An Adaptive and Preemptive Algorithm for Faster Handoffs in WLAN

Transcription

1 An Adaptive and Preemptive Algorithm for Faster Handoffs in WLAN Vijay Ukani Computer Engineering Department Institute of Technology, Nirma University Ahmedabad, Gujarat, India M. M. Gore Computer Science and Engineering Department Motilal Nehru National Institute of Technology Allahabad, Uttar Pradesh, India December 6, 2006 Abstract IEEE Wireless LANs have shown enormous growth in recent times. Due to the mobility promises shown by them, seamless roaming has become a critical component of the whole deployment. A handoff occurs when a Mobile Station moves away from transmission range of one Access Point and enters into transmission range of another Access Point. The current handoff procedure does not satisfy the jitter requirements for real time multimedia applications like VoIP. In this paper we propose a new handoff algorithm which reduces handoff latency to an extent to be seamlessly bearable by a multimedia application. We reduce the search phase latency by periodically scanning channels one at a time and caching the results. The cached results can be used be used for future handoff management. 1 Introduction Wireless network facilitates mobility with high speed information access. It has led to rapid growth in number of IEEE Wireless LAN deployments for internet and local network access. There are currently three IEEE WLAN standards[1]: a, b, and g. The IEEE a operates in 5 GHz ISM band and has 32 channels. The b and g operates in 2.4 GHz ISM band and use 11 of possible 14 channels. Of these 11 channels only 3 do not overlap in frequency range as shown in figure 1. In this paper we are concerned with IEEE b. The IEEE WLAN operates in two modes viz, Infrastructure and Ad-hoc mode. In Ad-hoc mode there is no central entity. Whenever two Mobile Stations (MS) come in communication range of each others, after detecting each other they establish a peer-to-peer communication link between them. Work conceived and completed during authors association with MNNIT, Allahabad as a masters student Figure 1: Channel Layout of b WLAN In Infrastructure mode of WLAN, there are Access Points (AP) acting as Base Stations which makes a single point of communication for all MS associated to it. The MSs and its associated AP forms a Basic Service Set(BSS). The BSS can be extended by connecting APs through a Distribution System (DS) to form Extended Service Set(ESS). Whenever an MS roams across APs, it makes handoff from current AP to a new AP in the range. During a handoff the MS leaves the association with old AP and establishes a new association with new AP. This process of locating a new AP and associating with it, generates latency. This latency may be long enough to break the ongoing communication or induce delay in transmission, which is not acceptable for real time multimedia applications. In this paper a new algorithm is proposed with which we were able to limit the handoff latency to a level where even VoIP application can work seamlessly. The rest of the paper is organized as follows. Section 2 details the operation of b WLAN, focusing on the handoff procedure and the related terms. Section 3 describes the existing methodologies to reduce handoff latency in WLAN. In section 4, a new algorithm for performing the handoff is presented. Section 5 demonstrates the simulation scenario and section 6 provides results and analysis. Finally, in section 7 some concluding remarks and scope for future work is presented.

2 2 The Handoff Process and Related Terminology As per ethernet technologies, roaming domain would mean a network, that connects devices that are capable of sending and receiving broadcast frames to and from one another. This domain is sometimes referred to as layer 2 network. In case of , APs that are in the same broadcast domain and configured with the same service set identifier(ssid) are said to be in the same roaming domain. IEEE natively supports layer 2 roaming ie. roaming within roaming domain or between APs having same SSID MAC does not support inter domain roaming as it is layer 3 roaming and MAC is layer 3 unaware. This doesn t mean layer 3 roaming is impossible with , some upper layer solutions like Mobile IP are required. The roaming user can maintain application connectivity as long as its layer 3 address does not change[5], which does not change in layer 2 roaming. Here, in this paper we are concentrating on layer 2 roaming. In the Infrastructure-based network, when an MS moves away from associated AP, the signal strength and signal-to-noise ratio decreases and at some point it falls below predefined threshold. This situation triggers MS to initiate handoff. The MAC layer handoff process can be divided into three distinct phases: Probe, Authentication, and Reassociation phase. The probe phase can be further categorized into 2 subphases: Detection and Search phase. In detection phase, the MS is responsible for detecting the need for handoff. Whenever an MS detects lack of activity on its transceiver, the possible reasons for this could be a collision, temporary signal fading or AP out of range [11]. The MS has to decide the possible reason and whether or not to go for handoff. In the search phase, after deciding to make handoff the MS should search the potential AP to which it can associate. This could be done by the MAC layer scan function. Scanning can be accomplished in either Active or Passive mode. In active scan the MS starts scanning for AP by broadcasting a probe request frame on one channel and waits for the probe response frame till minchanneltime and if no probe responses are heard during this interval then assuming that no APs are present on this channel, it tunes its transceiver onto next channel and repeats the same procedure. Else if probe responses were received till minchanneltime then it further waits till maxchanneltime, waiting and collecting more probe responses. After collecting all the probe responses, it starts scanning the next channel. After scanning all the channels, process all the received probe responses to determine the potential AP. This scenario is illustrated in figure 2. In passive scan mode the MS tunes its transceiver on one channel and passively listens to the beacons transmitted by the APs on that channel and after beacon interval has elapsed move to next channel. Using the information Figure 2: The Handoff Process using Active Scan contained in the beacon frames the MS decides to connect to an AP. The MS has to wait on each channel for beacon interval as more than one AP may be operating on one channel. Also standard makes it compulsory to scan all channels, passive scan tends to lengthen probe phase. So passive scan is not suitable as well as recommended for faster handoff. After locating an AP in the range the MS should authenticate itself to the new AP by sending an Authentication Request frame. The AP either accepts or rejects the identity of MS by sending an authentication response frame. After being authenticated the MS tries to reassociate with the new AP by sending a reassociation request to the new AP. The new AP then sends a reassociation response back to MS indicating the acceptance or rejection and using Inter Access Point Protocol(IAPP), it communicates with old AP for MS credentials and any buffered frames. It has been already proved in [8] that authentication process takes negligible time. Reassociation phase delay is also very small as compared to probe delay. Probe delay makes up almost 90% of the total handoff latency. So here we are concentrating on improving the probe phase algorithm so as to reduce total handoff latency. 3 Related Work A lot of work has been done to reduce the handoff latency when roaming across the roaming domain. Many new schemes for Mobile IP have been proposed for roaming between Wireless LANs. Here the focus is on roaming within the roaming domain ie. layer 2 roaming. Sangho Shin and A.S. Rawat in [10] focused on reducing handoff latency by using selective scanning and caching mechanism. The idea is to scan selected channels instead of all available channels in a single active scan. A channel mask is used to select a specific sub-

3 set of channels to be scanned, thus reducing the probe latency. Caching is the other technique used to further reduce the probe latency. The results of active scan are cached. This cached results are used for future handoff. Next time whenever MS wants to make handoff, it tries to locate an AP from the cache and try to reassociate with it. If not successful then again start selective scanning. This procedure reduces the probe phase latency but in any case the active scan has to be executed completely at least once, resulting into longer latency during that scan. As well as the selected channels are to be scanned at a stretch, during a handoff, thus the resulting delay would be sum of scan delay of the selected channels. A selective channel scanning mechanism was proposed by Hye-Soo Kin and Sung-Jea Ko [4], in which MS need not scan all channels during active scan but only the channels selected by neighbour graph of APs. In this approach also active scan of selected channel is involved, where in the selected channels are to be scanned in a single scan. While designing any handoff algorithm, the higher error rate of WLAN should be taken into consideration. The IEEE MAC protocol recommends active scanning for handoff process. The probe phase delay may increase considerably if a probe request or probe response packet is lost on the transit. A reliable active scanning procedure is proposed by Wei Li, Qing-An Zeng and D.P. Agrawal in [6]. The algorithm detects the loss of probe request and retransmits it and it assigns higher priority to management frames as compared to data frames. Hector Velayos and Gunnar Karlsson in [11] focused on reducing the detection phase, search phase and execution phase. The detection phase detects the need for handoff. This phase can be reduced by reacting quickly to packet losses and by using shorter beacon interval, in case of passive scanning. They tried to reduce the search phase by using active scanning instead of passive scanning. A. Mishra and W. Arbaugh in [8] described the detailed handoff process. They experimentally measured various components of the handoff process and proved that probe phase contributes dominantly in overall handoff latency, making almost 90% of it, while authentication and association phase contributing in few milliseconds. Here in the proposed approach we try to reduce the probe phase latency. F. A. Gonzalez and Jesus A. Perez in [2], suggested an accurate strategy for measuring actual handoff latency. They distinguished handoff latency from disconnection time and showed that actual disconnection time is very less as compared to handoff time. Most of the work described above are based on complete or selective active scanning. There are certain drawbacks when using complete/selective active scanning. Complete/selective scanning needs to scan all the available/selected channels. Each channel scan adds channel scan delay to probe latency. A new algorithm is presented in section 4, which aims at resolving this drawback and further improving the probe phase latency. 4 The Proposed Approach In active scanning, MS has to spend minchanneltime or maxchanneltime on each channel, depending on channel activity. Thus the time spent on each channel may vary. In case of selective channel scan, the number of channels to be scanned may vary, resulting into variable latency. If N is the number of channels to be scanned, then Probe Time (P T ) could be delimited by N * minchanneltime < P T < N * maxchanneltime. The values of minchannel- Time and maxchanneltime is very crucial in calculating the total handoff latency. Two threshold power levels are defined viz Periodic threshold and Handoff threshold. If received power level of any frame (data or beacon), falls below Periodic threshold (T P ), the MS should start periodic active scan immediately. This is an indicator to the MS about the future handoff. Handoff threshold (T H ) is the power level, which when reached indicates that there is need to execute handoff procedure now. Handoff threshold (T H ) is defined in such a way that MS gets sufficient time to execute handoff procedure before leaving communication range of an AP. The communication range of an AP or MS is decided by the a threshold power level(t). The AP or MS drops a frame if it is received with power level below T. Propagation measurements in a mobile radio channel show that the average received signal strength at any point decays as a power law of distance of separation between transmitter and receiver[9]. The average received power P r at a distance d from the transmitting antenna is approximated by p r = p 0 ( d d 0 ) ple (1) where p 0 is received power level at a distance d 0 between transmitter and receiver. ple is the path loss exponent. 4.1 Estimation of T H & T P T H is the power level which when attained, the MS should start executing the authentication and association procedure with the selected AP. The procedure should be completed before leaving the coverage area of current AP. The maximum time taken for authentication and reassociation phase is calculated experimentally. This is the minimum time MS must spend within the communication range, after receiving frames with power T H. Assuming the maximum speed with which an MS can move, the distance (d) inside transmission range of an AP can be calculated as shown in figure 3. Power level at this point would be T H. Knowing the transmission range(d 0 ), threshold power level(t), and d, T H can be calculated using equation 1. The procedure is made adaptive by dynamically calculating the value of T P. T P is the periodic threshold below which MS should start scanning the channels periodically. It is a function of number of channels to be

4 Figure 3: Relationship between T, T H,and T P scanned periodically(maxchanallowed), the ScanPeriod, and the speed at which MS is moving away from AP. Its value should be set such that MS has sufficient time to periodically scan all channels to be scanned before reaching d h. The speed is estimated by measuring received power level of last 10 beacons and the time at which beacons are received. The distance inside d h, as shown in figure 3, at which periodic scan should start can be given by dist = speed (ScanP eriod+sd) maxchanallowed (2) where SD is the maximum time taken by MS to scan one channel. So the distance from AP where power level is T P can be given by dp = dh - dist. Using equation 1, T P can be calculated. If MS is moving with varying speed, the value of T P is updated at receipt of each beacon frame thus making the algorithm adaptive to the situation. 4.2 Algorithm Following are the notations used in the proposed algorithm. T H :- Handoff threshold T P :- Periodic threshold maxchanallowed :- Number of channels allowed to be scanned ScanPeriod :- Periodic scan interval channellist:- Array of preferred channels channelindex:- Index of channel scanned from channel- List array Algorithm 1 The Proposed Algorithm Initialization Phase: 1 Initialize channellist=1,6,11,2,3,4,5,7,8,9,10 2 channelindex = 0 Periodic Active Scan Phase: 3 Monitor received power of all frames including beacons 4 if (receivedpower < T P && receivedpower > T H) 5 Start active scan for channel with channellist index = channelindex 6 if(ap is detected till minchanneltime) 7 collect probe responses till maxchanneltime and 8 store entries for each AP in cache, sorted according to received power 9 associate back with previously associated AP 10 else 11 shift the channel to end of channellist 12 associate back with previously associated AP 13 channelindex = channelindex + 1 repeat steps 4-13 at an interval of ScanPeriod Reassociation Phase: 14 else if(receivedpower < T H) check cache 15 if(any entry present in cache for that AP) 16 try to reassociate with 1 st AP 17 if(succeeded) 18 completed 19 else 20 try to reassociate next AP until all entries are tried 21 else 22 start complete active scanning 23 Continuously receive beacon frames and keep updating T P. The algorithm can be divided into three phases as initialization phase, periodic active scan phase, and reassociation phase. In the initialization phase, each MS initializes a channellist for each AP, wherein the channels are arranged in order of preference. Initially the channel- List contains all the channels with channel 1,6, and 11 as the first three channel(as there are more chances of an AP present there). When MS moves away from AP, the Signal Strength(SS) of received frames decreases. MS continuously monitors the received power of all frames(data and beacons). When it falls below T P, MS starts periodic channel scan by scanning first channel from channellist. This channellist array is updated at every channel scan. If a channel is found idle during minchanneltime it gets shifted at the end of the list. Thus building a list of channels in channellist, arranged in the order of preference. The MS stores the positive results of periodic channel scan in cache. The AP cache consists of a table which uses current MAC address as the key. Corresponding to each entry in the table the MAC of adjascent AP alongwith the channel is stored. Other fields are the received Signal Strength and SSID. The AP cache has a size of eight. We are limiting the entries per AP to eight, meaning that for each AP it can store details of eight adjascent APs. The cache entries are invalidated as as soon a handoff occurs. After minchanneltime or maxchanneltime, depending on presence of an AP on that channel, the MS again associates back to the previously associated AP, thus resuming the communication from where it had stopped. The breakage of communication due to the this periodic channel scan is either minchanneltime or maxchanneltime plus the reassociation delay, needed to associate back to previous AP. The suggested optimal values, deduced from previous experiments, are approximately 6.5ms for min- ChannelTime and 11ms for maxchanneltime [12]. The simulation results shows that the time taken to reassoci-

5 ate is approximately 3-4ms. Thus the latency generated from this would be less than the jitter bearable by most jitter critical application like VoIP. (The maximum unnoticeable jitter for VoIP is 40ms [7]). The above procedure is repeated periodically untill specified number of channels (MaxChanAllowed) are scanned one by one or the handoff threshold T H has reached. When the received power level of beacons or data packets falls below T H, the MS makes use of the cache to find an AP with maximum power level and tries to associate with it. If association is successful, the to and fro traffic will be routed through the new AP. In case of failure to associate the MS tries to look for next best AP and try reassociating with it. This procedure is repeated untill all APs in the cache are tried. from that AP now. So it finds the next best AP and associates with it. Thus increasing the delay. The actual 5 Simulation We have chosen OMNeT++ as simulation environment. The reasons for choosing OMNeT++ can be found in [3]. Above OMNeT++, IPv6SuiteWithINET model containing IEEE WLAN MAC layer implementation is being used. The simulation scenario has three APs, connected through a distribution system. The parameters used for the simulations are summarized below: Parameter Value Threshold Power -96 dbm Handoff Threshold Power dbm minchanneltime 6.5 ms maxchanneltime 11 ms maxchanallowed 11 When MS initially boots up, using complete active scan, it associates with an AP. During the movement, it initiates a handoff. The Handoff time is measured as difference between the time at which received power of MS falls below T H and the time at which reassociation response from new AP is received. During this time the MS retrieves best AP entry from the cache and authenticates itself with the found AP. Also it sends reassociation request to AP and waits for reassociation response. The packet delay can be measured by sending a stream of data from MS to a wired station while it moves from one AP to another. This is done by using a ping application, sending ICMP requests at regular intervals. The round trip time of response packets were traced to find out the packet delay due to periodic channel scan and handoff. 6 Results and Analysis The main result found is that the MS will not perceive any interruption in applications like VoIP as the handoff time obtained from the simulation results is very less compared to acceptable jitter. The handoff time is approximately 3 ms on most occasion except one in figure 4. During that handoff, the MS tries to associate with best AP from cache but failed to do so, as it has moved away Figure 4: Handoff Delay break in communication would be more than the handoff time while scanning individual channels periodically. The actual time taken for scanning a channel is highly dependant on minchanneltime and maxchanneltime. Practically the values of minchanneltime and maxchannel- Time would vary for different vendors, thus varying the communication break. By taking the optimal values of minchanneltime and maxchanneltime this delay can be controlled. The results are demonstrated in 5. Figure 5: Channel Scan Delay due to Periodic Scan The packet delay is measured as the round trip time of ping request, sent from MS to a wired station. The minimum delay occurs during normal association, and it seems to be around 0.4 ms. Maximum delay occurs during handoff or periodic channel scan, so it is always around 11 ms or more. The maximum delay recorded is ms, which is due to periodic active scan for a channel on which AP is present. The readings are taken as minimum, average and maximum delay for 6 experiments and average of these experiments as shown in figure 6.

6 [5] LEARY, J., AND ROSHAN, P. Wireless LAN Fundamentals: Mobility. Cisco Press, January [6] LI, W., ZENG, Q.-A., AND AGRAWAL, D. P. A reliable active scanning scheme for the ieee mac layer handoff. In IEEE Radio and Wireless Conference, RAWCON 2003 (Hilton Boston Logan Airport, Aug ). [7] MIRAS, D. A survey on network qos needs of advanced internet applications. INTERNET2 - QOS WORKING GROUP (November 2002). Figure 6: Packet Delay 7 Conclusion and Future Scope We have measured, analyzed and suggested means to reduce the MAC layer handoff latency by using periodic channel scan and caching the results, instead of every time scanning channels actively. We have shown that using periodic scan, the possible APs for future handoff can be stored in cache. Cache access and associating with AP, found from cache, is very fast resulting into very small handoff latency. Also by taking periodic threshold as function of speed and number of channels to be scanned periodically, the algorithm is adaptive to the situation. The AP prediction can be further improved by considering direction of movement instead of only power level and maintaining the results in cache throughout lifetime of MS. The handoff time can be further reduced by using pre-authentication as the authentication algorithm. The algorithm can be made more adaptive by dynamically choosing the number of channels to be scanned periodically as all the channels with APs present on it would be placed near the head in the channellist after one periodic scan. [8] MISHRA, M. S. A., AND ARBAUGH, W. An empirical analysis of the ieee mac layer handoff process. ACM SIGCOMM Computer Communication Review (April 2003). [9] RAPPAPORT, T. Wireless Communications, second ed. Prentice Hall India, New Delhi, [10] SHIN, S., RAWAT, A., AND SCHULZRINNE, H. Reducing mac layer handoff latency in ieee wireless lans. Mobiwac 04 (October 2004). [11] VELAYOS, H., AND KARLSSON, G. Techniques to reduce ieee b mac layer handover time. Tech. rep., Royal Institute of Technology, [12] WAHARTE, S., RITZENTHALER, K., AND BOUTABA, R. Selective active scanning for fast handoff in wlan using sensor networks. References [1] GAST, M wireless networks, the definitive guide. O Reilly and Associates, USA, [2] GONZALEZ, F. A., PEREZ, J. A., AND ZARATE, V. H. Hams: Layer 2 handoff accurate measurement strategy in wlans st IEEE International Workshop on Wireless Network Measurements(WiNMee 2005) (April 2005). [3] J.LAI, E.WU, A.VARGA, SEKERCIOUGLU, A., AND EGAN, G. A simulation suite for accurate modeling of ipv6 protocols. Proceedings of the 2nd Interanational OMNet++ Workshop (January 2002). [4] KIM, H.-S., AND KO, S.-J. Selective channel scanning for fast handoff in wireless lan using neighbor graph. ITC-CSCC2004.