pclf: a practical cross-layer fast handover mechanism in ieee...

PCLF: A Practical Cross-Layer Fast Handover Mechanism in IEEE 802.11 WLANs

Yilin Song1,2, Min Liu1, Anfu Zhou1, Zhongcheng Li1, Qi Li3 1Institute of Computing Technology, Chinese Academy of Sciences, Beijing, P.R. China

2Graduate University of Chinese Academy of Sciences, Beijing, P.R. China 3Department of Computer Science, Tsinghua University, Beijing, P.R. China

E-mail: {songyilin, liumin, zhouanfu, zcli }@ict.ac.cn, [email protected]

Abstract—As is known to us, the handover latency of FMIPv6 in its predictive mode is given little concerns. However our previous work [4] shows that FMIPv6 may suffer long handover latency in its predictive mode, and [4] identifies three key issues raising such problems. In this paper, we propose a practical cross-layer fast handover management mechanism (PCLF) to address these issues and improve success rate of mobility prediction. To solve the problem, PCLF includes a smart link layer trigger, a TBScan algorithm, a TBAPS algorithm, a buffering support Bi-Binding scheme and the smart link event notification policy. Experiment results show that our mechanism can achieve reasonable mobility prediction and seamless handover with no interruptions on upper layer applications (VoIP) in IEEE 802.11 WLANs. The average handover latency is less than 50ms, the success rate of mobility prediction is 97.7% and no packet loss is observed.

Keywords- IEEE 802.11; FMIPv6; fast handover; cross layer; I. INTRODUCTION

In IEEE 802.11 WLANs, when an MN (mobile node) hands over to an AP in a different IP subnet, both the link layer handover and the network layer handover are needed. Since IEEE 802.11 only supports link layer handovers, MIPv6 is proposed to support network layer handovers. However, MIPv6 suffers long handover latency. To reduce the handover latency, some mobility prediction based schemes [1]-[3] have been proposed. FMIPv6 [1] is the most popular one. FMIPv6 reduces the handover latency by conducting the fast handover operations before the link layer handover. Since the time of the link layer handover is hard to control, FMIPv6 has two different modes, the predictive mode and the reactive mode, distinguished by whether the fast handover operations are completed before the link layer handover.

There are many works focus on analysis and improvement of FMIPv6 [4]-[10] in literature. However, as is known to us, the handover latency of FMIPv6 in its predictive mode is given little concerns relatively. Most of the existing works assume that FMIPv6 can achieve its best performance when the handover prediction is correct and the MN goes into its predictive mode. The link layer factors and the efficiency of interactions between the link layer and the network layer, which may affect the handover latency, are usually ignored. There are few works based on experiments in real test-bed. Moreover, in most previous experiment works, handovers are triggered through the pre-defined time table, while in practice handovers are actually triggered by link layer events.

[5] and [7] focus on analyzing the overhead cost introduced by FMIPv6 especially when an MN goes into the reactive

mode. [5]-[8] are conducted by simulation or analytical approaches. There are also existing works on measurement [9] and improvement [10] of FMIPv6 in test-bed. However, handovers are triggered by the pre-defined time table [9], thus, link layer factors that may impact the handover latency cannot be found or resolved. [10] presents that FMIPv6 may suffer long handover latency in the predictive mode. However, [10] does not identify or solve the inefficiency of interactions between the link layer and the network layer. And the proposed mechanism will suffer bad performance when the MN moves back and forth. Our previous work [4] identifies three key issues affecting handover latencies of the predictive FMIPv6 in IEEE 802.11 WLANs through experiments.

In this paper, considering both the link layer and network layer factors and their interaction efficiency, we propose a practical cross-layer fast handover management mechanism – PCLF. Our work focuses mainly on: 1) improve the success rate of mobility prediction; 2) reduce handover latency of FMIPv6 in its prediction mode. PCLF handovers are triggered through smart link layer triggers proposed by us, which allow the MN to start the related network handover operations right after the link layer prediction or the link layer handover. Using TBAPS, an MN can select a target AP through link quality tendency and network layer information, thus, the success rate of mobility prediction can be improved. And we propose TBScan, Bi-Binding and the smart link event notification policy to address the three affecting issues identified by [4] respectively. Experiment results show that our mechanism can achieve reasonable prediction and seamless handover with no interruptions on upper layer applications (VoIP). The average handover latency is less than 50ms, the success rate of mobility prediction is 97.7% and no packet loss is observed through all our experiments.

Paper organization: Section II reviews the issues affecting the predictive handover latency. PCLF is proposed in section III. Section IV and V describe the implementation of PCLF and the experiment results. Section VI summarizes the paper.

II. PROBLEM STATEMENT This section briefly reviews the phases of the predictive

FMIPv6 handover (Fig. 1) and the three key issues that we should address when designing a fast handover management mechanism presented in our previous work [4]. Important parameters are listed in Table. 1. Then we will discuss the design goals and our proposed scheme in the next section.

This work has been supported by the Beijing Nova Program, the National Basic Research Program of China (No. 2007CB310702) and the National Natural Science Foundation of China (No. 60803140 and No. 60970133).

978-1-4244-6404-3/10/$26.00 ©2010 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings

A. Phases of Fmipv6 Predictive Handover Procedure 1) Link Layer Prediction Latency - Tl2Pre In the link layer prediction phase, the MN has to complete

the AP scanning and the target AP selection. And Tl2Pre can be calculated by (1). tscan is the AP scanning delay.

l2Pr e 1 0 scanT t t t ChannelNum ChannelWaitTime= − ≈ = ⋅ (1)

2) Predictive Network Handover Latency - Tl3Pre In the predictive network handover phase, the fast handover

operations are completed. Tl3Pre can be calculated by (2). During this period, the MN can still send and receive packets.

l3Pre 2 1T t t= − (2) 3) Predictive Tunneling Latency - TPre-T The MN might still connect with the PAR when the packets

are tunneled to the NAR. In this situation, extra handover latency (TPre-T) is introduced, which can be calculated by (3).

Pr e T 3 2T t t− = − (3) 4) Link Layer Handover Latency - Tl2 Tl2 is defined as (4). tdisa, tauth and treass denote the delays for

the disassociation from the current AP, the authentication and the reassociation with the new AP respectively.

l2 4 3 disa scan auth reassT t t t t t t= − = + + + (4) 5) Sender Preparation Latency - TsPre

TsPre is system depended and is defined by (5). sPre 5 4T t t= − (5)

In summary, all these above phases except the predictive network handover phase can introduce latency to the total handover latency. Thus, the total handover latency of the predictive FMIPv6 can be calculated by (6).

total_ handover l2Pre Pre T l2 sPreT T T T T−= + + + (6)

B. Issues Affecting handover latency of predictive FMIPv6 1) The lack of assistance from the network entities, the

key issue affecting Tl2Pre during the link layer prediction phase and Tl2 during the link layer handover phase. Moreover, only instantaneous information of neighbor APs can be obtained during the prediction phase. Handover predictions based on instantaneous information may be wrong at a high probability.

2) The ambiguous link layer triggering time, the key issue affecting TPre-T during the predictive tunneling phase.

3) The inefficient interaction between the link layer and the network layer, the key issue affecting TsPre during the sender preparation phase.

MN

Tl2Pre

NAR PAR

HI

HACK

TPre-T

T l2

TsPre

T l3Pre

Tbuffer

TsPre

RtSolPrPrRtAdv

FBU

FBACK

Tgd

Td

Tu

t0

t1

t2

t3

t4

t5

RtSolPr: Router Solicitation for Proxy Advertisement

PrRtAdv: Proxy Router Advertisement

FBU: Fast Binding Update

FBACK: Fast Binding Acknowledge

HI: Handover Initiate

HACK:Handover Acknowledge

FNA: Fast Neighbor Advertisement FNA

Fig. 1. The Predictive Handover Procedure of FMIPv6. The red solid lines represent periods during which the MN cannot send or receive packets.

Table 1 Parameters for Analysis Tgd link going down trigger Td link down trigger Tu link up trigger

(THp,THho) threshold for link layer prediction and link layer handover t0 the moment when the MN begins to probe the neighbor APs t1 the moment when the MN determines the target AP t2 the moment when the PAR begins to forward packets for

the MN to the NAR and sends the FBACK to the MN t3 the moment the MN starts its link layer handover t4 the moment when the link layer handover finishes t5 the moment when the NAR receives the FNA from the MN

III. PROPOSED SCHEMES A. Our Design Goal

Only when all the three issues above are solved, can we achieve predictive fast handovers in real 802.11 WLANs. Thus, we propose PCLF, a practical cross-layer fast handover management mechanism, includes the following parts:

Smart Link Layer Triggers: provide interaction information between the link layer and the network layer.

TBScan and TBAPS: solve the first issue. TBScan, an AP scanning algorithm, is to reduce Tl2Pre during the link layer prediction phase and Tl2 during the link layer handover phase; TBAPS, a target AP selecting algorithm, is to improve the success rate of mobility prediction.

Buffering Support Bi-Binding Scheme: solve the second issue, reduce TPre-T during the predictive tunneling phase.

Smart Link Event Notification Policy: solve the third issue, reduce TsPre during the sender preparation phase. B. Overview of PCLF

In public WLANs, the engineers should know the neighbor channels (channels on which neighbor APs exist) of each AP as well as the prefix and the MAC address of the attached AR. The information mentioned above can be configured when APs are deployed. And we introduce these information into 802.11 management frames, which will be utilized by the following schemes proposed in this Section.

As is shown in Fig. 2, the MN moves away from its current AP, the RSSI gradually decreases. The main stages of PCLF:

1) Mobility Prediction Preparation: When the RSSI goes into Level 2, the MN goes into the TBScan process until the real link layer handover happens. Information such as the neighbor AP’s RSSI, the prefix and the MAC address of the attached AR are stored in a data base.

2) Predictive Handover: When the RSSI goes into Level 3, the predictive target AP is selected by the TBAPS using information collected by TBScan. If the target AP is in the different subnet from the current AP, a Tgd is sent to the network layer to start the predictive handover procedure. Then the MN sends a PBU and a BN (introduced by Bi-Binding) to the HA and the target AR respectively. Upon receiving the PBU, the HA starts to forward packets to the MN’s current and new locations simultaneously. Upon receiving the BN, the NAR starts to buffer packets for the MN.

3) Regular Handover: When the RSSI goes into Level 4, the link layer handover occurs; the target AP is selected by the TBAPS using information collected by TBScan. After the MN disassociates with the previous AP, the smart link event


MN

APs in range on CH1

ProR

eq

Level 2 Level 4

APs in range on CH6

Dis

aReq

Current AP(CH11)

Level 1

ProR

eq

ProRes

THs Level 3THp THho

Aut

hReq

Rea

Req

Tgd Td

Tu

AuthR

es

RSSI

ProRes

DisaR

es

ProRes

ProRes

TBScan

HA(Home Agent)

NAR

PBU B

A

BN

BN

A

BA

AN

ReaR

es

BU

Bi-Binding

New AP1(CH1)

Buffering Forwarding

AN

A

Key Thresholds: THs: threshold for TBScan THp: threshold for mobility predictions THho: threshold for link layer handoversIEEE 802.11 link layer handover related management frames: ProReq: Probe Request ProRes: Probe Response DisaReq: Disassociation Request DisaRes: Disassociation Response AuthReq: Authentication Request AuthRes: Authentication Response ReaReq: Reassociation Request ReaRes: Reassociation ResponseSignaling Messages used by Buffering support Bi-Binding Scheme: PBU: Predictive Binding Update BU: Binding Update BA: Binding Acknowledge BN: Buffering Notification BNA: Buffering Notification Acknowledge AN: Access Notification ANA: Access Notification Acknowledge

Fig. 2. link layer perspectives of handover procedure in our scheme. The red solid lines represent periods when the MN can not send or receive packets

notification policy is used to guarantee that the MN can send packets right after it reassociates with the new AP. Once the MN reassociates with a new AP, the Tu is sent to the network layer to start the regular handover procedure. And the MN sends an AN (introduced by Bi-Binding) and a BU (defined in standard MIPv6) to the new AR and the HA respectively. Upon receiving the AN, the NAR starts to forward packets to the MN. Upon receiving the BU, the HA stops forwarding packets for the MN to MN’s old location. C. Smart Link Layer Triggers

We introduce the prefix and MAC address of the target AR (can be extracted from our extended 802.11 management frames) into Tgd and Tu, so that FMIPv6’s RtSolPr can be saved, and the MN need not wait for a RA even if the prediction is wrong. And we introduce Tl2 into Tu. This latency can help the NAR to reduce the duplicated packets the MN may receive, which will be described in subsection F. D. Topology based Background AP Scanning Algorithm

- TBScan The main idea of TBScan is to divide the long AP scanning

phase into small pieces, which is similar to [14]. So that Tl2Pre and Tl2 can be eliminated. The channels are scanned one by one periodically. However, it will take a long time to update the quality of the entire neighbor APs. We improve [14] by introducing the neighbor channels, which can be obtained from AP’s association/reassociation response. Thus, the unnecessary channel scanning can be avoided and the MN can update its neighbor APs’ quality more quickly, and more information of each neighbor can be obtained. This provides the opportunity to make more reasonable target AP decision.

Each channel probing operation is called a TBScan session, and the delay of each TBScan session can be calculated by (7).

TBScan CHSwitch MinCHWait MaxCHWaitT 2 T (1 p(i)) T p(i) T= ⋅ + − ⋅ + ⋅ (7) Where TCHSwitch is the time to switch channels, p(i) is the

probability of one or more AP exist on channel i. TMinCHWait and TMaxCHWait represent the min and max waiting times on each channel defined by 802.11. E. Tendency based Target AP Selection Algorithm– TBAPS

The link quality information obtained during the TBScan is stored in a data base. And thus, the target AP can be selected

concerning both the recent updated RSSI and the tendency of RSSI, which is proved to be efficient in [15]. We improve the scheme in [15] by making the final decision with the assistant of the prefix information of each neighbor AP’s attached AR. The quality of each neighbor AP is calculated using (8).

i y i

y dw ai

i y i

y dw a

R S SI T h C N in the sam e subneth t /( I n )

QR S SI T h C N else

h t /( I n )

α β ε

α β

−⎧+ + ⎪ ⋅⎪= ⎨ −⎪ + ⎪ ⋅⎩

(8)

RSSIi is the most recent updated RSSI of APi, CNi is the number of consecutive RSSI above the Thy of APi. α, β, ε are the weight factors of the current RSSI, the tendency of RSSI and the network prefix. THy is the RSSI threshold, hy denotes hyteresis and tdw denotes the dwelling timer, Ia is the average interval of TBScan, n is the number of the neighbor channels. F. Buffering Support Bi-Binding Scheme

Since it is impossible for the PAR to know when exactly the MN will reach the NAR, our Bi-Binding scheme (similar to [2], we improve it by provide buffering and duplicated packets reducing support) forwards packets to the PAR and the NAR simultaneously. So that TPre-T can be eliminated. The Bi-Binding is carried out by the HA. New signaling messages are defined, including PBU, BN/BNA and AN/ANA. The PBU is used to notify the HA that both the predictive and previous bindings should be maintained and packets should be forwarded to MN’s current and new locations of the MN simultaneously. The NAR sets up a buffer entry and starts to buffer packets for the MN after the exchange of the BN and BNA with the MN, and starts to forward packets to the MN after the exchange of the AN and ANA. The MN records the Tl2 and informs it to the NAR by the AN. To reduce the duplicated packets the MN may receive, the NAR records the time it receives the AN (TAN), then the NAR checks the arrival time (Ti) of each buffered packets, only packets meet the requirement shown in (9) will be forwarded to the MN.

i AN l2T > T (T )δ− + (9) Where δ is the paths asymmetry adjusting factor, since the

paths from the HA to the PAR and the NAR are different. How to adjust δ adaptively will be investigated in our future work. On receiving a regular BU, the HA deletes the previous binding and stops forwarding packets to the MN’s previous location. A timer (20s) is used to deal with wrong predictions. G. Smart Link Event Notification Policy

To eliminate the TsPre of the NM under slow link event processing system (eg, linux-2.6.15), we propose a smart link event notification policy. When the MN disassociates from its current AP, a timer, 30ms is started (the max time for an MN to dissociate, authenticate and reassociate with a new AP observed in our experiments is 27ms). Only when the MN does not connect with a new AP within 30ms, the link down event will be sent to the link-watch module.

IV. SYSTEM DESIGN AND IMPLEMENTATION Fig. 3 illustrates the architecture of our proposed PCLF,

and only the modules added or extended by us are listed. The implementation of our schemes on the MN consists of two main parts: the L2HM (link layer handover management,


Fig. 3. Architecture of our fast handover management system.

based on [12]) and the L3HM (network layer handover management, based on [11]). In MN’s link quality monitoring module, the RSSI will be updated once a frame is received, the dwell timer, DWlinkQ, is used to alleviate the affection of RSSI fluctuations. The continuous time of the current AP’s RSSI below a certain threshold is recorded. Only when the dwell timer expires, the corresponding operations will be conducted. The link layer trigger is driven by events and is implemented based on Netlink. With triggers received, the L2-Trigger Process module operates with the expanded MIPL to finish the predictive and regular handovers. The MIPL daemon and the routing mechanism on the HA are expanded to support Bi-Binding. And we implement the packets buffering on the AR by using the netfilter mechanism. The arriving time of a packet is recorded to recognize and reduce the duplicated packets during the Bi-Binding process.

We do not extend IEEE 802.11 management frames, since the commercial APs we use cannot be modified. Instead, to evaluate the extended 802.11 management frames we propose, we configure each AP a special ESSID, eg, 42120010CC0202 60006/021478845d3a. Where the first three characters mean the neighbor channels, “421” is parsed as 010000100001, every bit represents one channel in 802.11b, and the rest characters provide the prefix and MAC address information. Table. 2 lists important parameters in test-bed. Perspicacious readers may notice: THho is -61; when RSSI is -61, the link is still usable. That is because our office is not big enough; RSSI can only range from -13 to -65 when setting an AP at location E in Fig. 4 and the MN moves from location E to F.

V. PERFORMANCE EVALUATION All our experiments are done in a 70m * 35m office area as

shown in Fig. 4. The location A, B, C, D, E, F and G are places where the APs are set during experiments.

The logical network topology is shown in Fig. 5. The main experiment devices include a HA, 3 ARs, 4 APs (NETGEAR WGR614 v7), a CN and an MN. The MN is a IBM T43 notebook equipped with an Intel 2200 Wireless NIC; all the devices except the APs are running Linux; the MN moves at a regular walking speed (1m/s~1.5m/s).

The efficiency of the target AP selection algorithm, the handover latency and the overhead introduced by our scheme are analyzed. To simulate real time upper layer applications (VoIP), the 100-byte UDP packets are sent to the MN at the interval of 20ms from the CN during all the experiments.

Table 2 Parameters in Our Test-Bed AP1 ESSID 42120010CC020260006/021478845d3a AP2 ESSID 40120010CC020260007/02147883eb8b AP3 ESSID 02120010CC020260008/02105cae71cd AP4 ESSID 42120010CC020260006/021478845d3a Default Scan Interval , TCHSwitch 500ms , 5.5ms THl, THh; Ih , Im , Il ; TMaxCHWait -51, -54; 300ms, 200ms,100ms; 15ms α, β,ε,THy, hy, tdw 1, 1, 0.5, -45, 5, 15s THs, THp, THho, DWlinkQ, δ -48, -56, -61, 6, 300ms

1) The target AP selection Algorithm We place AP1, AP2, AP3 and AP4 at location E, D, G and

F as is shown in Fig. 4. The MN first connects with AP1 and moves from location E to D. The TBScan is triggered at 40.9s, and the target AP selection is triggered at 72.1s. The AP4 in the same subnet as AP1 is selected. As we can see from Fig. 6, AP2’s signal quality is slightly better than AP4. However, choosing AP4 as the target AP can avoid the unnecessary network layer handover. The RSSI fluctuations of AP3 because of the obstacles do not result in a wrong prediction.

We repeated this experiment 50 times, the probability of the wrong prediction is 2.3% and 26.7% with and without TBAPS respectively. The result shows that our algorithm can make more reasonable target AP selection. As is described in [15] the overhead introduced is acceptable.

2) Service Interruption During the Handover We place AP1, AP2, AP3 (in different subnets) at location

E, D, G as is shown in Fig. 4. The MN first connects with AP1 and moves from location E to D. The Packet arrival latency is used to evaluate the service interruption during the handover. We calculate this latency by subtracting the packet sending time at the CN from the packet arrival time at the MN.

Fig. 4. Map and device locations for handover experiment

CNIntranet

Switch

AR1

MN

Foreign subnet 3

HubAP1(CH1)

AR3

AR2Foreign subnet 2

AP4(CH1)

AP2(CH6)

AP3(CH11)

HA

Foreign subnet 1

Fig. 5. Experiment Scenario. The link of backbone of the Intranet is

1000Mbps Ethernet and others are 100Mbps Ethernet.

-60

-50

-40

-30

-20

0 10 20 30 40 50 60 70 80

RS

SI

Time (sec)

THs

THp

THy

T1=40.8s T2=72.1

AP1AP2AP3AP4

Fig. 6. RSSI changing tendency during the experiment.


The performance of FMIPv6 with the original AP scanning algorithm in Intel 2200 driver is shown in Fig. 7, where t0, t1, t2, t3, t4 and t5, defined in section 2, are 65.87s, 66.43s, 67.52s, 71.14s, 71.61s and 72.14s respectively. The MN begins to probe its neighbor APs at t0 and the target AP is selected at t1. Packets are tunneled to the NAR at t2. However, the real link layer handover occurs at t3. And packets can not be received from the NAR until t5 when the MN can finally send the FNA to the NAR. The overall handover latency is 5.18s and 21 packets during the link layer prediction phase are lost. We repeated this experiment 50 times. The shortest, the longest and the average handover latencies observed are 1.6s, 15.3s and 5.6s respectively.

Fig. 8 shows the performance of our mechanism, where we walk on the same track with the same speed as that in the experiment of FMIPv6 with the original Intel 2200 driver. The TBScan is triggered at 38.7s. The target AP selection is triggered at 67.2s. The real link layer handover occurs at 71.61s and is finished at 71.62s. Then the MN can receive packets from the NAR after the exchange of the AN and ANA. The seamless handover is provided without any interruption on the upper layer applications. We repeat the experiment 50 times (handover latencies less than 50ms) and observe at most 9 duplicated packets received by the MN. Handover latencies introduced by the issues listed in section II are eliminated.

3) Overhead Introduced by Our Mechanism To evaluate the jitters introduced by TBScan, the CDF of

the packet inter arrival time of packets received by the MN under different scan interval are measured. As is shown in Fig. 9, when no scan operations are conducted, the probability that the packet inter-arrival time is larger than 27.3 is 0. And this probability increases to 3.5%, 4.2%, 8.9% respectively when the interval is 500ms, 300ms, 100ms. The biggest interval observed is 47ms. As is widely agreed [14], interruptions smaller than 50ms are undetectable by real time applications (VoIP) users. The average number of duplicated packets (50 experiments) introduced by Bi-Binding in our test-bed is 239. The duplicated packets can be greatly reduced by introducing IP multicast scheme, which is not the focus of this paper.

0

1

2

3

4

5

35 40 45 50 55 60 65 70 75 80

Pac

ket A

rriv

al L

aten

cy (

s)

Packet Arrival Time at MN (s)

original FMIPv6

0

0.2

0.4

0.6

0.8

1

64 66 68 70 72

t0 t1 t2 t3t4 t5

packtes loss

Fig. 7. Handover procedure of FMIPv6 and the original AP scanning

algorithm in Intel 2200 driver

0

0.2

0.4

0.6

0.8

1

30 40 50 60 70 80

Pac

ket A

rriv

al L

aten

cy (

s)

Packet Arrival Time at MN (s)

more big jitters

38.7 67.2 71.61

71.62

our scheme

Fig. 8. Handover procedure of our mechanism

0

0.2

0.4

0.6

0.8

1

0 5 10 15 20 25 30 35 40

CD

F o

f pac

kets

Packets Inter-arrival Time (ms)

no scan500ms interval300ms interval100ms interval

Fig. 9. CDF of packet inter-arrival time under different scan interval

VI. CONCLUSIONS In this paper, we propose a practical cross-layer fast

handover management mechanism - PCLF, which can address the three key issues affecting handover latency of predictive FMIPv6. PCLF is consisted of the smart link layer trigger, a TBScan algorithm, a TBAPS algorithm, the smart link event notification policy and a buffering support Bi-Binding scheme. Experiment results show that our mechanism can achieve reasonable prediction and seamless handover with no interruptions on upper layer applications (VoIP). The average handover latency is less than 50ms, the success rate of mobility prediction is 97.7% and no packet loss is observed.

REFERENCES [1] R.Koodli. “Fast handovers for mobile IPv6”, IETF RFC4068. [2] Karim El Malki, Hesham Soliman, “Simultaneous Bindings for

Mobile IPv6 Fast Handovers”, 2005. [3] R. Hsieh, Z. G. Zhou, and A. Seneviratne, ”S-MIP: A Seamless

Handoff Architecture for Mobile IP,” INFOCOM 2003. [4] Yilin Song, Min Liu, Zhongcheng Li, Qi Li, “Handover

Lantency of predictive FMIPv6 in IEEE 802.11 WLANs: A Cross Layer Perspective”, ICCCN 2009.

[5] Sangheon Pack, Yanghee Choi，“Performance Analysis of Fast Handover in Mobile IPv6 Networks”, PWC 2003.

[6] Jiang Xie, Ivan Howitt, and Izzeldin Shibeika, “IEEE 802.11-based Mobile IP Fast Handoff Latency Analysis”, ICC 2007.

[7] Xavier P´erez-Costa, Marc Torrent-Moreno, Hannes Hartensteinab. “A Performance Comparison of Mobile IPv6, Hierarchical Mobile IPv6,Fast Handovers for Mobile IPv6 and their Combination”, ACM SIGMOBILE Mobile CCR 2003.

[8] R. Hsieh, A. Seneviratne, “A Comparison of Mechanisms for Improving Mobile IP Handoff Latency for End-to-End TCP,” ACM MOBICOM 2003.

[9] K. Yong-Sung, K. Dong-Hee, B. Kyung-Jin, and S. Young-Joo, “Performance Comparison of Mobile IPv6 and Fast Handovers for Mobile IPv6 over Wireless LANs”, VTC 2005-Fall.

[10] Hancheng Lu, Xiaobo Zhou, Peilin Hong, “Improving the Performance of Fast Handovers in Mobile IPv6”, GLOBECOM 2007.

[11] http://www.mobile-ipv6.org/software/ (MIPL-2.0.1) [12] http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&Prod

uctID=1637&DwnldID=10138&strOSs=39&OSFullName=Linux*&lang=eng (Intel 2200 driver)

[13] H. Velayos and G. Karlsson, “Techniques to Reduce IEEE 802.11b MAC Layer Handover Time,” KunglTekniska Hogskolen, Stockholm, Sweden, Tech. Rep. 2003.

[14] Haitao Wu, Kun Tan, Yongguang Zhang, Qian Zhang, “Proactive Scan: Fast Handoff with smart Triggers for 802.11 Wireless LAN,” INFOCOM 2007.

[15] Liu Min, Li Zhongcheng, Guo Xiaobing, Dutkiewicz Eryk, “Performance Analysis and Optimization of Handoff Algorithms in Heterogeneous Wireless Networks,” Transactions on Mobile Computing, 2008.


pclf: a practical cross-layer fast handover mechanism in ieee...

Documents