Sie sind auf Seite 1von 6

A Review of Analytical Model for Hybrid PQ-WFQ Scheduling Scheme for WiMAX Networks

Daniel Ndiki ECE-249, Project Paper Review

Abstract—One of the most significant promises in today’s Hierarchical Schemes – a two layer scheduling structure
integrated packet-switched broadband wireless that combines scheduling for UL and DL bandwidth
networks (such as WiMAX) is providing Quality of simultaneously and iv) Channel aware schemes –a
Service (QoS) guarantees to data, voice and video modification of wired networks schemes that attempt to
applications. Scheduling schemes are used to ensure address WiMAX channel variability while addressing the
QoS guarantees for this wireless applications. Many QoS requirements [ CITATION Ndiki2010 \l 1033 ]5.
research efforts have been dedicated to performance Due to the varied QoS requirements of Applications that
evaluation of traffic scheduling schemes. In this transverses a WiMAX network, use of hybrid schemes i.e.
paper, we present a discussion of on the Analytical using two or more schemes is suggested [[ CITATION
Model for Hybrid PQ-WFQ Scheduling Scheme for Ndiki2010 \l 1033 ]5. The general idea is to differentiate
WiMAX Networks using Markov Chains traffic based on QoS requirements and then individually
select a best-fit scheduling algorithm that can adequately
I. INTRODUCTION allocate bandwidth to individual traffic flows .i.e., different
Worldwide Interoperability for Microwave Access scheduling algorithms tailored to the traffic requirement of
(WiMAX) is a part of IEEE 802.16[ CITATION 2005 \l each Service Class. In a hierarchical scheme, a two-layer
1033 ]1, [ CITATION 2008 \l 1033 ]2 family of broadband approach is used to schedule traffic, where in the first layer,
wireless access (BWA) network standards for fixed and scheduling algorithms separates and might give priority to
mobile application. WiMAX networks, like other broadband DL and UL traffic. In the second stage, classical or hybrid
wireless access networks multiplex different applications, schemes are used to schedule the various QoS Service
e.g., voice, video or data that have different quality of Classes. A number of hierarchical scheduling have been
service (QoS) requirements. In such an environment, these proposed in the literature, [ CITATION Li2007 \l 1033 ]3,
applications contend for resources such as bandwidth. [ CITATION Ndiki2010 \l 1033 ]5. Deficit Fair Priority
Scheduling schemes are used to resolve contention for these Queuing (DFPQ) is used in the first layer, together EDF,
shared resources by allocating resources to users and WFQ and RR, Priority Queuing (PQ) and WFQ, and EDF ..
determining their transmission priority, that is, a scheduler Our focus is on analytical work from [ CITATION
allocates resources and establishes the order in which Wang2009 \l 1033 ]6 which present a three-dimensional
information flows are served ensuring that QoS Markov Model for modeling PQ-WFQ hybrid scheduling
requirements of each information packet are guaranteed. system applied to multiple non-bursty and busrty traffic.
A base station (BS) several mobile stations (MS) by The rest of the paper is organized as follows. In Section II,
coordinating the communication in the downlink (DL) i.e. we describe PQ-WFQ, stating the requirements for different
communication from BS to SS, and uplink (UL) SS to BS. applications. We describe the Markov model for modeling
The BS controls access to the media by providing resources PQ-WFQ with non-bursty traffic for different classes of
in form of slots to the SS [ CITATION Li2007 \l 1033 ]3, traffic flow in Section III and discuss its application to no-
[ CITATION Abdel-Hamid2009 \l 1033 ]4. The busrty traffic modeled on Markov Modulated Poisson
management and control functionality including QoS and Process (MMPP) in section IV. We discuss the results in
bandwidth resource assignment is done at the MAC layer. section V. Finally, we draw conclusion in Section VI.
The MAC layer is connection oriented accommodating both
continuous and bursty traffic in order to support constant bit II. PRE-EMPTIVE PRIORITY POLICY USING PPQ-
rate, real-time variable bit rate and best effort traffic. The WFQ SCHEDULER
standard makes it feasible to address QoS requirements by A scheduling scheme can be implemented either as Pre-
defining five different QoS service classes: Unsolicited emptive priority- where the ongoing services is interrupted
grant services (UGS), Real time polling services (rtPS), whenever a higher priority packet enters the network or as
Extended real-time polling service (ErtPS), Non-real time non-pre-emptive priority- where an arriving high priority
polling service (nrtPS), and Best effort service (BS). packet does not interrupt on on-going service to low priority
Scheduling schemes architecture for WiMAX has not been packet. Pre-emptive priority is further classified into i)
standardized and thus provide an important mechanism for pre-emptive resume policy if the interrupted or pre-empted
vendors to differentiate their equipment. The choice of a packets can resume services from the point at which it was
scheduling algorithm will depend on its complexity and ease interrupted, ii) pre-emptive repeat policy – when the service
of implementation. interruption clears, the interrupted service repeats with the
WiMAX Scheduling architectures can broadly be same service period, and iii) pre-emptive but with different
classified into four categories: These are Classical Schemes- repeat policy. In each of these pre-emptive policy, all lower
that utilize simple scheduling mechanism such as the ones priority packets does not influence the stochastic lae of
proposed for wired networks, ii) Hybrid scheduling schemes higher priority process. Consequently, the presence of low
that use a combination of classic scheduling algorithms iii) priority packets has no effect on traffic parameters such as
1
stationery distribution of the queue size or waiting time for (i,j,k) to (i-1,j,k),(i, j-1,k), (i, j, k-1) indicates service rate µ,
the higher priority packet flows. µ2, µ3 respectively from queue 1, 2 and 3. Queue 2 and
The discussion that follows focuses on pre-emptive queue 3 traffic are only scheduled if queue 1 traffic is empty
resume schemes using PQ scheduling for higher priority i.e. state from (0,j,k) to,(0, j-1,k) or ( 0, j,k-1) correspond to
packets while WFQ scheduling is used to scheduling lower service rates µ2, and µ3 for queue 2 and queue 3
priority packets as present in [ CITATION Wang2009 \l respectively. The marginal state probability Pi that I
ci
1033 ]6. Priority Queuing (PQ) assigns multiple queues
packets of class ci are in the queue are based on joint state
different priority levels. A queue with a higher priority is
probability pijk and is given by
always processed ahead of a queue with a lower priority.
Queue assigned the same priority are processed in a round-
robin fashion. Weighted Fair Queuing (WFQ) makes
reference to GPS, and assigns available bandwidth to traffic
queues based on weights. For example, it can assign more
weight to nrtPS and less to BE traffic packets since nrtPS
has higher priority. It has good bounded delay, can
guarantee throughput and observes fairness (ref). However,
the need to serially search among all the packets in the
queue makes it complex to implement.

III. SYSTEM DESCRIPTION


The authors consider a single server (possibly a single
Base Station (BS)) queuing system with a hybrid PQ-WFQ
scheduling system as shown in the fig. I. It has queues Qi, i
=1,2,3 and where traffic classes Ci access respectively. The
queues have priority such that traffic Qi > Qi+1, i = 1, 2, 3,
and each is bounded by a corresponding finite capacity Li.
Therefore, traffic class c1 has the highest priority and its
scheduled first using PQ. Pre-emptive priority also allows c1
Fig. 3: State Transition Diagram
to interrupt service for the other lower priority queues c2 and
c3. This is similar to UGS or ErtPS traffic class in WiMAX.
c2 and c3 are scheduled using WFQ, where c3 is assigned a
c1
lower weight w3- similar to a non-real time service such as
nrtPS or BE.
c2
Packets weights for each queue is fixed, all packets are
assumed to have the same size, and have the same inter-
c3
service time µ-1, so that traffic flow c =2,3 will achieve
average service flow µc= µwc/(w2+w3). Note that, a space in
the buffer can be released only after packet has been served
completely. The mean marginal number of packets in queue is given
by

At steady state, the marginal throughput Tc is given by


arrival rate multiplied by marginal probability

And from Little’s Law, the marginal queuing delay Dc is


Fig. 1: PQ-WFQ scheduling scheme with Poisson Arrivals. given by Lc/Tc. Its noted that, an arriving packet for class ci
can be dropped only if its full, and hence the packet loss
The architecture is a PQ-WFQ scheduling process with probability is equal to the marginal state probability that
three class traffic is represented using the state transition there is no space available in the queue
(i,j,k) (0≤ i ≤ L, 0≤ j ≤ L, 0≤ k ≤ L) corresponding to the
situation where there are i, j, k packets of the three classes
respectively. The corresponding transition rates are λ1, λ2, λ3 While fairness for the two classes is given by Jain’s fair
corresponding to i+1, j+1 and k+1 transition, where packets index i.e.
from class 1, 2, and 3 arrives into the queue, while rate from
2
.

3
IV. MODEL WITH BURSTY TRAFFIC As illustrated, mean marginal throughput for class 2
To capture the effect of bursty traffic, the authors extend and 3 traffic flow changes in inverse proportions with
the model so that the arrival process of high priority traffic respect to changes in weight ratio. Increased class 2 weight
flow is modeled by a two-state MMPP as shown in fig. 2. ratio leads to increased queuing delay and packet loss for
MMPP allows for the capture of traffic sources that are class 3 (Fig 4-5), while fairness between class 2 and 3
bursty in nature by employing an auxiliary Markov process. increasing to a maximum when the weight ration is 60% and
In MMPP, the current state of the Markov process controls thereafter decreases. There is no explanation given, but
the probability distribution of the traffic. MMPP is a seems to agree with [ CITATION Fang2003 \l 1033 ]7
Poisson process whose parameters are controlled by a where the mean packet delay increases rapidly when the
Markov process, that is, when the underlying Markov state offered load reaches 65%.
is at state i, (i = 1,2), the message arrive according to Fig 6 traces the effect of class 1 burstiness on the
Poisson process with rates λi. The MMPP is specified by a performance metrics for class 2 and 3, where as expected,
infinitesimal generator matrix QMMPP and rate matrix Λ so high burstiness reduces performance metrics including
that [ CITATION Fang2003 \l 1033 ]7 fairness for class 2 and 3 traffic flow. Burstiness variation
is represented by the rate matrix fixed at
QMMPP ¿
[−rr 10 −rr 01]
Λ=
[ λ01 λ02] and then for the second part increased it to

where r0, r1 is the low and mean duration in the high


states of burstiness respectively.
The priority packet flow burstiness for class 1 traffic is while using the same infinitesimal generator QMMPP.
then captured by building a four dimensional Markov chains Correspondingly, there is no appreciable change as
on the two-three dimensional Markov chains proposed. The anticipated for class 1 traffic parameters.
transition between the corresponding states form one layer
to the other represent the transition of underlying Markov
chain in two-state MMPP. The steady state probability P
will have four dimensions including the added state, and
satisfies equations postulated in section III above. Joint
state probability pijks, the marginal state probability pic, and
other performance metric are calculated likewise.

V. RESULT ANALYSIS
The parameters used are set in table I and II. The
capacity of each queue and service rates µ are the same and
held constant. The weight ratio w 2/(w2+w3) which is varied
from 10% to 90% is compared to marginal performance
metrics such as mean marginal number of packets in queue,
throughput, mean queuing delay, packet loss probability and
fairness (fig. 3-7). Fig. 4 Mean Marginal Number of Packets in the Queue v/s
Weight Ratio

Fig. 5: Mean Marginal Delay v/s Weight Ratio

4
V. CONCLUSION on channel quality measurement provided by the SS to
The paper provides an analytical framework at the the BS- all of which will affect the arrival and service
performance of a hybrid scheduling scheme incorporating rates of the various traffic class flows.
PQ and WFQ schemes using a Markov models. The model WiMAX further to the PHY profile being
is extended to show the effect of traffic burstiness using a
adaptive, it also utilizes dynamically architecture that
four-dimensional MC- the result being that increased
burstiness degrades performance of lower priority class
varies the resources assigned UL and DL sub-frame,
traffic. However, since there is a set limit on the traffic for which the paper does not address, although one can
each of the traffic classes, and the analysis does not show foresee a extension to the busrty MMPP model – in
the outcome of increasing arbitrary class 1 traffic. this case, having 8 dimensions.
The class of high traffic flow that demands
enhanced QoS characteristics such as low delay, low
jitter etc are put in highest priority queues with the
lowest drop probability. However, as the downside of
these control measures is to it starve the performance
behavior of the rest of lower priority traffic flows, and
special consideration calls for an upper limit for the
rate, and burst size if appropriate, for these high
priority traffic. In this case, this is achieved by
defining queue limit, in other works, [ CITATION
Li2007 \l 1033 ]3, [ CITATION Ndiki2010 \l 1033 ] 5 an
admission control and packet classifier is suggested
that ensure that some classes of traffic flow such as
Fig.6 Mean Marginal Queuing Delay v/s Weight Ratio for
UGS are not subject to scheduling.
Bursty Traffic
Similarly, low priority packets such as BE and
The paper does not mention whether the MAC nrtPS have to contend for resources allocation by
architecture in use is FDD or TDD, and on whether sending random bandwidth requests (BW-REQ)
this analysis was done for UL or DL- on the BS on during the previous UL period [ CITATION Abdel-
Hamid2009 \l 1033 ]4. MS will collide if more than one
MS. For UL, a BS typically have sufficient
knowledge of the environment, where else it has to MS tries to access the same resources during the
rely the channel to interference and noise ratio (CINR) contention process.
information sent whenever the SS request bandwidth.
Similarly, the bursft profile assigned to an MS is based REFERENCESx
x

x
[1] Approved Draft IEEE Standard for Local and metropolitan area networks Corrigendum to IEEE Standard for Local and
Metropolitan Area Networks-Part 16: Air Interface for Fixed Broadband Wireless Access Systems (Incorporated into IEEE Std
802.16e-2005 and IEEE Std 802.16-2004/Cor 1-2005 E), 2005, IEEE Std P802.16/Cor1/D5.
[2] IEEE Draft Standard for Local and metropolitan area networks Part 16: Air Interface for Broadband Wireless Access Systems
((Revision of IEEE Std 802.16-2004 and consolidates material from IEEE Std 802.16e-2005, IEEE Std 802.16-2004/Cor1-2005,
IEEE Std 802.16f-2005 and IEEE Std 802.16g-2007)), 2008, IEEE Unapproved Draft Std P802.16Rev2_D4 Apr 2008.
[3] Bo Li, Yang Qin, Chor Ping Low, and Choon Lim Gwee, "A Survey on Mobile WiMAX [Wireless Broadband Access],"
#IEEE_M_COM#, vol. 45, no. 12, pp. 70-75, 2007.
[4] Y. Abdel-Hamid, F. Gebali, and T. A. Gulliver, "A QoS model for random access systems with multiple request classes," , 2009,
pp. 331-335.
[5] D. Ndiki, H. J. Helgert, and S. Hussein, "A Comparative Overview IEEE 802.16e QoS Scheduling Algorithms," , 2010, pp. 74-
79.
[6] Lan Wang, Geyong Min, D. Kouvatsos, and Xiaolong Jin, "An analytical model for the hybrid PQ-WFQ scheduling scheme for
WiMAX networks," , 2009, pp. 492-498.
[7] Xiaoyan Fang and D. Ghosal, "Performance modeling and QoS evaluation of MAC/RLC layer in GSM/GPRS networks," , vol.
1, 2003, pp. 271-275.
[8] Ke Yu, Xuan Wang, Songlin Sun, Lin Zhang, and Xiaofei Wu, "A Statistical Connection Admission Control Mechanism for
Multiservice IEEE 802.16 Network," , 2009, pp. 1-5.
[9] Mingyan Kou, Xiaoyi Wang, and Huimin Xu, "Channel-Aware Adaptive ARQ Packets for WiMAX," , 2009, pp. 1-4.
[10] G. Ciccarese, M. De Blasi, P. Marra, C. Palazzo, and L. Patrono, "A Packet Size Control Algorithm for IEEE 802.16e," , 2008,
pp. 1420-1425.
5
[11] IEEE DRAFT Standard for Local and metropolitan area networks Part 16: Air Interface for Broadband Wireless Access Systems
(Revision of IEEE Std 802.16-2004 and consolidates material from IEEE Std 802.16e-2005, IEEE Std 802.16-2004/Cor1-2005,
IEEE Std 802.16f-2005 and IEEE Std 802.16g-2007), 2009, IEEE Unapproved Draft Std P802.16Rev2/D9a March 2009.
[12] Chakchai So-In, R. Jain, and A.-K. Tamimi, "Scheduling in IEEE 802.16e mobile WiMAX networks: key issues and a survey,"
#IEEE_J_JSAC#, vol. 27, no. 2, pp. 156-171, 2009.
[13] Ying-Hsin Liang, Ben-Jye Chang, Sung-Ju Hsieh, and De-Yu Wang, "Analytical Model of QoS-Based Fast Seamless Handoff
in IEEE 802.16j WiMAX Networks," #IEEE_J_VT#, vol. 59, no. 7, pp. 3549-3561, 2010.
x

Das könnte Ihnen auch gefallen