QoS SCHEDULING FOR IEEE 802.16e MESH MODE USING

GENETIC ALGORITHMS

Neila Krichene and Noureddine Boudriga

Communication Networks and Security Research Lab, University of the 7th of November at Carthage, Tunis, Tunisia

Keywords:

IEEE 802.16e mesh mode, rtPS scheduling, Genetic algorithms.

Abstract:

IEEE 802.16e amendments for the mesh mode do not specify particular QoS mechanisms. However, each

Mesh Subscriber Station (MSS) transmits its own trafﬁc and forwards the trafﬁc of its children while each

trafﬁc ﬂow has its particular QoS constraints. As all forwarded trafﬁcs use the same link, one MSS may ex-

perience starvation or act selﬁshly. Besides, the MSS’s mobility adds important complexity as all the affected

ﬂows need to be re-routed, re-admitted and re-scheduled while respecting their original QoS requirements. In

this paper, we adopt the concept of assigning ﬁve virtual nodes to each MSS reﬂecting the ﬁve QoS classes

of IEEE 802.16e and we propose scheduling the uplink transmissions of the real-time Polling Service (rtPS)

trafﬁc class by adopting the Genetic Algorithms concept. We deﬁne the priority assigner component which

communicates with the scheduler of the mesh BS in order to set and update ﬂows’ priority and we optimize the

time spent by a ﬂow in the queue while implementing the mobility constraints within the mutation function.

1 INTRODUCTION

The recent years have been marked by a growing need

for providing advanced applications and Internet-

related services at high throughput and low costs

while guaranteeing the required QoS and a continuous

and open access to such services. In order to address

such need at the metropolitan scale, the IEEE 802.16e

amendments implement service differentiation and

adopt a connection-oriented philosophy within a mo-

bility context but they left many QoS functions un-

speciﬁed so that researchers and constructors can de-

sign and adopt the most suited mechanisms that ful-

ﬁll particular requirements. Moreover, IEEE 802.16e

amendments for the mesh mode do not specify par-

ticular QoS mechanisms as the trafﬁc of the mobile

Mesh Subscriber Station (MSS) along with the trafﬁcs

of all its children use the same link which has no ser-

vice or QoS parameters associated with it. Besides,

the mesh Base Station (BS) schedules the transmis-

sion and all packets originating from the mobile MSS

use the aggregate grant values regardless of their na-

ture and QoS constraints, (Kuran et al., 2006). As

all forwarded trafﬁcs use the same link, one mobile

MSS may experience starvation or act selﬁshly. Last

but not least, the MSS’s mobility adds important com-

plexity as all the affected ﬂows need to be re-routed,

re-admitted and re-scheduled while respecting their

original QoS requirements.

Note that optimal QoS provision within the IEEE

802.16e context can not be fulﬁlled without the deﬁ-

nition of optimal scheduling of space, frequency and

time resources over the air interface on a frame-by-

frame basis. Resources allocation and scheduling

need to dynamically adapt to the bursty and unfore-

seeable nature of the trafﬁc while providing a large

dynamic range of throughput to speciﬁc users based

on their demand without degrading the overall net-

work performances or causing starvation to particu-

lar users or trafﬁc ﬂows. The implemented schedul-

ing scheme should be priority-based in order to cor-

rectly distribute the available resources among the

various ﬂows depending on their QoS requirements.

It should also be simple, efﬁcient and fair with a

low computational complexity and needs to guaran-

tee the throughput and delay performance. Currently,

many research activities are conducted to propose op-

timized scheduling algorithms for both the point-to-

multi point mode and the mesh mode. Nevertheless,

designing efﬁcient scheduling methods for the mesh

mode remains a little bit harder due to the distributed

nature of the mesh mode and the constraints induced

by the multi-hop communication. Regarding optimiz-

ing the scheduling scheme, we propose to adopt the

Krichene N. and Boudriga N. (2010).

QoS SCHEDULING FOR IEEE 802.16e MESH MODE USING GENETIC ALGORITHMS.

In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 5-12

DOI: 10.5220/0002916900050012

 SciTePress

genetic algorithms concept which are relatively sim-

ple to implement and considered particularly appro-

priate for scheduling problems. To the best of our

knowledge, genetic algorithms have not been used to

solve trafﬁc related problems particular to mesh net-

works.

In this paper, we aim at optimizing the schedul-

ing of the uplink transmissions of the rtPS trafﬁc class

within the IEEE 802.16e mobile mesh context

while

handling the selﬁsh behavior of the nodes and the star-

vation issues through priority assignment. Our ap-

proach uses the Genetic Algorithms concept in order

to optimize the time spent by a ﬂow in the queue while

implementing the mobility constraints within the mu-

tation function. To the best of our knowledge, Gnetic

Algorithms have not been used before for optimizing

scheduling while respecting QoS constraints within

the wireless mobile mesh networking context.

Moreover, we dissuade mobile MSSs from act-

ing selﬁshly by augmenting the priority of their own

ﬂows as long as they forward the trafﬁcs of their chil-

dren; thus avoiding starvation. For that aim, we de-

ﬁne the priority assigner component which communi-

cates with the scheduler of the mesh BS in order to set

and update ﬂows’ priority and we propose a linear ap-

proach and an exponential approach to deduce the pri-

ority level that should be assigned to a ﬂow while up-

dating that value due to mobility. The rest of the paper

is organized as follows: ﬁrst, we overview the state of

the art regarding scheduling in the IEEE 802.16e con-

text. We then detail our proposed scheme. Finally, we

evaluate the performances of our proposition.

2 RELATED WORK

The IEEE 802.16e amendments for the mesh mode

deﬁne two modes of scheduling known as centralized

scheduling and distributed scheduling, (IEEE, 2006).

Distributed scheduling is adopted when a Mesh Client

(MC) has data to be transmitted to a neighboring MC

managed by the same mesh BS. In this case, nodes ne-

gotiate the distribution of transmission opportunities

in a pairwise fashion by using a three-way-handshake.

First, an MC wishing to change the transmission op-

portunity allocation for one of its connections should

send a request for transmission opportunities to its

neighbors using a Mesh Distributed Schedule (MSH-

DSCH) packet. One or more of the neighbor corre-

spond with a range of available transmission oppor-

tunities. The MC chooses a subset of available trans-

In fact, we ﬁnd that our approach is easily applied to

the new standard IEEE802.16j as the multihop relay mode

reduces the complexity of the mesh mode.

mission opportunities and acknowledges that it will

use them with a third MSH-DSCH packet. After the

execution of distributed scheduling, the mesh node

can transmit in the reserved timeslot without colli-

sion, (Ciao and al, 2005).

In the centralized scheduling, the mesh BS acts as

a scheduler and allocates transmission and reception

timeslots for each client station. First, a scheduling

tree rooted at the mesh BS is established. This tree

describes the routing path between each MC and the

mesh BS; its is also broadcasted to all MCs. Each

MC belongs to one tree layer and has a position num-

ber in that layer. The centralized scheduling operates

in two stages and the time period required to perform

both stages is known as the “scheduling period”. In

the ﬁrst stage, the mesh BS collects the bandwidth re-

quests from all MCs. In the second stage, the mesh

BS allocates then distributes the transmission and re-

ception schedule to all MCs within the Mesh Cen-

tralized Schedule (MSH-CSCH) messages. Note that

the data subframe description belongs to a frame after

the frame that the grant is sent,(Kuran et al., 2006).

Moreover, in centralized scheduling, the mobile MSS

needs to send one bandwidth request for each link

it has with the neighboring stations and all the re-

quests belonging to that MSS is sent within a unique

MSH-CSCH message, (IEEE, 2006). The grant sent

in the MSH-CSCH message indicates the amount of

data that a node can transmit independently with the

QoS requirements of the transmitted ﬂows.

Many centralized scheduling techniques for the

mesh WiMAX networks have been proposed. Those

may be without spatial reuse or with spatial reuse.

Spatial reuse enables the scheduler to assign the same

slot to non interfering links. The proposed techniques

also differ in whether they provide QoS guarantees

or not, consider fairness or not and suggest routing

schemes or not. For instance, authors in (Shetiya

and Sharma, 2005) propose routing and centralized

scheduling algorithms that guarantee per-ﬂow QoS

requirements to real-time and interactive data applica-

tions. More precisely, they separately schedule UDP

and TCP connections and compute the number of

slots required per ﬂow along the path and at each node

per frame while assuming an OFDM-based physical

layer. The number of slots is computed with regard to

the ﬂow characteristics such as the end-to-end packet

drop probability. Once the mesh BS assigns the com-

puted number of slots to the nodes, the nodes provide

the required slots to its different queues in a weighted

round robin manner. For TCP trafﬁcs, slot allocation

is proportionally fair to the minimum bandwidth re-

quirements of the nodes.

In (Mai et al., 2009), authors designed a special

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

bandwidth requests for UGS trafﬁc while adding the

weight of delay when scheduling different Subscriber

Stations (SS) with the same service type. More pre-

cisely, authors in (Mai et al., 2009) propose an Ex-

pedited Queue (EQ) scheduling scheme which con-

siders both per-hop BW-REQ and end-to-end route

path in order to provide absolute QoS guarantees for

UGS trafﬁc. When the sender SS of UGS trafﬁc re-

quests BW-Req, the BS assigns the slots based on

the requested slots and the number of hops within the

route path; thus reducing overhead and end-to-end de-

lays. Moreover, authors in (Mai et al., 2009) assign

to SSs within the same service type but with higher

load higher priorities. They also reduce the access de-

lay of real-time ﬂows including UGS ones by giving

higher priorities to data frames that have been wait-

ing a longer time in the queue. Nevertheless, they

do not differentiate between the SS’s own trafﬁc or

the SS’s children trafﬁc. The performance of the pro-

posed techniques is evaluated by considering the aver-

age delay and the delay variation (ms) versus the ﬂow

data rate of total input trafﬁc (Mbps).

In (Ghosh et al., 2008), authors survey multiple

centralized scheduling techniques for the mesh and

the PMP modes. Some interesting ideas may be high-

lighted such as ordering the assigned slots to reduce

jitter, ranking links based on their satisfaction with the

schedule in the previous iteration (satisfaction s = rate

achieved with the previous schedule / required band-

width of the link), scheduling ﬂows based on a prior-

ity value proportional to the node’s load and through-

put requirements or scheduling the ﬂows while re-

specting the fairness constraint or the transmission

power constraint. Moreover, authors of (Ghosh et al.,

2007) deﬁne a metric called “Schedule Efﬁciency” as

the proportion of the weighted measure of the admit-

ted ﬂows to the weighted measure of all ﬂows seek-

ing admission in order to compare the performance of

their scheduling algorithm to other scheduling tech-

niques.

Authors of (Belghith and Nuaymi, 2008) com-

pare ﬁve scheduling algorithms which are the Round

Robin (RR), the maximum Signal-to-Interference Ra-

tion (mSIR), the Weighted RR, the combination of

the Temporary Removal Scheduler (TRS) and the RR

(TRS+RR) and the combination of the Temporary Re-

moval Scheduler and the mSIR (TRS+mSIR) by con-

sidering the number of delivered data packets versus

trafﬁc load, the number of the served SSs per frame

versus trafﬁc load. They also compared their pro-

posed technique called modiﬁed maximum Signal-to-

Interference Ration (mmSIR) and the original mSIR

by considering the mean sojourn time versus the traf-

ﬁc load. Nevertheless, the considered algorithms are

mainly designed for the PMP mode of WiMAX.

Authors of (Kuran et al., 2006) propose a Service

Adaptive QoS (SAQoS) approach in which the mesh

BS assigns ﬁve virtual node IDentiﬁers (node IDs) in-

stead of one to each MSS. The virtual nodes IDs rep-

resent the ﬁve scheduling classes of the IEEE 802.16e

standard and each of these virtual nodes requests

bandwidth individually according to its requirements.

Consequently, the mesh BS will handle the received

requests independently. As in mesh mode, a separate

request must be sent to the MBS for each hop, authors

of (Kuran et al., 2006) order their mesh BS to allocate

the same amount of bandwidth to each link the traf-

ﬁc uses to reach the mesh BS and the same allocation

differentiation is valid for downlink trafﬁc to MSSs

with hop count more than one. Authors of (Kuran

et al., 2006) also propose a Fair Adaptive Base Station

Scheduler (FABS) that bases its scheduling decisions

on each MSS’s current request and the grants given

to all MSSs in the network. In (Mogre et al., 2008),

authors jointly optimize the routing, scheduling and

bandwidth savings within the IEEE 802.16 context

using network coding while reducing the computa-

tional costs.

The stated scheduling methods are highlighted

and compared to our proposed method in the table 1.

3 PROPOSED SCHEME

3.1 Problematics, Assumptions and

Goals

The IEEE 802.16e mesh mode have several charac-

teristics that render complex the QoS provision. First,

each MSS not only transmits its own trafﬁc ﬂows but

also forwards the trafﬁc ﬂows of its children. There-

fore, the ﬂows owned by the forwarding MSS may

experience starvation as the available bandwidth on

the links is shared between all forwarded trafﬁcs.

Besides, a MSS may act selﬁshly to beneﬁt from

the available bandwidth. Meanwhile, when a MSS

moves, all the ﬂows that were managed by it need

to be re-routed, re-admitted and re-scheduling while

compensating the induced processing delays in order

to meet the original QoS requirements. Particularly,

the priority assigned to the affected ﬂows needs to

be dynamically adjusted while the uplink scheduling

needs to be dynamically revised in order to optimize

the delays spent in the queues while meeting the QoS

requirements despite mobility.

Our proposed scheme adopts the idea of the ﬁve

virtual nodes ID assigned to each MSS, (Kuran et al.,

QoS SCHEDULING FOR IEEE 802.16e MESH MODE USING GENETIC ALGORITHMS

Table 1: Comparing some scheduling techniques.

Scheduling Method Scheduling Policy Priority Policy Layer of

scheduling

Implementation

Performance criteria

addressed

Mobility

Involvement

Routing and

scheduling for IEEE

802.16d mesh mode

(Shetiya and

Sharma, 2005)

Centralized scheduling

for real-time and non

real-time ﬂows

-Trafﬁc based: priority to UDP

trafﬁc over TCP trafﬁc

-Does not treat selﬁsh behavior and

starvation issues

Physical:

calculates the

number of slots

to be allocated

-Average bandwidth

provided for each

ﬂow versus required

average bandwidth

No mobility

considerations

Combined

scheduling QoS

framework for IEEE

mesh mode (Mai

et al., 2009)

Combines centralized

scheduling and distributed

scheduling

-Within the same service type, the

SS with higher load has higher

priority

-Trafﬁc based: Highest priority for

UGS trafﬁc

-Higher priority is given to real-time

data-frames that waited a longer

time in the queue

Cross layer

approach

involving the

MAC layer and

the IP layer

-Average delay and

delay variation (ms)

versus the ﬂow data

rate of total input

trafﬁc (Mbps)

No mobility

considerations

Enhancement of of

the maximum

Signal-to-

Interference Ratio

(mSIR)(Belghith and

Nuaymi, 2008)

Schedules rtPS ﬂows for

the Point to Multi-Point

(PMP) mode

If it does not serve an SS having

unicast request opportunities, it

gives priority to other SSs having

higher SIR

Physical:

involving the

quality of the

link

Mean sojourn time

versus trafﬁc load

No mobility

considerations

Service Adaptive

QoS approach

(Kuran et al., 2006)

-Assigns 5 virtual node

identiﬁers for each QoS

class

- Proposes a BS scheduler

for the centralized

scheduling of the mesh

mode

Not addressed Physical: grants

bandwidth to

links

- Service delays of 5

ﬂows of involved

SSs

Partially

considered

Our method Minimizes the sojourn

time in the queue using

genetic algorithms for

rtPS trafﬁc

- Trafﬁc based and history based:

Own rtPS ﬂows’ priority depends on

the amount of children’s forwarded

trafﬁc

-Addresses selﬁshness and

starvation

- The ﬂow’s priority is incremented

when the managing node leaves the

route due to mobility

MAC layer:

adds a new set

of functions

-Number of rescued

ﬂows in each round

-Number of genomes

that persist until the

last round

-Delay is reduced

since waiting time is

optimized

Flows affected

by mobility are

rescued and

their QoS

constraints are

revised in order

to meet the

original QoS

requirements

2006), and considers the rtPS trafﬁc class. Each rtPS

ﬂow is characterized by its minimum reserved trafﬁc

rate (bits/s), its maximum sustained trafﬁc rate called

also peak rate (bits/s), its maximum latency (s), and

its priority, (IEEE, 2006). According to the mobile

WiMAX speciﬁcations, rtPS ﬂows are generated by

the third class of applications that includes streaming

media, (Forum, 2006). Guidelines specify bandwidth

values ranging between 5 kbits/s and 2 Mbits/s while

latency values are not speciﬁed, (Forum, 2006). Pri-

ority ranges between 0 to 7 where higher numbers in-

dicate higher priority, (IEEE, 2006). rtPS ﬂows that

we consider in our scheme have a size ranging from

20 Mbits to 340 Mbits, a minimum reserved trafﬁc

rate raging between 500 kbits/s and 2 Mbits/s, a maxi-

mum sustained trafﬁc rate ranging from 500 kbits/s to

2 Mbits/s and a maximum latency value of 680 sec-

onds. We also assume that a route has always been

established to the mesh BS so that intermediate MSSs

are known to the source and that every intermediate

MSS issues its own request in order to forward the

correspondent ﬂow. We also assume that the origi-

nal QoS values are set by the source and are integrity

protected. Besides, each intermediate MSS should

update the maximum latency value (i.e., the updated

value is speciﬁed in a different ﬁeld) when issuing its

request while taking into consideration the time spent

by the ﬂow before reaching that MSS; the updated

value should always be smaller than the original one.

Our proposed scheme intends to schedule the up-

link transmission of rtPS trafﬁc while a priority will

be set for every ﬂow in order to encourage the MSS

forwarding the trafﬁc of their children. That prior-

ity value is managed by a new component that we

call “Priority Assigner” which communicates with

the scheduler of the mesh BS. We intend to optimize

the time spent by a ﬂow in the queue while respecting

the ﬂow’s QoS requirements. We adopt the Genetic

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

Algorithms concept in order to solve the scheduling

problem.

3.2 Priority Assignment

We deﬁne the Priority Assigner component at the

mesh BS level as depicted by Figure 1. The sending

MSS generates a rtPS transmission request while

indicating the minimum reserved trafﬁc rate, the max-

imum sustained trafﬁc rate and the maximum latency

values without specifying a priority value. The Prior-

ity Assigner component will get the request and up-

date it by setting the priority value depending on the

history of the issuer. More precisely, the Priority As-

signer component maintains a table with these entries

(NID, T

, Amounto f ForwardedTra f f ic, Priority)

where NID refers to the issuer’s node Identiﬁer,

refers to the instant when the issuer has entered

under the coverage of the managing mesh BS and

Amounto f ForwardedTra f f ic refers to the amount

of the trafﬁc that the issuer has forwarded so far

for its children. The Priority Assigner component

assigns a priority value to the issuing MSS which is

computed by the following

amounto f ForwardedTra f f ic

t − T

(1)

then adjusted to be in the interval [0,7] depending on

8 thresholds to be deﬁned in order to cope with the

standard amendments. The priority value of that MSS

is then communicated to the scheduler so that it can

update the priority of the rtPS ﬂow lastly generated by

that MSS then schedule it with regard to the assigned

value.

Figure 1: The priority assigner interactions.

Note that when the mesh BS scheduler schedules

the transmission of a ﬂow (identiﬁed by a CID) by a

MSS (identiﬁed by a NID) and that ﬂow is not the own

ﬂow of the forwarding MSS, the scheduler communi-

cates with the Priority Assigner component to update

the Amounto f ForwardedTra f f ic value. When one

ﬂow is affected by the mobility of an intermediate

node on the route, the priority of that ﬂow is incre-

mented independently of its issuer.

Mobile WiMAX parameters are stated in (Forum,

2006). For instance, the mobile nodes can have a

speed reaching 120 km/h. Moreover, the distance be-

tween two BSs is about 2.8 km. That means that a mo-

bile WiMAX node moving at 120 km/h needs 84 sec-

onds to cross the distance between two BSs. In order

to adjust the priority value to be in the interval [0,7],

we propose a linear approach and an exponential ap-

proach. Regarding the linear approach, we assume

that a mesh node may stay under the coverage of the

mesh BS for a certain period of time during which its

should transmit a certain amount of trafﬁc for its chil-

dren in order to increase its own trafﬁc’s priority. The

time and trafﬁc size are augmented linearly by adding

chosen values and the obtained values are mapped to

8 levels of priority. The numerical scenario that we

propose assumes that after 85 seconds, the mesh node

should forward 85/2 Mbits for its children in order

to increase the priority of its own trafﬁc by one. The

assigned priority values are given by the table 2. For

illustration purpose, when a node stays for 85 seconds

under the coverage of the same mesh BS and forwards

less than 42.5 Mega bits for all its children, indepen-

dently of their number, all the ﬂows issued by that

node will have 0 as priority. When that node stays

100 seconds and forwards less than 85 Mega bits for

its children, the priority remains 0 but when the node

stays 100 seconds and forwards between from 42.5 to

85 Mega bits, the priority is set to 1.

Table 2: Priority assignment in case of a linear approach.

Amounto f Forwarded

Tra f f ic(Mega bits)

thet − T

(seconds)

Assigned

Priority

340 680 7

297.5 595 6

255 510 5

212.5 425 4

170 340 3

127.5 255 2

85 170 1

42.5 85 0

Regarding the exponential approach, we assume

that a mobile mesh node which transmits less than

42.5 Mega bits for its children during 85 seconds im-

plies that its own ﬂows will be assigned 0 as prior-

ity. These forwarded trafﬁc and time values will be

doubled in order to increase the assigned priority as

depicted by the table 3.

We also deﬁne the “Integrity Veriﬁer” component

shown by the Figure 1 which role is to verify whether

a request is issued by its owner or by an intermediate

MSS on the route. The CID of a request that is not

QoS SCHEDULING FOR IEEE 802.16e MESH MODE USING GENETIC ALGORITHMS

Table 3: Priority assignment in case of an exponential ap-

proach.

Amounto f Forwarded

Tra f f ic(Mega bits)

thet − T

(seconds)

Assigned

Priority

5440 10880 7

2720 5440 6

1360 2720 5

680 1360 4

340 680 3

170 340 2

85 170 1

42.5 85 0

issued by the owner of the ﬂow along with the node

ID of the forwarding node are communicated to the

scheduler so that the latter can update the correspon-

dent Amounto f ForwardedTra f f ic value.

3.3 Genome Modeling, Fitness,

Mutation and Cross-over

We model a genome as a list of ﬂows. Each ﬂow is

characterized by its size, its minimum reserved trafﬁc

rate, its maximum sustained trafﬁc rate, its maximum

latency, its priority value and the delay that the ﬂow

spends in the queue before being scheduled for trans-

mission d

CID,NID

. We assume that we have N ﬂows

to be scheduled and we use the Genetic Algorithms

concept to minimize the delay that a ﬂow spends in

the queue before being scheduled for transmission.

We calculate the ﬁtness of a genome by

∑

CID,NID

(2)

and we deﬁne the optimal genome among M ones as

the one having

Min

(

∑

i=1

CID,NID

)

and we intend to optimize

CID,NID

= T

− T

(3)

where T

is the instant at which the request is served

and T

is the instant at which the request was issued

subject to constraints:

maximumSustainedRate < available BW

and

MinimumReservedRate < available BW

We always verify that

CID,NID

< ML − λ

where λ is the maximum propagation delay needed

for the node to receive the schedule then transmit the

ﬂow till the mesh BS.

When a MSS leaves the network, its own ﬂows

are no longer scheduled for uplink while the ﬂows of

its children should be rescued by neighboring MSSs.

Rescuing a ﬂow by a neighboring MSS induces a de-

lay that should be taken into consideration when the

rescuing MSS issues an uplink transmission request

regarding that ﬂow. Therefore, the rescuing MSS

should decrement the ML value and increment the

priority of the rescued ﬂow. The mutation operator

reﬂects the mobility of MSSs. More precisely, the

mobility of a MC implies

sMSS

= ∞,T

s j

= T

s j

− 10 (4)

(where j is the index of a child of the quitting MSS)

and

= p

+ 1 (5)

(where p is the priority and j is the index of a child

of the quitting MSS). The number of chromosomes of

an individual is given by the Formula

Total BW o f meshBS

minimal BW required bya f low

(6)

4 PERFORMANCE EVALUATION

We implemented our scheduling genetic algorithm

by generating a 500 genomes population where each

genome is composed by 250 ﬂows and a ﬁtness. Ev-

ery node has 5 parent nodes and and each parent node

ID is comprised between 1 and 25. In every round

of the algorithm, the cross-over procedure is applied

followed by the mutation procedure which randomly

affects 100 genomes. The mutation consists in elimi-

nating 100 nodes of the mutant genomes; that is 100

nodes quit the network due to mobility. The 100

eliminated nodes are randomly chosen and their ﬂows

have a T

value equal to 999999 which is the synonym

for an inﬁnite value. After that, the resulting new pop-

ulation is sorted in order to keep the best 500 genomes

having the best 500 ﬁtness values. The pre-described

round is executed 100 times and for each round we

keep the ﬁtness of the best genome in population. It

is worth noticing that an arbitrary number of simula-

tions may be considered as we need just to launch the

genetic algorithm. However, we opted for consider-

ing three simulations only as we have noticed that the

difference between the obtained curves is tiny and in

order of 1.25%. This means that even if we launch

a greater number of executions, the obtained results

will be similar and that the conclusions that we can

have in the light of three simulations are relevant.

The results obtained after executing the genetic al-

gorithm while adopting the numerical scenario stated

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

in 3.2 are plotted in the Figures 2 and 3 below. We

may notice that the genetic algorithm converges after

maximum 10 executions when assigning the priority

of the ﬂows using the linear approach as depicted by

Figure 2. When adopting the exponential approach,

the algorithm converges also after maximum 10 exe-

cutions as depicted by Figure 3. Note that the chosen

numerical scenario ﬁxing the maximum latency, the

minimum reserved trafﬁc rate and the size of the for-

warded trafﬁc values inﬂuences the number of rounds

needed for convergence. However, that number has

not exceeded 20 rounds for more than 10 different nu-

merical scenarios that we probed. Note also that we

have tried other numerical scenarios and we obtained

curves having nearly the same shape with a rapid con-

vergence. We think that the rapid convergence of the

genetic algorithm is mainly due to the fact that we

perform multiple veriﬁcations on the generated ﬂows

and we state conditions on the values characterizing

the scheduled ﬂows. We also conclude that the ﬁtness

values of the genomes in both Figures 2 and 3 are very

close and that they converge to nearly the same value.

Figure 2: Execution of the genetic algorithm when adopting

the linear approach.

Figure 3: Execution of the genetic algorithm when adopting

the exponential approach.

In order to evaluate the performances of our

scheduling technique, it is interesting to evaluate the

number of rescued ﬂows in each round, that is those

affected by mobility, and to consider the number of

initial genomes that persist until the last round. We

notice that the number of rescued ﬂows when adopt-

ing the linear approach is very reduced in the ﬁrst 10

rounds as depicted by Figure 4, then it raises exponen-

tially until the 23

round, after that it converges after

the 45

round. This behavior persists when adopt-

ing the exponential approach as depicted by Figure 5.

However, the number of rescued ﬂows stabilizes after

the 38

round in this case. We notice that adopting ei-

ther the linear or the exponential approach for the pri-

ority assignment does not seriously affect the number

of rescued ﬂows; nevertheless, the convergence is ob-

tained rapidly with the exponential approach. More-

over, we conclude that after a particular x number

of rounds, which is also equivalent to executing the

mutation function on 100 ∗ x genomes of the retained

population, it becomes difﬁcult to rescue more ﬂows

while respecting their QoS constraints. This conclu-

sion is relevant even after only three simulations for

the same reasons stated earlier.

Figure 4: Number of rescued ﬂows in each round when

adopting the linear approach.

Regarding the number of persistent genomes, we

notice that after 6 to 8 rounds, it becomes null whether

we adopt the linear approach or the exponential ap-

proach as depicted by Figures 6 and 7. This means

that the genetic algorithm rapidly produces better

genomes and that it completely renews the population

leading to the creation of a better scheduling scheme

that optimizes the sojourn time of the rtPS ﬂows. The

obtained results regarding the number of persistent

genomes conﬁrm the rapid convergence of the algo-

rithm.

QoS SCHEDULING FOR IEEE 802.16e MESH MODE USING GENETIC ALGORITHMS

Figure 5: Number of rescued ﬂows in each round when

adopting the exponential approach.

Figure 6: Number of persistent genomes in each round

when adopting the linear approach.

5 CONCLUSIONS

In this paper, we addressed the scheduling problem of

the rtPS ﬂows within the IEEE 802.16e mesh context

while minimizing the sojourn time of such ﬂows in

the queues; thus leading to a reduction in the transmis-

sion delays and to a better QoS provision. To that aim,

Figure 7: Number of persistent genomes in each round

when adopting the exponential approach.

we built an ad hoc genetic algorithm and we showed,

through the simulations, that our genetic algorithm

converges rapidly. In other words, we rapidly reach

an optimized scheduling scheme that minimizes the

transmission delays of rtPS ﬂows. Moreover, we pro-

posed a novel scheme for assigning priorities to the

transmitted ﬂows then dynamically updating such val-

ues in case of mobility in order to encourage the mesh

nodes forwarding the trafﬁc of their children while fa-

voring ﬂows that need to be rescued when the manag-

ing parent node quits the network. That scheme may

adopt a linear approach or a linear approach which

both lead to comparable performances. The results

provided in this paper are encouraging; therefore we

plan to further analyze the performance in the network

by adapting additional metrics and we also plan to in-

vestigate the effects of mobility independently from

the effects of the prioritization scheme.

REFERENCES

Belghith, A. and Nuaymi, L. (2008). Comparison of wimax

scheduling algorithms and proposal for the rtps qos

class. Technical report, TELECOM Bretagne.

Ciao, M. and al (2005). Modeling and performance analysis

of the distributed scheduler in ieee 802.16 mesh mode.

In MobiHoc’05, Urbana-Champaign, Illinois, USA.

Forum, W. (2006). Mobile wimax- part i: A technical

overview and performance evaluation. Technical re-

port, Wimax Forum.

Ghosh, D., Gupta, A., and Mohapatra, P. (2007). Admission

control and interference-aware scheduling in multi-

hop wimax networks. In IEEE MASS.

Ghosh, D., Gupta, A., and Mohapatra, P. (2008). Schedul-

ing in multihop wimax networks. Mobile Computing

and Communication Review, 12(2).

IEEE (2006). Ieee standard for local and metropolitan area

networks part 16: Air interface for ﬁxed and mo-

bile broadband wireless access systems amendment 2:

Physical and medium access control layers for com-

bined ﬁxed and mobile operation in licensed bands

and corrigenda 1. IEEE.

Kuran, M. S., Y1lmaz, H. B., Alagz, F., and Tucu, T.

(2006). Quality of service in mesh mode ieee 802.16

networks. In International Conference on Software in

Telecommunications and Computer Networks, pages

107–111, Split-Dubrovnik, Croatia.

Mai, Y.-T., C-C.Yang, and Wen, C.-J. (2009). A new cut-

through mechanism in ieee 802.16 networks. In World

Academy of Science, Engineering and Technology.

Mogre, S., D’Heureuse, P., Hollick, N., and Steinmetz, M.

(2008). Core: Centrally optimized routing extensions

for the ieee 802.16 mesh mode. In 33rd IEEE Confer-

ence on Local Computer Networks.

Shetiya, H. and Sharma, V. (2005). Algorithms for rout-

ing and centralized scheduling to provide qos in ieee

802.16 mesh networks. In WMuNeP.

WINSYS 2010 - International Conference on Wireless Information Networks and Systems