You are on page 1of 5

Low Energy Priority Oriented Adaptive Polling

for Wireless Networks


T. D. Lagkas, G. I. Papadimitriou, Senior Member, IEEE, P. Nicopolitidis, and A. S. Pomportsis
Department of Informatics, Aristotle University
Box 888, 54124 Thessaloniki, Greece

Abstract—The widespread use of the wireless local area is a centralized access protocol, however, no bandwidth
networks (WLANs) and the increasing use of multimedia reservation is required. It efficiently supports simultane-
network applications have increased the need for integrating ous real-time and background traffic, by taking into ac-
time bounded and best effort traffic in the WLAN environ- count traffic priorities and the current status of the sta-
ment, while saving battery power for the mobile devices. The
IEEE 802.11e workgroup is standardizing a new access
tions, while providing low energy consumption. It should
mechanism called Enhanced Distributed Channel Access be noticed that despite the fact that EDCA is a distributed
(EDCA). EDCA seems capable of differentiating the traffic, scheme, most network scenarios that consider real-time
however, it exhibits great overhead that limits the actually traffic assume infrastructure topology with the use of an
available bandwidth and degrades the overall performance. Access Point (AP) for packet relay and interconnection to
This work proposes an alternative protocol in place of the backbone network. LEPOAP tries to exploit this com-
EDCA. The Low Energy Priority Oriented Adaptive Polling mon topology by using the AP for access control. This
(LEPOAP) is a collision free MAC protocol, implements a
paper assumes that stations are able to communicate di-
power saving mechanism, employs traffic priorities, and is
able to provide QoS for all types of multimedia applications, rectly when in range, however the model where the AP
while efficiently supporting background data traffic. acts as a packet forwarder could be also used. According
LEPOAP compared to EDCA, provides higher utilization, to [1], the IEEE 802.11e access model also provides a
consumes lower energy, distributes network resources to the Direct Link Protocol (DLP) as an extra feature.
stations adapting to their real needs, and generally exhibits This paper is organized as follows. Section 2 presents
superior performance. the IEEE 802.11e EDCA function. In Section 3, LEPOAP
is analyzed, focusing on the polling scheme, the priority
Index Terms—EDCA, 802.11e, LEPOAP, WLAN MAC
model, the station choice algorithm, and the power saving
mechanism. Section 4 presents our simulator, the network
scenario, and the simulation results, which prove the effi-
I. INTRODUCTION
ciency of LEPOAP by comparing it with EDCA. Section

I N the past few years, the wireless LANs have become a


significant part of the local area network market. At the
same time, multimedia network applications such as voice
5 concludes the paper.

over IP, video conference, and video on demand are in- II. IEEE 802.11E EDCA CONTROL
creasingly employed by network users. Furthermore, since
The DCF MAC protocol employed by the legacy IEEE
mobility has become essential nowadays, it is important to
802.11 access scheme does not support QoS. However,
ensure low energy consumption for the mobile devices in
some modifications that enhance partial QoS support have
order to extend their battery operation time. Thus, modern
been proposed. The need for QoS in modern WLANs has
WLANs need to efficiently integrate voice, audio and
led to the formation of the IEEE 802.11e workgroup [1].
video with background data traffic, while saving power.
The mandatory access scheme for the IEEE 802.11e chan-
The Distributed Coordination Function (DCF), which is
nel access mechanism is the EDCA protocol. EDCA is the
used by the legacy IEEE 802.11 wireless networks, has
QoS enhanced version of the DCF protocol. When a sta-
been enhanced by the IEEE 802.11e [1] workgroup with
tion needs to transmit a packet and the channel is busy, it
QoS support, forming EDCA, which is the essential part
waits until the medium becomes idle and then defers for
of the IEEE 802.11e MAC protocol. However, it causes
an extra time interval, called Arbitrary Distributed Inter-
high network overhead, which degrades the system’s per-
frame Space (AIFS). If the channel stays idle for the AIFS
formance, thus, efficiently serving multiple sources of
interval, the station then starts the backoff process by se-
different types of traffic becomes very difficult. Specifi-
lecting a random number of slots from a contention win-
cally, in a wireless environment, where bandwidth is
dow (CW). The minimum size of CW is CWmin, while
scarce and links are unreliable, we need high channel
the maximum is CWmax. The differentiation of the traffic
utilization and efficient traffic differentiation in order to
is based on the use of four buffers. Each buffer corre-
provide QoS. Furthermore, EDCA adopts no power sav-
sponds to an access category (AC) which carries packets
ing mechanism for the mobile devices.
of specific user priorities (UPs). The highest access cate-
In literature, there are various MAC protocols proposed
gories are more probable to gain access, since they are
for different kinds of networks [2]–[8]. This work pro-
assigned low CWmin, CWmax and AIFS values. An addi-
poses the Low Energy Priority Oriented Adaptive Polling
tional RTS/CTS (Request To Send/Clear To Send)
which can be built into the scheme defined by 802.11e. It
mechanism is defined to solve the hidden terminal prob-
lem. This scheme, however, increases the overhead and a) t+tPOLL+tNO_DATA
does not provide a completely collision-free medium. t +2tPROP DELAY
AP
The above model provides only minimal QoS. The NO_DATA
backoff procedure leads to waste of bandwidth, however,
POLL POLL
it is necessary in order to avert collisions. Furthermore, (poll to a possibly
A
the “hidden terminal” problem leads to collisions despite t+tPOLL+tPROP DELAY different station)
of the backoff mechanism. The use of the RTS/CTS hand-
b)
shake limits this problem, however, it increases the net- t+tPOLL+tSTATUS t+tPOLL+tDATA
t +2tPROP DELAY +2tSTATUS+4tPROP DELAY
work overhead. When a wireless transceiver neither AP
transmits nor receives, it stays in idle mode, which is still STATUS
(ack)
high power consuming, as it will be discussed later. Con- POLL
clusively, EDCA definitely enhances DCF with QoS sup- A STATUS
(ack–nack)
port, however, it is shown that it can actually serve only
DATA POLL
limited traffic of low QoS requirements. For these rea-
B
sons, we alternatively propose the LEPOAP protocol t+tPOLL+tSTATUS (poll to a possibly
+tDATA+3tPROP DELAY different station)
which greatly reduces the network overhead and opti-
mizes the priority model, providing significantly stricter
c) t+tPOLL+tMAX_DATA
QoS and generally higher performance, while limiting t +2tSTATUS+4tPROP DELAY
energy consumption, as it is shown in the next sections. AP

III. LEPOAP PROTOCOL POLL


A (poll to a possibly
t+tPOLL+tPROP
A. Polling Scheme DELAY different station)

In LEPOAP, the AP polls the stations in order to give Fig. 1. The polling scheme of the LEPOAP protocol
them the opportunity to transmit. The polling scheme
eliminates the collisions and causes low overhead. The certain that it will not collide with a possible on going
protocol uses the POLL, NO_DATA, and STATUS con- transmission. When the POLL packet is received success-
trol packets, with transmission duration tPOLL, tNO_DATA, and fully by the polled station, but then the AP fails to receive
tSTATUS, respectively. A STATUS packet is marked as any feedback, it waits for the maximum polling cycle
ACK or NACK, accordingly. The duration of a DATA similarly to the previous case. The duration of this polling
packet is tDATA and the propagation delay is tPROP_DELAY. cycle is t POLL + t MAX_DATA + 2t STATUS + 4t PROP_DELAY ,
The polling events are depicted in Fig. 1.
- The AP polls an inactive station (Fig. 1a): The AP where tMAX_DATA is the duration of the largest allowed
sends POLL to the station at time t and waits for feed- DATA packet. At the end of this cycle, it is certain that
back. The station responds with a NO_DATA packet, the medium is idle in any case. When such a failure oc-
which is received by the AP at curs, the AP lowers the probability to choose this station
in the new polling procedure assuming a bad link between
t + t POLL + t NO_DATA + 2t PROP_DELAY . Then, the latter
them. However, it is most likely that the AP eventually
initiates a new polling procedure. receives some feedback either from the polled or the des-
- The AP polls an active station (Fig. 1b): The AP sends tination station.
POLL to the mobile station at time t and waits for feed- The polling scheme provides efficient network feed-
back. The station replies with a STATUS packet marked back and low system overhead. The control packets aim at
as ACK, which carries the destination address and the size keeping the stations informed of the network’s status and
of the following DATA packet. Then, the polled station minimizing the idle intervals. The AP needs to monitor
starts transmitting the DATA packet directly to the desti- the transmissions, so that it can proceed to the next poll
nation. Upon successful reception, the destination broad- right after the completion of a communication. Thus, it
casts a STATUS packet marked as ACK. Otherwise, if the has to be aware of the actual duration of the specific poll-
reception fails but the station has realized that the specific ing cycle. In order to gain this knowledge, the AP just has
packet is destined to it, it responds with a STATUS packet to successfully detect the NO_DATA packet or the
marked as NACK. The transmission of a NACK is not STATUS ACK packet, which contains the duration of the
wasted time, since either way the stations had to wait for a following data transmission, or the DATA packet from the
possible ACK. There can be a new poll at polled station or the STATUS ACK-NACK packet from
t + t POLL + t DATA + 2t STATUS + 4t PROP_DELAY . It should the destination station. In all these cases, the AP is aware
of the polling cycle duration and can proceed to the next
be noticed that we consider variable DATA packet size, poll without waiting the maximum polling cycle duration.
thus, tDATA is not static. Actually, when POLL is successfully received, then it is
- The communication fails (Fig. 1c): In case the station most likely that the AP will obtain the necessary feed-
does not successfully receive the POLL packet, the poll- back, since the link is probably in good state.
ing fails. The AP has to wait for the maximum polling Furthermore, our purpose is to poll stations that are ac-
cycle before proceeding to a new poll, since it has to be tually active in a way that QoS is provided. Thus, the AP
needs to be aware of the stations’ status in order to avoid i=0
wasting time at polling inactive stations, favor the high
priority traffic, and provide fairness. Moreover, the con- All buffers i<4
are empty No
trol packets employed in the polling scheme are also ex- No
ploited by the power saving mechanism, which is exam- Yes Yes
ined later. Abort Buffer i Yes
i++
is empty
B. Priority Model
No
The priorities concept adopted by LEPOAP is the same Choose buffer accord-
PPR[i]: normalized buffer priority
ing to the normalized
with the one employed by 802.11e in order to retain com- final probabilities
patibility. However, LEPOAP uses a new method of and send its earliest PB[i]: normalized number of
choosing which packet to send. This method takes into generated packet buffered packets
account the access category of each buffer, so that high
P[i] = WPR×PPR[i] + WB×PB[i]
priority traffic is favored. Furthermore, in our effort to non-normalized final probabil-
provide low delays, low packet drops due to buffer over- ity to choose buffer i
flow and fairness among the ACs, we consider the number
of packets contained in each buffer before choosing a
packet. Specifically, heavy loaded buffers have higher Fig. 2. Flow chart of the packet choice mechanism
probabilities of transmitting. Lastly, the earlier generated
packets are transmitted first. implement high performance polling.
The packet choice mechanism is depicted in Fig. 2. Ini- C. Station Choice Mechanism
tially, we examine if there are any buffered packets, oth-
The algorithm that returns the station to be polled
erwise, the polled station replies with a NO_DATA mes-
mainly takes into account the stations’ priority score. Spe-
sage. Then, each one of the four buffers is examined in
cifically, it is most probable to poll a station with high
order to calculate its normalized priority (PPR) and nor-
priority traffic and large number of buffered packets, ac-
malized number of buffered packets (PB). Assuming that
cording to the definition of the priority score presented in
the priority of buffer i is p[i] = i + 1 , so that it is not the previous subsection. The second factor affecting the


3
null for AC[0], then: PPR [i] = p[i] k =0
p[k ] . Also, AP’s decision on the polled station is the time elapsed
since the last poll of each station (τ). Specifically, in order
if b[i] is the number of packets in buffer i, then: to provide fairness and avoid the total exclusion of sta-


3 tions that appear to be inactive for quite long, the algo-
PB [i] = b[i] k =0
b[k ] .
rithm favors to some degree the stations that have not
Furthermore, the buffer priority and the number of been polled for a long time. Furthermore, the AP, which
buffered packets should contribute differently to the final also participates in the channel contention, is assigned a
buffer choice probability (P). Thus, we use the weights higher probability of getting access, since it usually plays
WPR (default value 6) and WB (default value 2) for PPR and a central role in the network communications. Lastly, it
PB, respectively. The non-normalized final choice prob- should be noticed that the AP halves a mobile station’s
ability for buffer i is: priority score, when it receives no feedback after polling
P[i] = WPR × PPR [i ] + WB × PB [i ] . The normalized it, assuming that the link between them is in bad state.
In Fig. 3, the flow chart depicts the operation of the

3
choice probability is: P[i] k =0
P[k ] . Finally, the ear- proposed algorithm that returns the station to be polled.
Initially, we check if the AP has any buffered packets. If
liest generated packet in the chosen buffer is transmitted.
not, then it is not included in the station choice procedure.
The AP has to be well informed of the stations’ buffers
Then, the priority score of each considered station j is
status, in order to efficiently decide which one to poll.

M −1
However, this feedback should not cause excessive over- normalized: PP [ j ] = PS [ j ] l =0
PS [l ] , where M is
head. Therefore, we exploit the use of the ACK and
the number of stations considered by the algorithm. The
NACK messages in the polling scheme. Specifically, the
time elapsed since the station’s last polling is also normal-
STATUS packet apart from acknowledging receptions, it

M −1
also carries its source’s priority score, which is an indica- ized: PT [ j ] = τ [ j ] l =0
τ [l ] . The non-normalized
tion of the status of the station’s buffered traffic. The pri- final probability of polling
station j is:
ority score depends on the priority of each buffer and the
number of packets it contains. For station j, the priority
PPOLL [j] = WPR × PP [ j ] + WT × PT [ j ] , where WT
(default value 1) is the weight of the contribution of the P

3
score is: PS [j] = p[k ] × b[k ] . So, every time a factor. If the examined station j is the AP, then its non-T
k =0
STATUS packet is broadcasted, the AP examines it in normalized final probability of getting channel access is
order to update the stored priority score of the transmit- multiplied by the factor WAP (default value 10):
ting station. This way, our model tries to provide efficient PPOLL [j] = W AP (WPR × PP [ j ] + WT × PT [ j ]) .
feedback for the AP with minimum network overhead.
Lastly, the AP decides which station will be given access
The gathered priority scores are then used by the AP to
according to each station’s normalized polling probability,
which is for station j equal to the quantity: The scenario considers one bidirectional voice commu-


M −1 nication and one bidirectional video communication be-
PPOLL [ j ] l =0
PPOLL [l ] .
tween the AP and each mobile station. Furthermore, there
is a bidirectional TCP flow between any two adjacent
D. Power Saving
mobile stations. The simulation duration is 60 sec, every
The limited battery power of the mobile devices makes communication lasts for 30 sec, and a new traffic flow is
energy conservation essential. According to the power added every second. We simulated 14 WLAN topologies,
saving mechanism adopted by LEPOAP, a station transits starting with 2 mobile stations and finally reaching 28
to a low power “doze” mode, when it is certain that it will mobile stations with a step of 2.
not be given permission to transmit nor any data will be The characteristics of traffic derived by the analysis of
sent to it. real traces. Regarding the background traffic we use TCP
Specifically, any station which successfully detects the flows with features typical for file transmissions (VBR,
STATUS packet broadcasted by a polled station (Fig. 1b) ~1200 kbps, no delay bound considered, UP: 1). The
becomes aware of the duration and the destination of the voice communication is based on the G.726 codec (CBR,
forthcoming data transmission. Thus, if the given station 32 kbps, delay bound: 75 ms, UP: 6). The new H.264 co-
is not the destination, it transits to doze mode in order to dec is used for live video transmission (CBR, 600 kbps,
save power. It should be noticed that this power saving delay bound: 200 ms, UP: 5).
scheme does not degrade the network performance at any We evaluated the capability of LEPOAP and EDCA of
degree, since stations stay in doze mode only during defi- handling voice, video, and background traffic simultane-
nitely inactive intervals. ously. Most voice and video packet drops are caused by
lifetime expiration, while TCP drops are due to buffer
overflow.
IV. SIMULATION RESULTS As far as voice traffic is concerned, in Fig. 4, it can be
We developed a simulation environment in C++ to seen that LEPOAP exhibits lower packet delays and loss
compare LEPOAP against EDCA. The considered physi- rates when the throughput is lower than 550 kbps, while
cal layer protocol is the IEEE 802.11g. The IEEE 802.11e EDCA performs better for higher values of the voice
super-frame is employed by our scheme. The condition of throughput. However, both protocols achieve similar
any wireless link was modeled using a finite-state ma- maximum throughput. Despite the fact that LEPOAP
chine with three states (Good, Bad, Hidden). The statisti- eventually causes higher packet delays, it keeps them be-
cal analysis is based on the “sequential simulation” low 18 ms which is very satisfactory for voice communi-
method. cations. The conclusion is that both schemes are able to
provide QoS in voice transmissions. This behavior is due
The AP has collected the to the fact that LEPOAP conserves recourses to serve
stations’ priority scores video and TCP traffic as well, while EDCA favors voice
packets to such a degree that seems unable to simultane-
M=N,j=0 ously serve video and TCP flows. Fig. 5 shows that
LEPOAP performs significantly better than EDCA when
dealing with live video. Actually, EDCA suffers from so
The AP has No
M=N-1
many packet losses that it cannot really support live video
buffered
packets when the video throughput is over 3 Mbps. In Fig. 6, it is
shown that LEPOAP can efficiently support bandwidth
Yes demanding background transmissions the same time it
Choose a station provides QoS for voice and video. The total power con-
to poll according to
the normalized j<M sumption of all stations is depicted in Fig. 7. It is shown
No that LEPOAP achieves significant power saving, espe-
final probabilities Yes
PP[j]: normalized cially for high throughput values. The power values are
priority score 1.65 W, 1.4 W, 1.15 W, and 0.045 W in transmit, receive,
idle, and doze modes, respectively [9].
PT[j]: normalized time of last poll
V. CONCLUSION
PPOLL[j] = WPR×PP[j] + WT×PT[j]
non-normalized final probability LEPOAP is a more deterministic protocol than EDCA.
to poll station j The polling model eliminates the “dead” idle periods,
while providing minimum overhead with optimized feed-
No back and zero collisions. Furthermore, it adapts to the
This is j++
the AP
traffic requirements and supports QoS, by considering not
only the traffic priorities, but also the time elapsed since
Yes the last poll and the amount of buffered bits. It also pro-
PPOLL[j] = WAP× PPOLL[j] vides fairness by averting medium domination. LEPOAP
efficiently integrates background and real-time traffic,
Fig. 3. Overview of the station choice mechanism, where N is the
total number of stations including the AP while it saves power.
40

Total Power (Watts)


V oic e LEPOAP
30
EDCA
20
Average Packet Delay (ms)

10
15 0
0 0.2 0.4 0.6 0.8 1
10 LEPOAP
Total Load (ratio) EDCA

5 Fig. 7. Total power consumption of the mobile devices


versus total network load

0.08
0.06 800 REFERENCES
0.04 600
0.02 400 [1] IEEE 802.11e WG, “Draft Amendment to Standard for Information
Average Packet Loss Rate 0 200 Voice Throughput (kbps) Technology--Telecommunications and Information Exchange Be-
tween Systems-LAN/MAN Specific Requirements-- Part 11 Wire-
Fig. 4. Voice Traffic: Average packet delay and packet loss less Medium Access Control and Physical Layer specifications:
rate versus voice throughput in LEPOAP and EDCA Amendment 7: MAC Quality of Service (QoS) Enhancements,”
D13 2005.
[2] T. D. Lagkas, G. I. Papadimitriou, P. Nicopolitidis, and A. S. Pom-
portsis “POAC-QG: Priority Oriented Adaptive Control with QoS
Guarantee for wireless LANs,” Proceedings of IEEE EUROCON,
Video LEPOAP p. p. 1858-1861, Serbia & Montenegro, Belgrade, Nov. 22-24,
EDCA 2005.
[3] P. Nicopolitidis, M. S. Obaidat, G. I. Papadimitriou and A. S. Pom-
portsis, “Wireless Networks”, Wiley, January 2003.
Average Packet Delay (ms)

[4] T. D. Lagkas, G. I. Papadimitriou, and A. S. Pomportsis, “QAP: A


120
QoS supportive Adaptive Polling Protocol for Wireless LANs,”
100 Computer Communications, Elsevier, (in press), 2005.
80 [5] P. Nicopolitidis, G. I. Papadimitriou, A. S. Pomportsis, “Learning-
Automata-Based Polling Protocols for Wireless LANs”, IEEE
60
Transactions on Communications, vol.51, no.3, pp. 453-463, 2003.
40 [6] Imrich Chlamtac, Marco Conti, Jennifer J.-N. Liu, “Mobile ad hoc
20 networking: imperatives and challenges,” ELSEVIER Ad Hoc Net-
works 1, 13-64, 2003.
[7] Ajay Chandra, V. Gumalla, and John O. Limb, “Wireless Medium
0.4 Access Control Protocols,” IEEE Communications Surveys, pp. 2-
0.3 15, 2000.
12000 [8] Ian F. Akyildiz, Janise McNair, Loren Carrasco, Ramon Puigjaner,
0.2 10000
8000 “Medium Access Control Protocols for Multimedia Traffic in Wire-
0.1 6000
4000
Average Packet Loss Rate 0 2000 Video Throughput (kbps) less Networks,” IEEE Network Magazine, vol. 13, no.4, pp. 39-47,
Jul/Aug 1999.
Fig. 5. Video Traffic: Average packet delay and packet loss [9] Eun-Sun Jung, N.H. Vaidya, “An energy efficient MAC protocol
rate versus video throughput in LEPOAP and EDCA for wireless LANs,” in Proc. INFOCOM, IEEE Vol. 3, 23-27 June
2002, Page(s):1756 – 1764.

TCP
LEPOAP
EDCA
Average Packet Delay (ms)

8000

6000

4000

2000

0.3 12000
0.2 10000
8000
0.1 6000
4000
0 2000
Average Packet Loss Rate Traffic Throughput (kbps)

Fig. 6. TCP Traffic: Average packet delay and packet loss


rate versus TCP throughput in LEPOAP and EDCA

You might also like