In recent years, wireless LANs have been widely deployed for public mobile Internet services. Public wireless LANs can provide high-speed Internet connectivity using portable devices such as laptop computers and personal digital assistants (PDA).
An emerging application to wireless LAN system could be Voice over IP (VoIP), which requires networks to support real-time transmission such as bounded latency and jitter. One of the key issues for VoIP application to support real time servise is providing the Quality of Service (QoS) that meets the user expectations. This is recognized as the single biggest challenge in providing high-quality voice on wired IP-based networks. In wireless networks, one of the major issues affecting QoS is to minimize the service disruption during handoffs of the mobile nodes.
It is recognized that an intra-subnet handoff across access points (APs) in an administrative domain, so-called layer-2 handoff, normally brings a service disruption of several hundreds of mini-seconds (ms), and an inter-subnet handoff across different domains, so-called layer-3 handoff, would be disrupted in periods of several seconds.
However, to meet the requirement of real-time services, the maximum latency would be limited to less than 100 ms. Therefore, lots of investigation devoted to minimizing the handoff latency, which is defined as the period while a mobile node is unable to receive application traffic due to the handoff.
Among the researches focusing on minimization handoff latency, schemes based on prediction of next AP for a layer-2 handoff, or next access router (AR) for a layer-3 handoff are proposed in lots of literatures. Neighbor graphs, which are defined as a specific data structure that stores information of set of neighboring APs or/and ARs with respect to current AP or/and AR, are one of major approaches to predict
next APs or/and ARs to handoff to. With aids of neighbor graphs, mobile nodes in a layer-2 handoff will probe the channels in which at the least one of the neighboring APs exists. Moreover, mobile nodes can terminate probing the channel immediately after the last neighboring AP has responded rather than wait until maximum channel time. Therefore, the probe phase in layer-2 handoffs can be significantly shortened with the help of neighbor graphs [1].
The reassociation latency in layer-2 handoffs can be reduced an order by the scheme that the current AP caches secure information of mobile nodes to all of the APs in neighbor graphs prior to occurrence of layer-2 handoffs [2]. The 802.1x reauthentication to new AP would also bring significant latency to layer-2 handoffs.
By means of proactive key distribution to all of the APs in neighbor graphs are reported to effectively eliminate the 802.1x reauthentication latency from 800 ms up to 10 ms [3].
The layer-3 latency can be effectively reduced to a level of layer-2 handoffs with aids of neighbor graphs. The neighbor-casting scheme sets up temporary links from current AR to all ARs in neighbor graphs in order to maintain the connection to the mobile node via the relay of current AP to next AP in a layer-3 handoff. Therefore, the latency induced by a layer-3 handoff is eventually reduced to the same level as disruptions caused by the layer-2 handoff [4].
The neighbor graphs include information of all possible APs/ARs that a mobile may handoff to from the current AP/AR. The size of a neighbor graph (i.e., number of APs/ARs in a neighbor graph) may be large enough to cause severe overhead to wired networks. Size of a neighbor graph reflects the number of potential APs to be probed for a mobile node to complete the probe phase in a layer-2 fast handoff. In both context caching and neighbor-casting schemes, the context of the mobile node and the arriving packets for the mobile node during a layer-3 handoff have to be duplicated as
many copies as the number of APs/ARs in neighbor graphs. Therefore, the size of a neighbor graph stands for the overhead to wired bandwidth in a fast handoff scheme.
In proactive key distribution scheme, the pairwise master keys (PMK) are generated and distributed to all APs in neighbor graphs. The size of a neighbor graph thus responds to the overhead and the security risk caused by abuse of key distribution.
To reduce the size of a neighbor graph, some schemes such as selective neighbor caching [5] and frequent handoff region (FHR) [6] are proposed. Moreover, global positioning system (GPS) is proposed to update the topology information of a mobile node dynamically for the sake of predicting the next AP among APs, in a neighbor graph before a handoff is initiated. Based on the assumption that next AP and next AR are known prior to the occurrence of a handoff with aids of a neighbor graph and real time topology information, a cross-layer fast handoff scheme is proposed to minimize overall latency for a layer-2 and layer-3 handoff to less than 50 ms in the experiment results [7].
It is generally recommended to configure a neighbor graph with a distributed manner such that each AP/AR stores and maintains its own neighbor graph. The neighbor graph can be automatically learned by the individual AP/AR with the reassociation frames and registration packets from various mobile nodes. Distributed neighbor graphs require the supports from neighboring APs/ARs as well as the mobile nodes roaming through the coverage of APs/ARs.
In fact, not all of the mobile nodes have to invoke for seamless handoffs. For example, a laptop with a wireless network interface card (NIC) may be considered as a mobile node since it may be carried from the coverage area of an AP/AR to another.
With its volume and weight, the user may seldom play real time application (such as VoIP) while the laptop is being carried across different wireless LAN. Therefore, seamless handoffs may be not a mandatory function for a laptop with wireless NIC
unless the notebook is especially designed for used in a vehicle. A hand carried device such as personal digital assistants (PDA) may run only in non-real-time applications, such as web browser or specific remote database queries. A hand held device for non-real-time applications may neither need feature of seamless handoffs since disruptions of few hundreds mini-seconds is generally tolerable to users. A typical example that really invokes for seamless handoffs is recognized as media streaming, such as a WiFi VoIP, with both attributes of hand held devices and real-time application. Since only particular clients request for seamless handoffs, neighbor graph scheme may induce unnecessary overhead by non-seamless-handoff nodes because neighbor graph scheme is implemented in AP that provides universal service to all of its clients. For example, a stationary station may work as a client of an AR with neighbor graph function. By ignoring the non-seamless-handoff attribute, an AR may buffer and forward the packets to all its neighboring ARs in case a sudden termination from the client is misinterpreted as an event of handoff, so that misinterpretation as well as bandwidth waste may occur.
Millions of IEEE 802.11 APs/ARs as well as mobile nodes have been deployed without supports to neighbor graphs. One of the reasons for the popularity of IEEE 802.11 wireless LAN is due to the features of simple configuration and low cost. To support neighbor graph mechanism, unfortunately, increases complexity and cost to re-configure an IEEE 802.11 wireless LAN. Moreover, a new protocol will be required to enable cooperation among APs/ARs and mobile nodes though some of mobile nodes do not require seamless handoffs from wireless circumstance.
The factors described above somehow explain the difficulty for neighbor graph mechanism to be implemented in existed wireless networks, though researches have proven with experiments that seamless inter-subnet handoffs can be eventually achieved with integral fast handoff schemes benefit by neighbor graph mechanism.
In this thesis, the so-called “discrete scan”, a simple and effective scheme to collect the set of potential next APs in a layer-2 handoff is proposed. A mobile node with discrete scan scheme executes the process of passive scan a certain time ahead of handoff initiation. To avoid disruptions lasting longer than 50 ms, process of passive scan is divided into smaller pieces of sniffing periods; therefore, the probe delay in layer-2 handoff is decomposed into pieces of disruption that users are unable to sense.
Discrete scan is designed to implement in a media streaming application such as WiFi VoIP that typically desires function of seamless handoffs. Benefit by features of dedicated hardware and relative low working bit rate (usually less than 100 Kbps bidirectional) compared with its network interface card (NIC) (11Mbps), discrete scan scheme can take place of neighbor graphs without collaboration from APs as well as other mobile nodes.
The characteristic that a mobile node spends its own effort to predict the next AP to fasten its handoff may allow the seamless handoff schemes to utilize to discrete scan. Without any support from AP, the mobile nodes using discrete scan can perform a fast handoff by elimination of probe latency from a layer-2 handoff. The confusion for an AP to provide a seamless handoff service to a node without such demand can be avoided because discrete scan will be only implemented in the nodes that may invoke seamless handoffs.
Compared with neighbor graphs, the set size of candidate next-APs collected by discrete scan scheme is much smaller than that by neighbor graphs. As shown in Figure 1.1, discrete scan scheme discovers only a subset of neighboring APs of which signals can reach to the mobile node. Taking advantage of examined information from frames received in discrete scan, such as received signal strength (RSS), MAC address and network allocation vector (NAV), several mechanisms to assist the mobile node to select the nearest or/and the best AP among the candidates is
introduced in this thesis.
With discrete scan as well as its auluxiary mechansims, the mobile node is able to select the most appropriate AP to handoff to before handoff is initiated. Thus, latency induced by probe phase in a layer-2 handoff is significantly reduced to a ProbeRequest from the mobile node and a ProbeResponse from the selected AP, without modification to the existed infrastructure of IEEE 802.11 wireless LANs.
Since the latency spent in probe phase is recognized as more than 90% of overall layer-2 handoff latency, discrete scan with its auluxiary mechansims expect to reduce 90% of latency in a layer-2 handoff. With such significant improvement but without modification required to the current system, the realization of discrete scan scheme for a mobile node is reasonably much easier than that of neighbor graph scheme.
Set of neighboring APs
Current AP
Candidate next-APs
discovered by discrete scan
Figure 1.1 Set of neighboring APs and set of next-AP candidates in discrete scan.
Discrete scan is supposed to support seamless layer-3 handoffs with the fast handoff schemes proposed for neighbor graphs. Substituting for the functions of a
neighbor graph and its extensions, discrete scan with auxiliary mechanisms proposed in this thesis helps a mobile node determine the next AP in advance of the trigger of a handoff. The only insufficiency for discrete scan scheme to take place of a neighbor graph lays on the lack of the ability to discover next AR because discrete scan is absolutely designed for layer-2 operation. However, a lookup table built-in the current AP, that outputs next AR with input of the next AP, will simply address the problem.
One of major concerns on discrete scan scheme is its impact on the service QoS received in the periods of discrete scan, because discrete scan decomposes the latency of probe phase in a layer-2 handoff into pieces. Benefit by the high transmission priority of VoIP traffic in wireless EDCA environment, the simulation results in this thesis show that QoS degradation due to discrete scan is controlled within a tolerable value for a real-time application.
The rest of this thesis is organized as follows. Chapter 2 introduces the background of latency in layer-2 and layer-3 handoffs. The relative works involved in neighbor graphs are introduced briefly in this chapter as well. In Chapter 3, we elaborate discrete scan scheme and the mechanisms to select next AP. Besides, in Chapter 3, the performance and impact on service QoS are evaluated by numerical results based on analytical model of DCF wireless LAN environment, and those in EDCA environment are demonstrated with simulation results in Chapter 4. Finally, the conclusion and future works are presented in Chapter 5.