Content uploaded by Wei Koong Chai
Author content
All content in this area was uploaded by Wei Koong Chai
Content may be subject to copyright.
Abstract—The rapid emergence of multimedia applications in the
Internet has highlighted the need for service differentiation in
broadband satellite networks, which aim at being an integral part
of the broadband network infrastructure. This paper presents a
novel scheduler, called SWTP (Satellite Waiting Time Priority),
to provide relative service differentiation in a DVB-RCS
geostationary (GEO) satellite system where the network is
structured to support a finite number of ordered service classes.
We advocate the adoption of the proportional service
differentiation model in the satellite domain to provide
proportional delay differentiation to different traffic classes. The
lightweight nature of the model makes it especially suitable for
satellite systems as it minimizes computational cost by doing away
with mechanisms such as admission control and resource
reservation. Simulation results suggest that the SWTP scheduler
can effectively and consistently provide proportional delay
differentiation in satellite networks.
I. INTRODUCTION
EXT generation networks are expected to be largely
heterogeneous and encompass multitude of networking
technologies. Broadband satellite networks aim at being
an integral part of this global communication infrastructure.
The rapid emergence of multimedia applications and the
growth of wireless data services have magnified the need for
quality of service (QoS) provision mechanisms that will satisfy
consistently the diverse service requirements across different
network segments. Given the expectation that future data
networks will provide service differentiation at the level of
service class, efficient and flexible yet simple and robust
mechanisms that serve the service differentiation purpose in
satellite networks become mandatory.
Service differentiation and QoS provision are not new
research topics and in fact, much research effort has been
devoted to them over the last couple of decades. Integrated
Services (IntServ) [1] and Differentiated Services (DiffServ)
[2] stand at the two edges of the proposed frameworks’ range;
the former invokes admission control and an explicit
reservation protocol (Resource Reservation Protocol – RSVP
[3]) to provide hard guarantees to individual traffic flows,
whereas the latter, more lightweight framework segregates
traffic into classes and provides them with softer QoS
guarantees. More recently and in between IntServ and
DiffServ, various proposals h ave emerged, which try to
compromise the robustness of provided QoS guarantees with
scalability [4]. The proportional differentiated services (PDS)
framework is one of these proposals [5], which opts to provide
relative QoS guarantees, namely the performance spacing
amongst different traffic classes is both predictable and
controllable by the network manager [5].
In this paper, we essentially adopt the PDS model into the
satellite network context; in particular, we introduce a
scheduler that draws on the Waiting Time Priority (WTP)
scheduler [6] and can serve the PDS model under the specific
requirements of satellite environments. The following section
gives a brief overview of the PDS model and reviews the work
carried out on the WTP scheduler in the context of the specific
service differentiation framework. In Section III we outline the
satellite system architecture to which we apply the service
differentiation model. Section IV formulates the bandwidth
allocation task under the particular service differentiation
requirements and presents our pr oposed solution , which is
evaluated with simulations in the section that follows. We
conclude the paper in Section VI.
II. THE PROPORTIONAL DIFFERENTIATION SERVICE MODEL
AND T HE WAITING TIME PRIORITY SCHEDULER
A. PDS Model
The PDS model is a lightweight framework for relative
QoS provision to a finite number of service classes. Assume
that the network wants to support N service classes. Each class
is associated with a differentiation parameter (DP), i
r. If i
σ
is
the performance metric of interest for class i or a proper
function of it, e.g., throughput, the inverse function of delay or
packet loss, then the PDS model requires
{}
.1,; Nji
r
r
j
i
j
i ∈∀=
σ
σ
(1)
We number classes in decreasing priority order, i.e., the
lower the class index, the better the service provided to it, and
normalize all DPs with reference to the highest priority class,
which is assigned DP equal to 1:
1...0 121 =<<<<< −rrrr NN
In this paper, the performance metric under consideration
is the average queueing delay i
d of class i. Hence, the model
requires that
{}
Nji
d
d
j
i
j
i 1,; ∈∀=
δ
δ
(2)
where i
δ
is the Delay Differentiation Parameter (DDP) of
class i.
Scheduling for Proportional Differentiated Service Provision
in Geostationary Bandwidth on Demand Satellite Networks
Wei Koong Chai
†
, Merkourios Karaliopoulos
‡
and George Pavlou
†
†Centre for Communication Systems Research, University of Surrey, Guildford, Surrey, GU2 7XH, UK
‡New Technologies and Applications Group, Teletel S.A., Athens, Greece
Email: {W.Chai, G.Pavlou}@surrey.ac.uk, M.Karaliopoulos@teletel.gr
N
matter experts for publication in the IEEE GLOBECOM 2005 proceedings.This full text paper was peer reviewed at the direction of IEEE Communications Society subject
IEEE Globecom 2005 0-7803-9415-1/05/$20.00 © 2005 IEEE3722
The WTP scheduler, on the other hand, originates from
Kleinrock’s Time Dependent Priorities queue [6]. Packets
from different classes enter first-in-first-out (FIFO) queues,
each associated with a DDP. The priority of each packet in the
queue is dependent on its DDP, i
δ
, and its waiting time in the
queue. If
()
t
wp
i is the waiting time of a packet p of class i at
time t, then its time-dependent priority Pi(t) is given by
() ()
i
p
ii t
w
t
P
δ
⋅=
()
p
p
itt
w
τ
−= (3)
where p
τ
is the arrival time of packet p.
The WTP scheduler always estimates the time-dependent
priorities of the head-of-line packets in all queues and
transmits the packet with the highest priority in a non-
preemptive manner.
B. WTP as an Enabler of the PDS Model
In [5], the WTP algorithm has been shown to approximate
closely the PDS model for heavily loaded wired networks with
Pareto traffic source. The WTP scheduler is then further
examined in [7] where the feasibility region of the scheduler is
characterised and an iterative algorithm to determine the
control parameters for obtaining the desired delay ratios is
proposed. In [8], three WTP variants are proposed, namely
maximum WTP (MWTP), variance WTP (VWTP) and
counting WTP (CWTP), that besides achieving the PDS model
objectives, also aim at reducing the absolute packet queueing
delay by considering the packet waiting times and packet
transmission times. A WTP scheduler that includes an adaptive
parameterisation scheme is proposed in [9] to approximate the
model in moderate load conditions. The Scaled Time Priority
(STP) [10] is proposed as a WTP variant with lower
complexity. In [11], a controller using fuzzy rules has been
introduced to reduce the effect of low priority class upon
higher priority ones in a proportional relative DiffServ
network. Based on the well-known Little’s Law, the authors in
[12] propose a scheduling mechanism that jointly controls the
delay and throughput metrics.
The PDS model has also attracted attention in the wireless
domain. By taking into account the wireless channel errors,
reference [13] presented the wireless WTP (WWTP)
scheduler. It tries to achieve PDS by providing higher
bandwidth compensation for higher priority classes when the
channel is in error state. In [14], a cross-layer WTP (CWTP)
scheduling algorithm (also named as distributed WTP (DWTP)
[15]) has been presented for wireless local area network
(WLAN). The Neighbourhood Proportional Delay
Differentiation (NPDD) [16] model is another WTP variant on
achieving PDS in the wireless domain. By keeping running
averages of the delays for the local and neighbouring nodes,
NPDD computes an index (ratio between local and
neighbouring delays), which is then used to map to the fixed
level of medium access control (MAC) priorities.
However, to the best of the authors’ knowledge, this is the
first work on realizing the PDS model in the satellite domain.
The WTP scheduler was first designed for terrestrial wireline
networks. There, the scheduler only needs to schedule the
departure of each contending packet locally. In wireless and
satellite domain, the access to the transmission medium is
often controlled in a distributed manner by a MAC protocol.
Thereby packets from one node may contend with packets
from other nodes, so that WTP sch eduler variants proposed in
the context of these networks cannot be applied directly to the
satellite domain.
Moreover, there are several fundamental architectural and
environmental differences that impede the adaptation of WTP
variants proposed in terrestr ial wireless networks to satellite
networks supporting dynamic bandwidth allocation
mechanisms. Firstly, for a bandwidth on demand (BoD)-based
satellite architecture, resource has to be requested by the
satellite terminals before they can make use of it, so that the
scheduler ends up scheduling requests for resource rather than
packets. Secondly, there is a non-negligible propagation delay
between the satellite terminals and the scheduler that may,
depending on the access control algorithm, inflate the waiting
time of a packet in the queue of the satellite terminal. The
impact of this semi-constant delay has to be taken into account
by the scheduler in providing relative service differentiation.
III. SYSTEM ARCHITECTURE
The system architecture under consideration is an Internet
Protocol (IP)-based broadband multimedia geostationary
(GEO) satellite network with resource allocation mechanisms
analogous to those described in the Digital Video Broadcasting
– Return Channel via Satellite (DVB-RCS) system standard
[17]. However, the discussion and the scheduler applicability
are not limited to DVB-RCS networks. Fig. 1 illustrates the
main nodes of the network architecture: -
• Satellite(s) – The satellite used is assumed to be equipped
with on-board processor (OBP) and the scheduler is located
on-board.
• Traffic Gateway (GW) – In line with the DVB-RCS
definition, GWs are included to provide interactive services
to networks (e.g., Internet) and service providers (e.g.,
databases, interactive games etc.).
Int er net
Sat FW
Sat RT
GW
ST
ST
ST
ST
ST
ST
ST
Service / Content
Pr o v i de rs
Fi
g
. 1 Reference satellite s
y
stem r esembli n
g
the DVB-RCS architecture
[
17
]
.
matter experts for publication in the IEEE GLOBECOM 2005 proceedings.This full text paper was peer reviewed at the direction of IEEE Communications Society subject
IEEE Globecom 2005 0-7803-9415-1/05/$20.00 © 2005 IEEE3723
• Satellite Terminal (ST) – STs represent the users. They
may serve one (residential) or more users (collective).
Time Division Multiple Access (TDMA) is used for the
forward path whereas on the return path, Multi-frequency
TDMA (MF-TDMA) is assumed. In a MF-TDMA frame, the
basic unit of the link capacity is the TS with multiple TSs
grouped in TDMA frames along several frequency carriers. In
this paper, we consider fixed MF-TDMA frame whereby the
bandwidth and duration of successive TSs is static.
The BoD process used in this work is derived from [18].
The procedure involves two main stages: the resource request
estimation and resource allocation. The main entities involved
are the BoD entity and BoD scheduler. The BoD entity is
located at the ST and handles all packets of the same class,
which are stored in the same queue i.e. there will be x BoD
entities in a ST if this ST supports x classes. BoD entities
periodically send slot requests (SRs) to the BoD scheduler,
when there are new packet arrivals at their queues. Upon
reception of SRs, the BoD scheduler allocates TSs to each
requesting BoD entity based on a certain scheduling discipline
and policies set up by the network operator. It then constructs
and broadcasts the burst time plan (BTP) that contains all the
allocation information to the BoD entities. Both stages are
repeated with a period equal to a multiple number of TDMA
frames ns, whilst one BoD cycle consists of k TDMA frames,
where k is an integer multiple of ns. Fig. 2 gives the BoD
timing diagram, which also describes the basic tasks involved.
IV. SWTP SCHEDULING DISCIPLINE
A. Problem Statement
Consider a satellite network with M geographically
distributed BoD entities serving N different service classes,
each one associated with a DDP, δi. BoD entity m, responsible
for service class i, periodically sends slot requests SRm
i. Let
i
ddenote the average queueing delay for packets of class i. In
this network, the bandwidth allocation task is formulated into
an online resource management problem as follows:
“Given a finite capacity C and the set of slot requests
{}
MmNiSRm
i≤≤≤≤ 11, , how does the BoD scheduler
allocate resources to BoD entity m so that for a given set of
DDPs,
{}
i
δ
, the PDS model objective of Eq. (2) is achieved.”
B. SWTP Algorithm
We consider an adaptation of the WTP algorithm, called
Satellite Waiting Time Priority (SWTP), as the BoD scheduler
that will serve Eq. (2). The basic idea of SWTP is that instead
of scheduling individual packets as in terrestrial networks, the
SWTP schedules the resource requests SRs from BoD entities.
B.1. SWTP Resource Request Rules
Since the scheduler and the queues are not physically co-
located, information regarding the waiting time of packets has
to be communicated to the BoD scheduler. The BoD entity
adds a timestamp to each request, which is used by the BoD
scheduler to estimate the request priorities and schedule them
according to (3) and (4). Here lies the second difference of our
scheduler from other WTP variants. Since each SR is
submitted for a batch of packets, i.e., the new arrivals within
the latest resource allocation period (RAP), BoD entities have
several alternatives for computing the SR timestamps. Unlike
terrestrial networks, SR timestamps do not necessarily
correspond to the arrival times of head-of-line packets as the
SWTP determines the aggregate priority of the whole request.
Formally, if m
i
Q is the set of newly arrived packets, i.e.
packets that came within the last RAP at the queue i of BoD
entity m, q its cardinality, and j
τ
the arrival time of packet j,
q
j
≤≤1, indexed in increasing order of arrival times, then the
BoD entity m may compute at time t the SR timestamp m
i
ts
using the subsequent rules:
1. According to the arrival time of the last packet that arrived
in the queue during the last RAP: qi tts
τ
−=
2. According to the arrival time of the first packet that arrived
in the queue during the last RAP: 1
τ
−= ttsi
3. According to the sample mean of the arrival times of all
packets that arrived in the queue during the last RAP:
∑
=
⋅−=
q
j
ji q
tts
1
1
τ
The first rule corresponds to the “worst” case, since the
request priority will be defined by the packet with the least
waiting time in the queue. Conversely, the second rule
maximizes the request priority by considering the packet with
maximum waiting time in the queue (best case). The third rule
is effectively a compromise between the former rules by
considering the waiting times of all newly arrived packets.
B.2. SWTP Resource Allocation
Satellite
(BoD
Scheduler)
ST
(BoD Entity)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
frame number
SR(1)
SR(2)
SR(3)
SR(4)
SR(5)
SR(6)
BTP
(SR(1))
BTP(SR(2))
BTP
(SR(3))
BTP(SR(4))
BTP
(SR(5))
(1) Esti mate and send
res our ce r eques t ( SR ).
(2) Update request buffer (3) Allocate Resources
(4 ) Proc ess receiv ed BTP (5) Activate new BTP
1 BoD cycle
propagation delay
Fig. 2 BoD timing diagram
matter experts for publication in the IEEE GLOBECOM 2005 proceedings.This full text paper was peer reviewed at the direction of IEEE Communications Society subject
IEEE Globecom 2005 0-7803-9415-1/05/$20.00 © 2005 IEEE3724
The allocation process is centralized. The BoD scheduler
computes the priority of each SR stored in the request buffer.
The priority
()
k
Pm
i allocated to SR m
i at kth RAP is given by
() ()
()
α
δ
+⋅= k
w
k
PSR
ii
m
i (4)
whereby α is a constant added to the waiting time of packets to
account for the propagation delay of BTP and the processing
delay of BoD entities, while
()
m
i
SR
itstk
w−= and m
i
ts is the
timestamp information encoded in each SR based on on e of the
resource request rules presented in section B.1. Compared to
other WTP schedulers, there are two significant differences in
the computation of the priority. Firstly, there is an addition of
α to account for the exact packet transmission time due to the
fact that there is non-negligible delay between the allocation
time and the actual packet transmission time. Secondly,
alth ough the equation for
()
k
wSR
i remains unchanged, it must
be noted that t here refers to the time when the BoD scheduler
is actually computing the priority i.e., t no longer correspond to
the packet service instance as in terrestrial network. Note that
in the considered BoD system, the functions can be in terms of
k instead of t due to the periodical nature of the system.
At each allocation period, the SWTP scheduler allocates
TSs by considering requests in decreasing priority order:
requests can be full y satisfied as long as they do not exceed the
available capacity. Therefore, at kth period, the SWTP will first
allocate TSs to the SR with the highest priority among all SRs
buffered at the BoD scheduler.
The process proceeds until all TSs for the MF-TDMA
frame have been allocated. Those requests that are not
considered or partially satisfied will be buffered for the next
allocation period. When the next RAP arrives, the priorities of
these buffered requests are recalculated to account for the
additional waiting time of the request at the scheduler queue.
The algorithms executed by the BoD entities and
scheduler are given in Table I and II respectively. At the BoD
scheduler, the allocation process is divided into two steps.
Firstly, for each request, a priority metric has to be computed.
Then based on these priorities, TS allocations are determined.
Finally, the allocation information is broadcast to BoD entities
in the form of the BTP.
V. SIMUL ATION RESULTS
The capacities for all links in this paper are configured to
be 2048kbps. Unless explicitly stated otherwise, the network is
set to have DDPs: 1, 1/2, 1/4, 1/8. The IP packet size used in
all simulation is 500 bytes, while MAC frames are of size 48
bytes with additional 5 bytes header. Unless explicitly stated
otherwise, all simulations use Poisson traffic sources.
A. Evaluation of Slot Request Rules
Simulation runs of 100s with four service classes under all
three SR estimation rules described in section IV (B.1) have
been carried out. Fig. 3 plots the achieved differentiation ratio,
normalized with respect to the target ratio
δδ
ji . Intuitively,
the more complex rule 3 would perform most accurately.
However, simulation shows that all three rules give similar
performance in terms of achieving the accurate proportionality
in delay. All of them yield ratios close to the ideal value
(=1.0). Rule 1 actually achieves better performance though
only very marginally.
The performance of each slot request rule has also been
assessed in short-term. Fig. 4 shows the individual packet
delay upon departure for all three rules i.e. the delay suffered
by each packet when departing from the ST is logged down
and plotted. Three observations can be made: -
1. Delay – All three achieve similar delay performances.
2. Consistency – only rule 1 satisfies the requirement “higher
class always performs better than lower class”. The
requirement is violated by the other two rules.
3. Fluctuation – Comparatively, less fluctuation is seen for
rule 1. Smaller packet delay deviations are generally
desirable for real-time streaming video applications.
It is found that rule 1 performs the best. Firstly, it emulates
most closely the PDS model; hence satisfying the
“controllability” property of the model. Secondly, it performs
consistently adhering to the order of different service class.
Hence, it satisfies the “predictability” property of the model.
Thirdly, it gives smaller packet delay variation. Therefore, it is
the best option for real-time multimedia applications. From
here onwards, all simulations use SR estimation rule 1.
B. SWTP in Achieving PDS Model
Fig. 5(a) shows the queueing delay for each service class,
while Fig. 5(b) presents the corresponding delay ratios under
constant bit rate traffic. The ideal value for the ratios is 0.5 for
all cases. From the plotted results, it is clear that the SWTP
scheduler can indeed emulate closely the PDS model.
Since the PDS model requires that the spacing between
any two service classes adhere strictly to the ratio of the DDPs
for specified service classes, the scheduler should not be
dependent on the traffic distribution between service classes.
Fig. 5(c) shows the result of this test at utilization, U=95%.
TABLE I: SWTP ALGORITHM FOR BOD ENTITIES
1. BEGIN (k = request time)
2. get the number of newly arrived packets in the
current RAP,
()
k
qm
i
3. compute resource request,
()
k
SR m
i
4. read
δ
i
5. compute
()
k
ts m
i
6. send request packet [ST id,
()
k
SR m
i,
()
k
ts m
i,
δ
i]
7. END
TABLE II: SWTP ALGOR ITHM FOR BOD CONTRO LLER
1. BEGIN (k = allocation time)
2. for each request
{compute
()
k
Pm
i based on
()
k
wSR
i and
δ
i}
3. while (TS available)
{allocate
()
k
SR m
i TS to the request with the
highest
()
k
Pm
i unconsidered request}
4. buffer all unconsidered requests
5. broadcast new BTP
6. END
matter experts for publication in the IEEE GLOBECOM 2005 proceedings.This full text paper was peer reviewed at the direction of IEEE Communications Society subject
IEEE Globecom 2005 0-7803-9415-1/05/$20.00 © 2005 IEEE3725
The ideal value is 0.5 and in all simulations, the achieved
ratios are very near to this value.
C. Controllability of SWTP
This section illustrates the capability of SWTP in
accurately controlling the spacing between different service
classes. Three sets of DDPs have been defined below.
• Set A – [1, 1/2, 1/4, 1/8]
• Set B – [1, 1/2, 1/3, 1/4]
• Set C – [1, 1/4, 1/5, 1/6]
Simulations with U= 95% have been conducted based on
these DDP sets and the results given in Fig. 6 shows the
normalized ratios of all three cases. With the ideal value as 1.0,
it can be concluded that SWTP indeed is able to control the
class spacing. However, due to the long propagation delay, the
spacing between the highest and lowest DDP should not be too
large to ensure reasonable delay for the lowest class.
D. Predictability of SWTP
The beh aviour of SWTP in short timescale is investigated
to ensure that the predictability requirement of the PDS model
is satisfied. Fig. 7 shows the individual packet delays upon
departure of four-class scenario for a period of 100ms. The
graph shows that SWTP can consistently provide the
appropriate spacing for the service classes.
E. Feasibility Region of SWTP
Up to this point, the results given are all obtained from
simulation runs under high network load. However, it is found
that, similar to other WTP schedulers, SWTP suffers from the
same problem of n ot being able to provide service
differentiation when running under low load. Fig. 8 shows the
delay ratios for a four-class network against different levels of
network utilization. Similar to [5], SWTP cannot maintain the
service class spacing defined by the DDPs when the network
load is lower than 80%. Although the performance of each
service class still maintains the pre-defined order, the spacing
between the classes deviates fr om the settings provided.
However, service differentiation is only needed when the
network load is high. Forcing service differentiation by
depriving customers of lower priority available bandwidth is
unnecessary. Maximizing the utilization of satellite bandwidth
is of greater importance here.
VI. CONCLUSIONS
This paper studies the problem of providing proportional
service differentiation in GEO BoD satellite networks. It
advocates the adoption of the proportional differentiation
model within the satellite domain and proposes as well as
evaluates a novel scheduler in this model context.
Simulation results show that the SWTP scheduler can
serve the desired service differentiation model in different
settings. Three slot request estimation rules have been devised
and evaluated. It is found that rule 1 exhibits the best
performance. Regarding resource allocation, the SWTP
scheduler is shown to be capable of providing proportional
service differentiation. It is also able to operate in conformance
with two important properties of the PDS model; namely the
controllability and predictability properties. The spacing
between classes can be controlled via the DDP of each service
class, while higher classes always performs consistently better
than lower classes. The feasibility region of the scheduler has
also been presented.
ACKNOWLEDGMENT
This work is performed within the framework of the SatNEx
project, funded by European Commission (EC) under the
Research Framework Program 6. The financial contribution of
the EC towards this project is greatly appreciated.
REFERENCES
[1] R. Braden, D. Clark and S. Shenker, “Integrated Services in the Internet
architecture: an overview,” RFC 1633, June 1994
[2] S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang and W. Weiss, “An
archit ecture for Differ entiat ed Ser vice,” RFC 2475, December 1998
[3] J. Wroclawski, “The use of RSVP with IETF Integrated Services,” RFC
2210, September 1997
[4] M. Welzl and M. Mu hlhau ser, “S calability and quality of service: a
trade-off?,” IEEE Communications Magazine, pp. 32-36, June 2003
[5] C. Dovrolis, D. Stiliadis and P. Ramanathan, “Proportional differentiated
services: delay differentiation and packet scheduling,” IEEE/AC M
Transactions on Networking, vol. 10, no. 1, pp. 12-26, Feb 2002
[6] L. Kleinrock, Queueing Systems. New York: Wiley, 1976, vol. II
[7] M. K. H. Leung, J. C. S. Lui and D. K. Y. Yau, “Adaptive proportional
delay differe ntiate d services: characterization and performa nce
evaluation,” IEEE/ACM Transactions on Networking, vol. 9, no. 6, pp.
801-817, Dec 2001
[8] Y. Lai, “Packet scheduler s to provide proportional delay differentiation
and reduce packet queuein g delay simultanou esly,” IEEE International
Conference on Communications, vol. 4, pp. 1968-1972, June 2004
[9] L. Essafi, G. Bolch and A. Andres, “An adaptive waiting time priority
scheduler for th e proport ional di fferentiation model,” in Proc. of the
High Performance Computing Symposium, 21 Sept 2000
[10] H. Ngin a nd C. T ham, “ Achievi ng proportional delay differentiation
efficiently,” Computer Communications, vol. 27, issue 2, pp. 153-161,
Elsevier Science, Feb 2003
[11] S. Patchararungruang, S. K. Halgamuge and N. Shenoy, “Optimized
rule-based dela y proportion adjustment for proportional differentiated
services,” IEEE Journal of Selected Areas in Communication, vol. 23,
no.2, pp. 261-276, Feb 2005
[12] S. Sankaran and A. E. Kamal, “A combined delay and throughput
proportional scheduli ng scheme for differentiated services”, IEEE
Globecom, Nov 2002
[13] M. R. Jeong, K. Kakami, H. Morikawa and T. Aoyama, “Wireless
scheduler providing relative delay differentiation”, in Proc. of the 3rd
International Symposium on Wireless Personal Multi media
Communications, Bangkok, Thailand, pp. 1067-1072, Nov 2000
[14] Y. Xue, K. Chen and K. Nahrstedt, “Achieving proportional delay
differentiation in wireless LAN via cross-layer scheduling,” Journal of
Wireless Communications and Mobile Computing, pp. 849-866, Wiley-
InterScience, Nov 2004
[15] Y. Xue, K. Chen and K. Nahrstedt, “Distributed end-to-end proportional
delay di fferentiatio n in Wireless LAN”, IEEE International Conference
on Communications, vol. 7, pp. 4367-4371, 20-24 June 2004
[16] K. Wang, “Quality of service assurances in multihop wireless networks,”
PhD. Dissertation, University of Wisconsin-Madison, 2003
[17] ETSI EN 301 790, “Digital Video Broadcasting (DVB); Interaction
channel for satellit e distribution systems,” ETSI European Standard
(Telecommunications series), EN 301 790 V1.3.1 (2003-03)
[18] G. Açar, “End-to-end resource management in geostationary satellite
networks,” PhD. Dissertation, University of London, Nov 2001
matter experts for publication in the IEEE GLOBECOM 2005 proceedings.This full text paper was peer reviewed at the direction of IEEE Communications Society subject
IEEE Globecom 2005 0-7803-9415-1/05/$20.00 © 2005 IEEE3726
c1/c2 c2/c3 c3/c4 c1/c3 c2/c4 c1/c4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Request Strategy Comparison
Class x / Class y
Normalized Ratio
Strategy 1
Strategy 2
Strategy 3
Fig. 3. The behaviour of t he sch eduler under the thr ee SR rules i s similar.
010 20 30 40 50 60 70 80 90 100
0
2
4
6
8
10
12
14
16
time (sec)
Individual Packet Delays upon Departure (sec)
Individual Packet Delays upon Departure vs time
Class 1
Class 2
Class 3
Class 4
(a)
010 20 30 40 50 60 70 80 90 100
0
2
4
6
8
10
12
14
16
18
time (sec)
Individual Packet Delays upon Departure (sec)
Individual Packet upon Departure vs time
Class 1
Class 2
Class 3
Class 4
(b)
010 20 30 40 50 60 70 80 90 100
0
2
4
6
8
10
12
14
16
18
time (sec)
Individual Packet Delays upon Departure (sec)
Individual Packet upon Departure vs time
Class 1
Class 2
Class 3
Class 4
(c)
Fig. 4. Per-packet delay by SWTP scheduler under (a) SR rule 1, (b) SR rule 2, (c) SR rule 3. Only SR rule 1 obeys the “predictability” property of the PDS model.
550 600 650 700 750
0
2
4
6
8
10
12
14
Average Queueing Delay Vs Input Traffic Rate
Input Traffic Rate (kbps)
Average Queueing Delay (sec)
class 1
class 2
class 3
class 4
(a)
550 600 650 700 750
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Delay Ratio Vs Input Traffic Rate
Input Traffic Rate (kbps)
Delay Ratio
class 1 / class 2
class 2 / class 3
class 3 / class 4
(b)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Average Delay Ratio vs Class Load Distribution (U = 95%)
Class Load Distribution (1/2/3/4)
Average Delay Ratios
Class 1 / Class 2
Class 2 / Class 3
Class 3 / Class 4
40/40/10/10
10/40/40/10
10/10/40/40
10/40/10/40
40/10/40/10
40/10/10/40
25/25/25/25
(c)
Fig. 5. (a ) Queueing dela y of different service class following the specified spacing of the model; (b) the corr esponding delay ratios achieved whereby they are closed
to the ideal delay ratios; (c) SWTP emulating the PDS in different load distributions with all valu es achieved close to the ideal value.
Set A (1,1/2,1/4,1/8) Set B (1,1/2,1/3,1/4) Set C (1,1/4,1/5,1/6)
0.7
0.8
0.9
1
1.1
1.2
Normalized Ratio vs DDP Set
DDP set
Normalized Ratio
Class 1 / Class 2
Class 2 / Class 3
Class 3 / Class 4
10 10.01 10.02 10.03 10.04 10.05 10.06 10.07 10.08 10.09 10.1
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Individual Packet Delays upon Departure Vs time
time (sec)
Individual packet delays upon departure (sec)
class 1
class 2
class 3
class 4
70 75 80 85 90 95 100
0
0.5
1
1.5
Utilization vs Delay Ratio
Utilization (%)
Delay Ratio
Class 1 / Class 2
Class 2 / Class 3
Class 3 / Class 4
Fig. 6. SWTP with 3 sets of DDPs: all normalized
delay ratios achieved are close to the ideal value
Fig. 7. Short time scale behaviour of SWTP
showing its predictability property
Fig. 8. Feasibility region of SWTP similar to other
WTP schedulers
matter experts for publication in the IEEE GLOBECOM 2005 proceedings.This full text paper was peer reviewed at the direction of IEEE Communications Society subject
IEEE Globecom 2005 0-7803-9415-1/05/$20.00 © 2005 IEEE3727