer
occupancy profile ?j(n) for the jth FACH queue is given by:
?
⋅≤
=Λ
j
jj
σλ
where σj is the buffer occupancy threshold, providing a safe
bound for the buffer length,
queue length of the jth FACH at current TTI.
?
?
?
?
⋅>
⋅
j
jj
j
n if
n
nif
n
σλλ
λ
*
σλλ
*
j
j
*
j
)(
)(
)(1
)(
,
N.1,...,n
J...., 1,j
=
=
(5)
)(n
j λ
denotes the instantaneous
The date rate profile is calculated as the ratio of the service
required/guaranteed data rate against the mean data rate at
current time. The instantaneous priority of each queue is
affected proportionally by the difference between the mean
transmitted data rate and the required data rate of each queue.
Let γj
the data rate profile Γj(n) of the jth FACH queue is defined as:
1 if ( )n
γγ
?
≤
??
Γ=?
?
??
where
γ
denotes the mean data rate of jth FACH achieved
until TTI n, which is determined as:
nN
TnSn
⋅
=
1
where Sj,k represents packet size for k th packet in queue j.
*denote the guaranteed data rate for the jth FACH queue,
*
jj
j
*
jj
*
j
( )n
( ) if
γ
( ) >n
j
n
γ
γγ
,
N. 1,...,n
J....,1,j
=
=
(6)
j( ) n
tti
k
kjj
l
j
=?
)(
,
)(
γ
,
N. 1,...,n
J...., 1,j
=
=
(7)
Similar to the queuing delay profile, the packet loss available
at the SAT-Hub is also confined to the packet loss due to buffer
overflow, although the packet loss in the propagation path is the
most crucial factors impacting the QoS performance.
Nevertheless, the packet loss due to buffer overflow is the single
metric that can be monitored and controlled by the RRM entity.
Let ?j
overflow for the jth FACH queue, ?j is the packet loss rate
threshold for the jth FACH queue. The packet loss rate profile
Ξj(n) is defined as:
*
jj
1 if ( )
( )
( ) if
( ) >n
ξξ
ξδ
⋅
? ?
j( ) n
ξ
denotes the mean PLR of jth FACH achieved
until TTI n, which is defined as:
*denote the acceptable packet loss rate due to buffer
j
*
jj
*
j

j
j
j
j
n
n
n
ξξδ
⋅
ξ
δ
⋅
?
??
≤
Ξ=?
?
,
N.1,...,n
J....,1,j
=
=
(8)
where
)()(
)(
)(
nNnN
d represents the total number of packets that are
dropped due to buffer overflow for the jth FACH until TTI n.
nN
n
l
j
q
j
d
j
j
+
=
ξ
,
N.1,...,n
J...., 1,j
=
=
(9)
where Nj
In this paper, we consider the throughput as the buffer
throughput at SAT-Hub, which is obtained by dividing the total
bits successfully scheduled and delivered to the physical
channel for radio frame transmission with the total bits arrived
in a specific FACH queue until current time. Let ηj
target throughput for the jth FACH queue, the throughput profile
Ηj(n) for the jth FACH queue is given by:
*denote the
*
jj
j
*
jj
*
j
1 if ( ) n
( )n
( ) n
ϕ
⋅
if ( ) >n
j
j
j
j
ηηϕ
η
η
ηηϕ
?
??
?
? ?
≤⋅
Η=?
⋅
,
N.1,...,n
J....,1,j
=
=
(10)
j( ) n
achieved so far, which is defined as:
a
jjj
BBn =
)(
η
where Bj
successfully scheduled for transmission for the jth FACH until
current TTI, Bj
arrived in the jth FACH so far.
η
denotes the mean throughput of jth FACH that has been
s
, j = 1, ...,J; n = 1,...,N.
s represents the total number of bits that are
(11)
a represents the total number of bits that are
C. Flexibility and scalability
In the above context, we assume all the contributing profiles
influence the APF in an equal way during the session
transmission. However, fixed setting upon all performance
criteria may not work well in provisioning multimedia data
with diverse QoS demands and fast-varying traffic dynamics,
the performance gain achieved in one profile may sacrifice the
performance on other profiles, which may be even more
important for the specific service. The proposed AMQ
algorithm provides a tuning ability over essential performance
profiles to further optimize the scheduling performance. By
observing the QoS preferences specified by service and the
behaviours of queuing status, the AMQ scheduling entity
dynamically adjusts the following “tuning knobs” on a
TTI-scale: 1) queuing delay threshold (σj), 2) PLR threshold
(?j), and 3) throughput threshold (?j). By selecting an
appropriate combination of the above thresholds for each queue,
the serving orders of competing flows can be effectively
managed. According to the sensitivity preferences of service
QoS classes, through giving flexible weights to different
profiles in terms of delay, PLR and throughput, it is therefore
possible to adaptively select the scheduling policy to allow for
different treatment of diverse QoS demands and to maintain
optimal resource utilization. For example, the σj is preferred to
be set higher for delay-tolerant PLR-sensitive service, whilst
preserving a target ?j, ?j. Some applications have stringent
constraints on the achieved throughput rather than PLR, thus
the scheduler should apply lower ?j for better throughput
performance whilst releasing the constraints set by σj, ?j.
From the viewpoint of implementation, the proposed AMQ
algorithm introduces extra computation complexity due to its
nonlinear (with loop iterations for selection sort operation) and
nondeterministic (with unpredictable variable) nature. In order
to examine the scalability of the proposed AMQ algorithm, the
Big O notation [9] is employed for determining the involved
computational complexity. We assume that there are n sessions
to be transmitted to UEs in a number of multicast groups,
located within multiple sectors of a satellite beam. We consider
a single typical TTI period, with all the tuneable thresholds
already assigned for current TTI. Derived from the worst case
scenario, where the processing time is the most expensive
among all possible scenarios, with the input size of n, the
involved computational time complexity (i.e. running time)
1930-529X/07/$25.00 © 2007 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.
Authorized licensed use limited to: University of Surrey. Downloaded on February 5, 2010 at 10:24 from IEEE Xplore. Restrictions apply.
4616

Page 5

0
5
10
15
20
25
Mean delay (seconds) .
MLPQ
23.6 1.11.71.0 0.910.6
DDQ
15.41.3 1.61.11.1 9.1
AMQ
13.41.1 1.61.1 0.98.7
FACH 1 FACH 2 FACH 3 FACH 4 FACH 5 FACH 6
0
1
2
3
4
5
Mean jitter (seconds)
MLPQ
3.40.8 1.70.7 0.94.1
DDQ
1.40.71.40.60.82.2
AMQ

1.80.51.00.5 0.52.3
FACH 1 FACH 2 FACH 3 FACH 4 FACH 5 FACH 6
0
2
4
6
8
10
12
14
16
18
Mean delay (seconds) .
FACH 1

FACH 2FACH 3FACH 4FACH 5FACH 6
scenario 1: streaming:0.1s,hot download:0.2s,cold download:0.8s
scenario 2: streaming:0.02s,hot download:0.2s,cold download:0.8s
scenario 3: streaming:0.1s,hot download:0.4s,cold download:0.8s
scenario 4: streaming:0.1s,hot download:0.2s,cold download:2.0s
Mean queuing delay for FACH queues under different delay thresholds
(a) (b) (c)
Figure 3. Queuing delay/jitter statistics for AMQ scheduling at the RLC buffer in SAT-Hub.
TABLE I.
S-CCPCH id
S-CCPCH bit rate
FACH id
Streaming
Hot Download
Cold Download
RADIO BEARER MAPPING CONFIGURATION (KB/S)
1
384
123
- 256 64
64 - -
- - -
23
384 384
6
-
-
384
45
256
-
-
128
-
-
required for MLPQ and DDQ are O(n) and O(n2) respectively,
whilst the AMQ algorithm requires an overall complexity of
O(n2), featuring typical quadratic statistics.
IV.PERFORMANCE EVALUATION
A.Simulation Methodology
In order to evaluate the performance enhancement of the
proposed cross-layer packet scheduling scheme, a system-level
simulator implementing the S-DMB system has been
developed with ns2 and MATLAB. Taking advantage of its
available built-in code blocks, relying heavily on the C++ code
modules, we developed additional code modules implementing
S-DMB specific features. The AMQ packet scheduling
mechanism is physically implemented in the SAT-Hub
(Node-B) employing the S-DMB functions, supporting three
types of QoS classes, namely: 1) real-time video streaming, 2)
hot download, and 3) cold download [10]. The streaming
traffic model applies publicly available trace files for video
streaming traffics. Traffic characteristics associated with hot-
and cold- download services -or, push-and-store services-
follow the ns-2 Pareto distribution, with different traffic
priority assigned. In addition, we choose different guaranteed
data rate in order to examine the performance between users
with different rate requirement.
Our link budget simulation results provide the Eb/No v.s
BLER look-up curves of each FACH. The simulation period is
set as 1000s or 12500 TTIs. Various queuing delay threshold
values are applied and examined for the specific scenario,
showing the range of the performance gain against tuning the
delay threshold parameter.
A wide variety of traffic mix scenarios and physical channel
capacities are evaluated via simulations, we select an indicative
scenario, where 6 individual MBMS sessions with diverse QoS
profiles in terms of service type, data rate, and QoS constraints
are considered for broadcast transmission; each session is
carried by a single FACH queue. Three Secondary Common
Control Physical CHannels (S-CCPCHs) are used for carrying
heterogeneous multimedia services, the considered radio bearer
mapping scenario is given as Table I. We compare the
performance of the proposed AMQ packet scheduling with
those of MLPQ and DDQ in this paper. Several main
parameters, which have significant impact upon the overall
system performance, are analyzed and discussed in the
following.
B.
Queuing delay evaluation
In Fig. 3(a), the queuing delay performance for AMQ is
compared to MLPQ and DDQ for all the allocated real-time
streaming and download sessions. Rather than achieving lower
download delay by scarifying streaming delay performance in
DDQ case, the proposed AMQ managed to deliver download
sessions with even further lower queuing delay whilst
maintaining the similar performance on streaming sessions.
From the viewpoint of human perception, it is worth noticing
that the delay variation of MBMS streaming service shall be
limited, to preserve the time variation between information
entities (i.e. packets) of the stream [11]. As seen from Fig. 3(b),
although the background services suffer from higher jitter for
AMQ than DDQ, a considerable performance gap with respect
to queuing jitter is achieved for all streaming services, which
makes it an attractive solution for real-time jitter-sensitive
streaming service.
Fig. 3(c) investigates the range of performance gain obtained
by adjusting variable delay threshold values. By tuning the delay
threshold value for a specified QoS service class, the AMQ is
capable of optimizing the delay performance amongst
competing flows. For example, in comparison with Scenario 1,
cold download FACH 6 suffers from worse delay in Scenario 4
when its delay threshold is increased from 0.8 second to 2.0
second, but this leads to the performance gain on the streaming
and hot download FACHs.
C. Channel utilization and buffer throughput
The impacts of AMQ on the performance of channel
utilization and throughput are studied. Herein the channel
utilization refers to the ratio obtained by dividing the total
information bits transmitted over the air with the maximum
supported capacity bits for considered physical channels. the
From Fig. 4(a), by adaptively re-utilizing wasted resources
among sessions with diverse QoS class, it isobserved directly
that AMQ has managed to offer better resource utilization over
the existing schemes.
1930-529X/07/$25.00 © 2007 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.
Authorized licensed use limited to: University of Surrey. Downloaded on February 5, 2010 at 10:24 from IEEE Xplore. Restrictions apply.
4617