Você está na página 1de 5

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the IEEE ICC 2009 proceedings

An Efficient MAC Layer Handoff Scheme for


WiFi-based Multichannel Wireless Mesh Networks∗
Zhenxia Zhang and Azzedine Boukerche
PARADISE Research Laboratory
SITE, University of Ottawa, Canada
Email: {zzhan036, boukerch}@site.uottawa.ca

Abstract—Compared to traditional wireless networks, Wireless layer handoff latency of IEEE 802.11b/g [7] protocols is
Mesh Networks (WMNs) are more efficient in terms of deploy- about hundreds of milliseconds, which makes it unsuitable
ment, configuration, and maintenance. However, connecting all for seamless roaming in WMNs. Thus, many works have
routers through wireless connection in WMNs results in remark-
ably lower bandwidth of the network backbone. Multichannel proposed to decrease MAC layer handoff latency in IEEE
technology, where non-interfering channels are used to enable 802.11b/g protocols. However, migrating these schemes to
mesh routers to send and receive packets simultaneously, has been multichannel WMNs cannot meet the requirements of a real-
extensively adopted to improve the throughput of WMNs, and time application. In multichannel WMNs, two neighboring
providing seamless roaming in multichannel WMNs has become ARs may have to send packets through multihops since they
an important topic in WMN research. In this paper, a novel
MAC layer handoff scheme is proposed as a means of minimizing use different channels, which means that data sent between
handoff latency in WiFi-based multichannel WMNs for seamless the neighbors will encounter higher delay and loss ratio than
communication in real-time applications. By designing a dynamic WMNs using one channel. Thus, it is especially important to
grouping algorithm for channel selection and allowing mesh optimize the MAC layer handoff solution for the multichannel
routers to switch channels for probe message reply, this new WMNs.
scheme can shorten the waiting time for the detection of available
access routers, decrease loss ratio of data packets during handoff, In this paper, we propose a novel MAC layer handoff
and consequently achieve smooth handoff in the MAC layer. scheme to decrease scan time for multichannel wireless mesh
networks. The access routers can switch to the channel pre-
I. I NTRODUCTION defined by the mesh client in order to send back the probe
response messages. Therefore, after the mesh clients broadcast
Wireless Mesh Networks (WMNs) are an important com-
the probe request messages in a channel, they switch to the
ponent of the next generation networks for providing wireless
next channel and broadcast the probe request messages without
connection for mobile terminals. Most WMNs have backbones
waiting. Channels are categorized to priority-based groups,
that are composed of mesh routers in order to forward packets
and after broadcasting the probe request messages in all of
[1]. In a WMN, all of the connections are wireless; therefore,
a group’s channels, the mesh client returns to the original
the backbone of a WMN has less bandwidth than that of
connection channel to continue communication. Moreover, the
a wireless network using wired infrastructure. To increase
probing process can be stopped before the whole scan is
the capacity of the backbone, mesh routers equipped with
finished in order to diminish latency, provided a certain AR has
multiradios are allowed to work in different channels [2]. Thus,
better signal strength than at the optimized level. Using this
the mesh routers can send or receive packets in different fre-
scheme, we believe that the handoff latency during the MAC
quencies simultaneously by utilizing non-interfering channels.
layer handoff can be minimized, and that smooth handoff can
Many works have proposed to increase the throughput of the
be realized. To our best knowledge, this is the first optimized
multichannel WMNs[3][4][5][6].
MAC layer handoff protocol for multichannel wireless mesh
WiFi is the most common technology for wireless connec-
networks.
tion. In WiFi-based WMNs, handoff occurs very often during
The rest of this paper is organized as follows. In Section
roaming due to a short transmission range. Handoff therefore
2, background and some protocols for improving handoff
plays an important role in supporting seamless roaming. A
performance are presented. The detailed scheme is proposed
whole handoff process can be divided into two phases: the
in Section 3. Section 4 describes the simulation and provides
MAC layer handoff and the network layer handoff. In the MAC
some discussion on its performance. Finally, Section 5 sum-
layer handoff, the mesh client selects a new AR based on the
marizes and concludes this paper.
signal quality and reassociates with the new AR in physical
connection. In the network layer handoff, the mesh client II. BACKGROUND AND R ELATED W ORKS
re-establishes the logical connection with the correspondent
In IEEE 802.11 [7] protocol, the MAC layer handoff can
nodes by finding a new routing path. The original MAC
be divided to three phases: scan, reauthentication and reasso-
∗ This work is partially supported by NSERC, Canada Research Chairs ciation [8]. There are two kinds of scan, active and passive.
Program, ORF, OIT/Distinguished Researcher Award, EAR Award. Using active scan, the mobile clients broadcast probe request

978-1-4244-3435-0/09/$25.00 ©2009 IEEE


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2009 proceedings

messages and wait for the response messages from the access terminals to reduce MAC layer handoff [15]. The obvious
points in each channel. Conversely, in a passive scan, the drawback is that using the second radio incurs extra cost.
mobile clients listen to the beacon messages from the access Recently, studies of handoff over WMNs focus mostly on
points in each channel. To guarantee that the mesh clients how to reduce the network layer handoff latency [16][17]
receive the beacon messages, the waiting time in a passive [18]. Only [19] tries to minimize the MAC layer handoff over
scan is longer than in an active scan. Therefore, active scan WMNs by letting mesh clients access the backhaul channel.
is a more reasonable option for seamless handoff according to A similar approach is taken in [13] which waives the waiting
most of the literature. time. [19] allows the mesh routers to provide different AP
Probing channels to select the next associated channel is channels. However, it assumes that all of the relay routers
the primary cause of handoff latency [8][9]. Most of the work in the same backhaul channel. Since the mesh routers
improvements made therefore try to minimize scanning time. are assigned different channels to improve the throughput in
The total scan time is affected by both the number of prob- multichannel WMNs, this scheme cannot be adopted.
ing channels and the waiting time for each channel. The
avoidance of scanning unnecessary channels is an efficient III. P ROPOSED S CHEME
approach to decreasing the scan time. Shin et al. abstract the In this section, our efficient MAC layer handoff scheme is
relationship between APs in order to reduce the number of proposed. In general, we allow the mesh routers to switch to
probing channels, by using neighbor graphs and non-overlap the mesh client’s communication channel so it can reply the
graphs [10]. Meanwhile, in [11], a channel mask is used to probe messages and waive the probe waiting time; further-
predict the probable channels for the next association, and more, we use a dynamic grouping algorithm to predict the
a caching structure is adopted to record the former handoffs most probable channels for reducing unnecessary scanning.
for prediction. Based on selective scanning, Mustafa et al.
propose pre-scanning in [12]. Rre-scanning in the background A. Detailed Design
is invoked and the top 5 APs are stored in the dynamic caching Three threshold of Received Signal Strength (RSS) are
structure. Consequently, when the handoff is triggered, the predefined, RSSmin , RSSres and RSSmax . RSSmin refers
mobile node can reassociate with the AP in the buffer, and to the minimum acceptable RSS for sending and receiving
the scan time is minimized. packets. The handoff process is triggered when it has been
Another means of reducing scan latency is to reduce the noticed that the current AR’s RSS, denoted as RSScurr , is
waiting time. M inChannelT ime and M axChannelT ime lower than RSSmin . RSSmax is used to terminate the probing
are two important parameters in active scan. When the mobile process before the whole scan is finished, to decrease scan
terminal waits for the responses, if neither responds and no latency. RSSres is introduced in order to waive unnecessary
traffic is detected during M inChannelT ime, the channel response messages. When the access routers receive the probe
is considered as empty. Otherwise, the mesh client must messages, the routers which notice that the RSS is greater
wait for M axChannelT ime to collect probe responses. This than the RSSres reply the probe request. RSSres equals to
said, waiting time in each channel is significantly affected the current RSS, RSScurr , plus a variable Δ used to prevent
by M inChannelT ime and M axChannelT ime. In [8], the ping-pong effect.
authors suggest that the M inChannelT ime could be 6.5ms All of the channels are categorized into three priority-based
and the M axChannelT ime could be 11ms. Velayos et al. groups: high priority (GroupH), normal priority (GroupN )
propose a formula to calculate the MAC layer handoff time; and low priority (GroupL). A higher priority increases the
based on their experiments, the search time of the active scan likelihood of the new access router working in that channel.
can be reduced by 20% when the M inChannelT ime is set to The group members are maintained by a dynamic grouping
1ms and the M axChannelT ime is set to 10.24ms [9]. In [13], algorithm.
Chintala et al. propose a novel scheme, called FHAP, to waive When the mesh client finds that the current RSS (RSScurr )
the waiting time using inter-APs communication. Instead of is lower than the predefined threshold (RSSmin ), it begins the
sending back the probe reply message, the APs relay the reply probing process. The mesh client scans the channels group-by-
messages to the original AP. Therefore, the mobile client does group on the basis of priority. First, the channels in GroupH
not need to wait for the reply, and it can directly switch to the are scanned. The mesh client sends a notification message
following channels. Since all of the connections in WMNs are to the current AR to claim that the client is about to start
wireless, multihop inter-AP communication has more latency scanning. Upon receiving the notification message, the AR
and a higher loss ratio than the wired backbone, especially in starts to buffer all of the packets sent to the client. The mesh
multichannel WMNs where less location distance does not client then switches to the first channel in GroupH, and broad-
mean fewer hops. This scheme is therefore not especially casts the probe message. The signal strength threshold for
suitable for multichannel WMNs. the response, RSSres , as well as the channel Channelcomm
Finally, smooth MAC handoff is introduced in [14]. The where the mesh client communicates with the former AR and
channels are divided into groups, and after scanning one group, a predefined variable Tdelay which indicates that the client will
the mobile terminal can return to the original channel to return to Channelcomm to collect the response messages after
continue communication. Multiradios are also used in mobile it, are encapsulated in the message. The mesh client does not
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2009 proceedings

A special scenario, where a group contains only one mem-


ber that is the current communication channel is considered
separately. The mesh client does not need to switch channels
and it can send and receive data packets when it is waiting
for a response; as a result, the AR does not need to buffer
the packets during the scan. Other steps are the same as steps
following the probing process described above.
B. Dynamic Grouping
Channels are divided into three groups based on the basis of
their priority. A higher priority means that the channel is more
likely to be selected as the next channel. Priority is measured
as a float number, and its range is [0, 2]. In addition, the
priority range for the high-priority group is (1.5, 2], the range
for the normal priority group is [0.5, 1.5], and the one of the
low priority group is [0, 0.5). The initial priority value of the
non-interfering channels is 2 and that of other channels is 0,
assuming that in a multichannel WMN, the non-interfering
channels are more likely to be used as the AR channel than
the other channels. Each time the scan is finished, the priority
value of each channel is recalculated. If that of the new channel
is the same as that of the former channel, its priority adds 3
* α, and α is the factor used to caculate the priority of the
new channel. If the new channel is one of the non-interfering
channels, other non-interfering channels’ priorities also add 3 *
α, assuming the other non-interfering channels are likely to be
used after the next handoff. The priorities of empty channels
Fig. 1. Flowchart of proposed MAC layer handoff in client side subtract β that is the factor used to caculate the priority of
the empty channel, and the priorities of other channels remain
the same. The procedure for calculating priority is shown in
need to wait for the response messages; it simply switches Algorithm 1.
to the next channel in GroupH and broadcasts the probe
message again. After the client broadcasts the probe messages Algorithm 1 Calculate priority
in all channels in GroupH, it switches to the communication 1: Ci , Cj : Channel i, j
channel to restore the connection, asks the former AR to send 2: Pi , Pj : Priority of channel i, j
the buffered packets and initiates the timer T imerGroupH . The 3:
4: for all Channel Ci do
ARs calculate RSS from the probe messages, and the ARs 5: if Ci is empty then
that have better RSS than RSSres initiate a timer T imerdelay , 6: Pi = Pi - β;
which will expire after TDelay . When T imerdelay expires, the 7: end if
AR switches to Channelcomm to send the response message 8: if Ci is the new channel then
back to the mesh client. If the mesh client finds an AR with 9: if Ci is also the old channel then
10: Pi = Pi + 3 * α;
better RSS than RSSmax , that AR is selected as the new 11: else
AR, T imerGroupH is cleared, and the whole probing stage 12: Pi = Pi + α;
is completed. Otherwise, the retrieved RSS is stored as the 13: end if
best RSS, and the AR with the best RSS is considered as 14: if Ci is one of the non-interfering channels then
the candidate AR. After T imerGroupH expires, the probing 15: for other non-interfering channel Cj do
16: Pj = 3 * α;
process progresses to the following stages, and channels in 17: end for
GroupN and GroupL are scanned. In the next steps, the same 18: end if
operation as in the first stage must be carried out. After the 19: end if
scanning is completed, if no AR has better quality than the 20: end for
predefined threshold, the AR with the best RSS is selected
as the new AR. Then, the mesh client reauthenticates and
reassociates with the new AR. After that, the MAC layer IV. P ERFORMANCE E VALUATION
handoff is finished and the handoff in the upper layer can In this section, performance evaluation is presented. We
be triggered. The flowchart of our scheme on the client side modify the Network Simulator - ns2 (Release 2.33) to simulate
is shown in Figure 1. four MAC layer handoff solutions: standard handoff using full
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2009 proceedings

250 250
Full Scan Full Scan
Selective Scan Selective Scan
FHAP FHAP
Our Solution Our Solution

200 200
Handoff latency (ms)

Handoff latency (ms)


150 150

100 100

50 50

0 0
0 16 32 64 128 256 384 512 0 16 32 64 128 256 384 512
Background Data Traffic (Kbps) Background Data Traffic (Kbps)

Fig. 2. Handoff latency using 1 AR channel Fig. 3. Handoff latency using 3 AR channels

250
Full Scan
Selective Scan
active scan [7], handoff using selective scan [11], FHAP [13], FHAP
Our Solution
and our own solution. There are 30 access routers randomly
200
deployed in our experiments. Different channels are used to
construct the relay backbone for increasing throughput. A
Handoff latency (ms)
Random Waypoint Mobility Model is used to generate the 150

mobile client’s movement path. The maximal velocity is 5


m/s, pause between movements is 1s, M axChannelT ime
100
is 11 ms, M inChannelT ime is 5 ms, channel switch delay
is 5 ms, probe delay is 0.1 ms, and radio range is 150
m. Moreover, α is set to 0.1 and β is set to 0.2. For each 50

experiment, 33 simulations are performed to calculate the 95%


confidence range, and each simulation runs 1000 s.
0
There are 14 channels designated in IEEE 802.11b/g [7], 0 16 32 64 128
Background Data Traffic (Kbps)
256 384 512

only 11 of which channels are used in North America. Channel


1, 6 and 11 are three non-interfering channels. To measure Fig. 4. Handoff latency using 11 AR channels
handoff latency, three scenarios must be considered: access
routers using 1 channel, 3 non-interfering channels, and all
11 channels as their AR channels. Moreover, background FHAP avoids waiting time for probe messages, it needs to scan
data traffic is established in the network in order to obtain all of channels; therefore, its performance is similar to that in
the relationship between handoff latency and the background the first scenario. Selective scan can predict the next available
traffic load. channels with almost full accuracy, and it uses less time to
In the first scenario, all ARs provide access service in finish handoff than full scan and FHAP. Unlike selective scan,
the same channel. The handoff latency in four schemes is except when reducing probing channels, our solution avoids
illustrated in Figure 2. The performance of the selective scan the waiting time by letting ARs switch channels when they
is similar to that of the full scan. When the selective scan send probe reply messages. Thus, our solution still has the
calculates the scan mask, the mask of the current channel is best performance in this scenario.
always 0, the selective scan in this scenario actually does a In the third scenario, all 11 channels are used as AR
full scan all the time. FHAP reduces the handoff latency by channels, access routers are assigned AR channels randomly.
avoiding waiting time, but it still exceeds the 50ms preferred The comparisons for handoff latency are illustrated in Figure 4.
for real-time applications. The handoff latency in our solution Although selective scan and FHAP here reduce handoff latency
is always less than 30ms, despite different background traffic compared with full scan, their latency is still higher than 50ms.
load, because our solution can accurately predict the next They are therefore not suitable for real-time applications such
channel. as VoIP. Our scheme has the least handoff latency, which is
In the scenario where 3 non-interfering AR channels are less than 50ms, since both the waiting time and the number of
used, the access routers provide connection for mobile clients probing channels are minimized. Moreover, our scheme scans
in channels 1, 6 and 11. In our opinion, this scenario would be less channels than selective scan through the use of dynamic
the most common one. In Figure 3, we can see the performance grouping.
comparison of the four schemes in this scenario. Although Inter-frame delay and packets loss ratio are two other
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2009 proceedings

50
Inter-frame delay designed to select the appropriate channel from priority-based
groups quickly. Experimental results prove that this scheme
40 can reduce handoff latency to 50ms or less, which meets the
requirements for general real-time applications.
Inter-frame delay (ms)

30 R EFERENCES
[1] I. F. Akyildiz, X. Wang, and W. Wang, “Wireless mesh networks: A
survey,” Computer Networks, vol. 47, no. 4, pp. 445–487, March 2005.
20 [2] A. Boukerche, “Algorithms and protocols for wireless and mobile ad
hoc networks,” 2008.
[3] A. Raniwala and T. cker Chiueh, “Architecture and algorithms for an
10 ieee 802.11-based multi-channel wireless mesh network,” in Proceeding
of IEEE INFOCOM 2005, vol. 3, March 2005, pp. 2223–2234.
[4] A. Boukerche, “Handbook of algorithms for wireless networking and
mobile computing,” 2005.
0
0 100 200 300 400 500 600 700 800 900 1000 [5] M. Nekoui, A. Ghiamatyoun, S. N. Esfahani, and M. Soltan, “Iterative
Data packet number cross layer schemes for throughput maximization in multi-channel wire-
less mesh networks,” in Proceedings of 16th International Conference
Fig. 5. Inter-frame Delay on Computer Communications and Networks, 2007., August 2007, pp.
1088–1092.
TABLE I [6] K. Karakayali, J. H. Kang, M. Kodialam, and K. Balachandran, “Joint
DATA PACKETS LOSS RATIO resource allocation and routing for ofdma-based broadband wireless
mesh networks,” in IEEE International Conference on Communications,
Scenario Packets sent Packets lost Loss ratio 2007, June 2007, pp. 5088–5092.
1 AR channel 10000 3.291 0.033% [7] “IEEE standard for information technology-telecommunications and
information exchange between systems-local and metropolitan area
3 AR channels 10000 6.289 0.063% networks-specific requirements - part 11: Wireless lan medium
11 AR channels 10000 8.742 0.087% access control (mac) and physical layer (phy) specifications,”
http://standards.ieee.org/getieee802/802.11.html, 2007.
[8] W. A. Arunesh Mishra, Minho Shin, “An empirical analysis of the
ieee 802.11 mac layer handoff process,” ACM SIGCOMM Computer
important parameters in real-time applications. To test them, Communication Review, vol. 33, no. 2, pp. 93–102, April 2003.
CBR traffic is established between two mobile terminals. One [9] H. Velayos and G. Karlsson, “Techniques to reduce the ieee 802.11b
client sends a 160-byte UDP packet every 20ms to the other handoff time,” in 2004 IEEE International Conference on Communica-
tions, vol. 7, June 2004, pp. 3844–3848.
client, to simulate the G.711 encoded/decoded VoIP stream. [10] M. Shin, A. Mishra, and W. A. Arbaugh, “Improving the latency of
In each simulation, 10000 UDP packets are fully sent. Figure 802.11 hand-offs using neighbor graphs,” in Proceedings of the 2nd
5 shows the inter-frame delay of 1000 data packets in a international conference on Mobile systems, applications, and services,
June 2004, pp. 19–26.
simulation using 3 AR channels, as an example. Since the [11] S. Shin, A. G. Forte, A. S. Rawat, and H. Schulzrinne, “Reducing mac
relay backbone is constructed by wireless connection and the layer handoff latency in ieee 802.11 wireless lans,” in Proceedings of
handoff occurs frequently, the jitter is easily observed in Figure the second international workshop on Mobility management and wireless
access protocols, October 2004, pp. 70–83.
5. However, the inter-frame delay is still almost less than 50ms. [12] N. Mustafa, W. Mahmood, A. A. Chaudhry, and M. Ibrahim, “Pre-
Moreover, the average packet loss ratio is shown in Table I. scanning and dynamic caching for fast handoff at mac layer in ieee
This loss ratio is acceptable for the real-time application. 802.11 wireless lans,” in IEEE International Conference on Mobile
Adhoc and Sensor Systems Conference, November 2005, pp. 1–8.
In summary, our scheme minimizes the handoff latency to [13] V. M. Chintala and Q.-A. Zeng, “Novel mac layer handoff schemes
less than 50ms in the three scenarios discussed above, and the for ieee 802.11 wireless lans,” in IEEE Wireless Communications and
inter-frame delay is also less than 50ms. In addition, the loss Networking Conference, March 2007, pp. 4435–4440.
[14] Y. Liao and L. Cao, “Practical schemes for smooth MAC layer handoff
ratio is very low, at less than 0.1%. Our scheme is therefore in 802.11 wireless networks,” in International Symposium on a World
very well-suited to real-time applications. of Wireless, Mobile and Multimedia Networks, June 2006, pp. 1–10.
[15] K. Ramachandran, S. Rangarajan, and J. C. Lin, “Make-before-break
V. C ONCLUSION mac layer handoff in 802.11 wireless networks,” in 2006 IEEE ICC,
vol. 10, June 2006, pp. 4818–4823.
In WMNs, constructing the whole network backbone with [16] V. Navda, A. Kashyap, and S. R. Das, “Design and evaluation of
wireless connections leads to various advantages but at the iMesh: an infrastructure-mode wireless mesh network,” in Sixth IEEE
expense of decreased bandwidth, short transmission range and International Symposium on a World of Wireless, Mobile and Multimedia
Networks, June 2005, pp. 164–170.
more frequent congestion. More occurrences of handoff are [17] Y. Amir, C. Danilov, M. Hilsdale, R. Musaloiu-Elefteri, and N. Rivera,
required in order to maintain the connection; providing seam- “Fast handoff for seamless wireless mesh networks,” in Proceedings of
less handoff in multichannel WMNs for real-time applications the 4th international conference on Mobile systems, applications and
services, June 2006, pp. 83–95.
has therefore become a challenging task. This paper presents [18] A. Boukerche and Z. Zhang, “A hybrid-routing based intra-domain
an efficient MAC layer handoff scheme for minimizing the mobility management scheme for wireless mesh networks,” in Proceed-
latency during handoff. Instead of replying to probe requests ings of the 11th international symposium on Modeling, analysis and
simulation of wireless and mobile systems, October 2008, pp. 268–275.
immediately, access routers are proposed to switch to the [19] Y. He and D. Perkins, “BASH: A backhaul-aided seamless handoff
client’s communication channel and send response after a scheme for wireless mesh networks,” in International Symposium on
delay, which allows the mesh client to broadcast a probe a World of Wireless, Mobile and Multimedia Networks, June 2008, pp.
1–8.
request without waiting. A dynamic grouping algorithm is

Você também pode gostar