Adaptive Region-Based Location Management for
PCS Systems
Shou-Chih Lo and Arbee L. P. Chen, Member, IEEE
Abstract—The personal communications services (PCSs) systems can provide ubiquitous and customized services. The key issue, which affects the performance of the whole system, is the location management. In this paper, we propose a region-based location strategy by taking advantage of the user’s movement behavior to improve the performance of the conventional systems. Each mobile user is associated with a set of regions, which are derived from the user’s movement patterns. The registration processes in the same region can be eliminated such that the cost of location management can be significantly reduced. Several design issues are studied by considering the workload balance and the call-to-mobility ratio for a user. The proposed strategy can be dynamically adjusted based on different system parameters and user behaviors. A performance analysis on the signaling cost and the database access cost is given to justify the benefits of this approach.
Index Terms—Location management, location registration, loca-tion tracking, personal communicaloca-tions service (PCS).
I. INTRODUCTION
T
ELEPHONE services have brought great convenience to our life. The communication quality and service contents have been extensively enhanced with technology advances in telecommunications. Personal communications services (PCSs), with the characteristics of personalized and ubiquitous services, have become an interesting topic [7]. In the PCS environment, the geographic area is divided into adjacent cells, each covered by a base station. A mobile user with a handset can connect to the local base station via a wireless channel to receive or send any message.Location management is an important task in order to com-municate with the user, who may move from one cell to another at anytime. There are two processes for location management: registration and location tracking (or call delivery). Registration corresponds to the process of updating the data about the user’s location, while location tracking corresponds to the process of searching the user’s current location by the registration data. In general, several cells constitute a location area (LA); a registra-tion is required as a user crosses the boundary of the LA. The
mobile switch center (MSC), which can be viewed as a bridge
to connect the wireless network and the wired network, may manage several LAs. To simplify the discussion, we assume there is only one LA within an MSC. The location management defined in IS-41 and the global system for mobile communica-tion (GSM) standards adopts a two-level hierarchical
architec-Manuscript received November 8, 1999; revised October 23, 2001. S.-C. Lo is with the Computer and Communication Research Center, National Tsing Hua University, Hsinchu, Taiwan, R.O.C.
A. L. P. Chen is with the Department of Computer Science and Information Engineering, National Dong Hwa University, Hualien, Taiwan, R.O.C.
Publisher Item Identifier S 0018-9545(02)02508-2.
ture, which is composed of a centralized home location register (HLR) and multiple visitor location registers (VLRs) [20]. The VLR, which is usually coupled with an MSC, contains loca-tion informaloca-tion for the users in its service area, while the HLR contains permanent information for all users and pointers to the current serving VLRs of the users. Any reader interested in the current and proposed protocols for location management in var-ious wireless systems is referred to [2].
In this paper, we consider the problem of how to efficiently locate a user under an environment characterized by the fol-lowing properties: small cells and high user density. This paper focuses on a special class of users whose movements follow cer-tain patterns. The main idea of our method is to form some re-gions for each user by grouping the LAs they frequently visit. The registration cost can be significantly reduced by decreasing the registration frequency as the user moves in the same region. We will discuss the suitable condition to put the restraint on the registration and provide a mechanism for the location tracking. Our primary contributions in this paper are as follows: we pro-pose an efficient location strategy by taking advantage of the user’s movement behavior; we construct a distributed HLR in-stead of a centralized HLR to balance the workload; moreover, we consider the ratio of the call arrival rate to the mobility rate [or call-to-mobility ratio (CMR)] for reducing the registration cost. The proposed strategy can be dynamically adjusted based on the number of regions, the degree of user mobility, and the system parameters such as the signaling cost between HLR and VLR.
The rest of this paper is organized as follows. In Section II, we present the location management defined in the IS-41 and give a brief introduction on other approaches. Section III presents the design for the region-based strategy in detail. In Section IV, the benefits of this method over IS-41 are indicated by performance analysis. Finally, we present the conclusion and future work in Section V.
II. RELATEDWORK
Location management is an important issue in PCS systems for efficiently locating a mobile user. In the IS-41 standard [8], the location information is managed by the HLR and the VLR. The respective flows for the processes of registration and loca-tion tracking are depicted in Fig. 1.
The identification number of VLR is periodically broadcast in each LA for assisting the user to judge whether a new LA is entered. If it is the case, a registration message containing the user’s mobile identification number (MIN) is sent to the new VLR. Then the registration message is forwarded to the HLR. At the HLR, another message is sent to the old VLR to reclaim 0018-9545/02$17.00 © 2002 IEEE
(a)
(b)
Fig. 1. Location management in the IS-41. (a) Registration process. (b) Location tracking process.
previously occupied resources (referred to as the deregistration process). When a call setup is going to be established, a re-quest to query the callee’s location is issued from the MSC to the HLR, then forwarded to the VLR where the callee currently visits. A routing number, such as a temporary location direc-tory number (TLDN) in the IS-41 or a mobile station roaming number (MSRN) in the GSM, is returned. Finally, a trunk is es-tablished by using the routing number.
The location management specified in the IS-41 standard has several disadvantages. The HLR becomes a bottleneck of the network system. Also, the registration process may incur a big waste for a user frequently crossing LAs while seldom receiving calls. This situation can be stated as the user’s having a low CMR. The CMR is defined as the number of calls to the user di-vided by the number of times the user crosses the LA boundary in a time unit. When the CMR is low, the number of registra-tion processes performed by the user is more than the number of location tracking processes performed by the system to lo-cate the user. An efficient location management strategy should minimize the costs of registration and location tracking. How-ever, the two costs have a tradeoff. The more location informa-tion we record (i.e., increase the frequency of registrainforma-tion), the less location tracking cost we need to pay and vice versa. The improvement on the location management can be considered in three aspects: the registration frequency, the registration cost, and the location tracking cost.
A. Techniques for Reducing the Registration Frequency
The number of registrations can be controlled by performing the registration process only when a certain quantity is reached. As illustrated in [1] and [6], the quantity may be measured in time, the number of moves, or the moving distance. These methods have either poor performance or high complexity in im-plementation. Another possibility is to define the region (group of LAs) and then perform the registration process as the user enters a new region. In [5], the region is predefined, which con-sists of the neighboring cells of the reporting center selected from the base stations. However, an optimal selection of the
reporting centers is an nameplate (NP)-complete problem in a hexagonal network configuration. A heuristic algorithm for se-lecting reporting centers was proposed in [10]. In [9], the region is preallocated by separating the PCS service area into several uniform groups. This approach has worse performance when a user frequently crosses the region boundary. In [26], the size of a region can be dynamically adjusted for a user according to the CMR value. Nevertheless, the region can only be shrunk or ex-panded in the form of a square shape. In [16], the optimal size and shape of a region is studied under a one-dimensional wire-less environment. In [19], the region is composed of the current and the previous LAs that the user has visited. This method is suitable only for users moving back and forth crossing two LAs. In [4] and [11], the region is generated by partitioning the movement patterns into groups. Hierarchical location servers are built over the regions. The search locating a user is per-formed either by the top–down or the bottom–up method to tra-verse the location servers. However, the complete signaling for the location management is not presented. Location manage-ment in the case where each user has only one region was ad-dressed in [21] and [24]. The region is defined as a list of LAs where the user is most likely to be. Registration is performed when a user enters an LA not recorded in the list. Since the con-tent of the list may change, retransmission of the list is required. Moreover, the search cost in a large region will be very high. In our approach, multiple regions are generated as in [11]. How-ever, a different search method to locate a user is adopted. Fur-thermore, we reduce not only the number of registrations but also the cost of a registration by using the CMR value.
B. Techniques for Reducing the Registration Cost
To reduce the registration cost, the technique of forwarding
pointer, by which the registration message is directed to the last
visited VLR, was introduced in [13]. Implicit deregistration was studied in [17], where the obsolete resource (such as TLDN or MSRN) is reused when no free resource is available.
C. Techniques for Reducing the Location Tracking Cost
For reducing the location tracking cost, typical techniques in-clude caching the search result of any outgoing call from an MSC for later use [12], replicating the location data in multiple VLRs [23], and constructing the location databases into a hier-archical architecture [3], [15], [25]. The search to locate a user in a region can be performed by broadcasting, by following the forwarding pointers that direct the moving tracks, or by sequen-tially searching each area in the order of probabilities [11]. The last method (named the list method) is better than others in most cases. The list method is adopted in our strategy for the location search. More complex search mechanisms were studied in [22].
III. REGION-BASEDLOCATIONSTRATEGY
In this section, a region-based location strategy is presented. First, we discuss various issues in the strategy design. Then, the architecture and the processing flows of location management for our proposed method are discussed.
A. Design Issues
1) The Regions: Two types of regions are defined based on a
user’s movement pattern. The hot region is a group of LAs that corresponds to a partition of the user’s movement pattern [11]. The PCS service area excepting the hot regions constitutes a single
cold region. Note that the regions are disjointed from each other.
We have proposed four strategies to derive the regions from the movementpatterns in [14]. To reflect the changes in user behavior, a simple method is to rebuild the regions periodically.
The location data are recorded by the following formats. The data about which region the user is in are recorded at the HLR, while the data about which LA in the region the user is in are recorded at a server associated with the region. When the user crosses the region boundary, both of these data are updated. To re-duce the registration cost, we decrease the registration frequency as the user crosses the LA boundary in the same region. In this situ-ation, a search within a region is needed to locate the user. To speed up the search, each LA in the region can be associated with a value that denotes the probability that the user may be there. Hence, the search is performed by sequentially searching each LA in the re-gion in the order of probabilities. The probability information can be either provided by the user or estimated by the user’s calling history. In certain conditions, such as when the user is in the cold region or with a high CMR, the search will be expensive. In these conditions, we perform the registration when a user crosses the LA boundary in the cold region or crosses the LA boundary in the hot region when the CMR is high.
2) The Options of the Registration: The location data at
the HLR are usually updated when the user crosses the region boundary. This update cost is high in the PCS system. For some location data in the HLR, they may not be used for location tracking when the mobility rate is much higher than the call arrival rate (i.e., CMR is low). Therefore, we can eliminate the registration at the HLR when the CMR of the user is low as a means of reducing the registration cost. That is, the registration message is first sent to the server of the newly entered region, and then the message is optionally sent to the HLR based on the current CMR value. If the location data at the HLR are not up to date, the location tracking should be performed by first sequentially searching each region of the user. Again, each region can be associated with a probability value to speed up the search.
3) The Distributed HLR: A centralized HLR can become a
bottleneck of the network system, which causes more call block-ings. It is intuitive to build several distributed HLRs with an appropriate data allocation [18]. Although full data replication can balance the workload on querying the HLR for location tracking, the consistency maintenance becomes a problem, es-pecially when a large volume of location updates is issued. An-other choice is to partition the HLR. In this case, a function like
global title translation (GTT) to decide which partition of the
HLR stores the location data, given the identifier of a mobile user, should be supported. This model is adopted in our strategy.
B. System Architecture
According to the above design principles, we propose a three-level hierarchical architecture of location management. The log-ical view of our proposed architecture is shown in Fig. 2.
Fig. 2. Three-level hierarchical architecture.
Each square at the bottom level of the architecture represents a VLR (or an LA). The distribution of regions for a user is also shown in the figure. Note that each user has its own set of re-gions. In the middle level of the architecture, each square rep-resents a region server (RS). Each the user’s region is managed by the nearest RS, as indicated by the vertical line in the figure. For example, hot regions 1, 2, and 3 are managed by RS , RS , and RS , respectively, in the figure, while the cold region is managed by RS . Generally, an RS may manage several re-gions of different users. The HLR is partitioned, and each par-tition is managed by a profile server (PS) in the upper level of the architecture. The user’s permanent information is stored in one of these PSs. The PS where the user’s permanent informa-tion is stored is called the user’s home PS. The placement of the PS depends on the population of mobile users with denser areas having more PSs.
Here, we give a brief description of how to perform the pro-cesses of registration and location tracking. Three movement cases in Fig. 2 are considered. When a user enters a new LA, the registration process is performed as follows. It should be de-cided first whether the user stays in the same region. If the user is still in the cold region (the case of movement 1), a registra-tion message is sent to the RS of the cold region (RS ), which indicates the user’s current residing LA. A further notification to the home PS (say, PS ) is needless. If the user is still in a hot region (the case of movement 2), the registration message is only sent to the RS of the current region (RS ) when the CMR is high, but no registration is performed when the CMR is low. If the user crosses any region boundary (the case of movement 3), a registration message is sent to the RS of the newly entered region (RS ). If the current CMR value is higher than the prede-fined threshold value, the RS then sends a registration message to the home PS (PS ), which indicates the user’s current serving RS. A record cancellation message is also sent from the new RS (RS ) to the old one (RS ).
Location tracking proceeds by first querying the home PS of the callee. From the PS, the current serving RS of the callee is found by first searching the recent serving RS (which is recorded in the PS) and then other RSs in the order of probability. The found RS then decides the current residing LA of the callee by searching first the recent residing LA (which is recorded in the RS) and then other LAs in the order of probabilities. The
se-TABLE I PS TABLE TABLE II RS TABLE TABLE III HANDSETTABLE
quential search is due to the optional registration at the PS and the RS.
In Section IV, we will do a performance analysis to show when it is better not to perform registration at the RS and the PS. As we can see, the CMR value of each user is an important basis to adjust the performance of the proposed strategy. To obtain this information, we can implement two counters that record the number of incoming calls and the number of movements for a user.
C. The Location Databases
We introduce the data structure used in our region-based lo-cation strategy. Four tables are given as follows.
The PS table, as shown in Table I, is stored in the PS. The recent RS field records the RS the user recently registers. The associated RSs field records a list of RSs belonging to the user. Each RS in the list is also associated with a value that denotes the probability that the user may be there. The RS table, as shown in Table II, is stored in the RS. The recent VLR field records the VLR in the region the user recently registers. Similarly, the associated VLRs field records a list of VLRs for the LAs in the region with probability values. The home PS field records the user’s home PS. The CMR history field describes the condition when the registration at the PS is not needed. The condition is described either by time periods or a value as the threshold.
In the user’s handset, the handset table, as shown in Table III, is stored. The second and third fields record the last visited VLR and RS, respectively. The CMR history field describes the con-dition when the registration at the RS is not needed. The region
table, which records the information about the regions of a user,
is also stored in the user’s handset. The data structure is shown in Table IV. Each region has a region number, a region server, and a list of VLRs for the LAs in this region. The last field for the cold region (numbered zero) is empty, as it is useless. The region table is used to identify the current residing region of the user, given the user’s current serving VLR.
TABLE IV REGIONTABLE
Fig. 3. A B-tree index for the region table.
A large region table will incur significant overhead in iden-tifying the region. To reduce this cost, a B-tree index structure can be constructed on the associated VLRs field in Table IV, as demonstrated in Fig. 3. Each VLR entry in the index structure constitutes a search-key value and a pointer for each search key. This pointer points to one of the entries in the simplified region table, where only the region number and the region server are recorded. The search proceeds by using the current VLR iden-tification (ID) as the search key. If the search is successful, one hot region is found. Otherwise, the user is in the cold region.
D. Processing Flows
The processing flows of the registration and location tracking of the proposed strategy are illustrated.
1) Location Registration: When a user enters a new LA, a
new VLR ID will be received and recorded in the field of last vis-ited VLR in the handset table. This ID will be compared with the associated VLRs of the regions in the region table for deciding to which region the newly entered LA belongs. If the user is in the same hot region and the recent CMR value is lower than the threshold recorded in the CMR history field of the handset table, no registration is performed. Otherwise, the registration process is performed as shown in Fig. 4. The steps are illustrated as fol-lows.
1) The handset sends the message MIN, old RS, current RS to the newly entered VLR. The first two fields in the message are acquired from the handset table, while the last field is the result of the comparison at the region table.
2) The newly entered VLR updates its database, indicating that the user is now staying in its coverage area. Then the message MIN, old RS, new VLR is sent to the current serving RS, as indicated by the current RS in the incoming message.
(a)
(b)
Fig. 4. The flow of registration. (a) Intraregion registration process. (b) Interregion registration process.
3) The current serving RS updates the recent VLR field, in-dicating the new VLR in the RS table for the user. Also, the current serving RS checks whether the user stays in the same region by referring to the old RS in the incoming message. Case 1 (intraregion registration): the user stays in the same region. Here, the old RS and the current RS in the message are the same. An acknowledgment is trans-mitted backward to the handset. The registration process is then finished. Case 2 (interregion registration): the user stays in a different region. Go to Step 4).
4) The current serving RS sends the message MIN, current RS to the home PS if the user’s CMR is higher than the threshold recorded in the CMR history field of the RS table. The address of the home PS is acquired by the home PS field.
5) The current serving RS sends a resource cancellation message to the last serving RS, as indicated by the old RS in the incoming message.
6) An acknowledgment is transmitted backward to the handset. The handset then updates the last visited RS field to the new value in the handset table.
The resource cancellation message to the last visited VLR is not shown in the above processing flow because we assume that the implicit deregistration process [17] is operated. The implicit deregistration process is really needed for the case where the user enters a new VLR but no registration is performed.
2) Location Tracking: Location tracking is performed as
shown in Fig. 5. The steps are illustrated as follows.
1) The handset sends the dial number (DN) of the callee to the current serving MSC.
2) The MSC then transmits the DN to a fixed switch where the global title translation is performed to get the address of the callee’s home PS.
3) The request for querying the callee’s location is for-warded to the home PS.
Fig. 5. The flow of location tracking.
4) The home PS searches for the current serving RS of the callee based on the PS table (the recent RS and associated RSs fields).
5) The RS then finds the current residing LA of the callee based on the RS table (the recent VLR and associated VLRs fields). The MSC should issue the paging message to confirm that the callee is in its coverage area.
6) If the callee is reachable in an LA, the routing number (RN) is then returned to the current serving MSC of the caller.
7) Finally, a trunk is established based on the RN. IV. PERFORMANCEANALYSIS
To evaluate the performance of the proposed strategy, we use an analytical model to explore the costs of registration and lo-cation tracking. We consider both the cost for transmitting sig-naling messages and the cost of database access in the analysis. As in [24], the analysis is performed for one user because the ef-fectiveness of this strategy is dominated by personal behaviors. Experiments are performed to compare the cost of the proposed strategy with that of the strategy defined in IS-41.
A. Performance Environment
Assume that all hot regions are of the same size and that there are LAs per hot region. The number of hot regions is de-termined by giving the total number of LAs in all hot regions and the region size in units of LAs. To reduce the complexity in the analysis, we assume that all hot regions are separate. The switches functioning as the GTT are colocated at the PSs.
We measure the signaling cost for a specified communication link by its delay value. Assume that these values can be acquired by a table lookup method. In Table V, we denote the delay value for each communication link as a variable. The item paging in the table denotes the cost to page a user in an LA. We define , , , and to be the costs for updating or querying the VLR, the RS, the PS, and the centralized HLR, respectively. We assume that in our analysis. The cost for per-forming the GTT is denoted as .
We introduce parameters for modeling the movement be-havior of the user in the following:
number of LAs in all hot regions; number of LAs in a hot region;
number of hot regions (that is, );
probability that the user is in region (note that region 0
TABLE V SIGNALINGCOSTNOTATIONS
Fig. 6. The user’s movement model.
probability that the user moves in the same region ; probability that the user is in LA of a region; assume
.
The movement behavior can be modeled as a state transition di-agram (see Fig. 6). The user may move in the same region or into another region. If the user is in the cold region, we assume that it has the same probability as the user’s moving into any hot region. Since hot regions are supposed separate, crossing between hot re-gions is impossible. Each node in the diagram represents a region, and each edge represents a movement. Symbols in the node and on the edge denote the probability values of a user’s being in the cor-responding region and the movement, respectively. If the user has the same behavior in a certain time period, we can view the move-ment transition as a steady state. By the flow balance for each state, we can derive the equations shown in
(1)
The performance measurement is based on the total signaling cost and the total database access cost spent on the registra-tion and locaregistra-tion tracking. Define and to be the average number of incoming calls to the user per unit time and the av-erage number of times the user changes LA per unit time, re-spectively. CMR can be expressed as . Let and denote the costs for registration and location tracking, respec-tively. The total cost per time unit can be computed by the
ex-pression . We separate the cost evaluation for
the proposed strategy into four cases according to the option of registration at RS and PS as listed in Table VI. In the following, we list the expressions for evaluating the costs in these cases:
1) RB1
TABLE VI
THEDIFFERENTOPTIONS OF THEREGISTRATION
2) RB1
3) RB2
4) RB2
Fig. 7. The effect of the number of hot regions (data set 1).
TABLE VII SYSTEMPARAMETERS
B. Experiments and Results
In this section, we observe the influence of different factors on the performance of the proposed strategy through the exper-iments. Also, we use the cost ratio of the proposed strategy to that of IS-41 as a comparison of the performance. That is, our proposed strategy outperforms IS-41 when the cost ratio is less than one. The values of signaling cost parameters and database access cost parameters are adopted from [9].
The settings of system parameters in our experiments are listed in Table VII. Here, we assume that the probabilities of appearing in the hot and cold regions have an 80–20 ratio. The probability distributions of a user’s appearing in each region and each LA within a region are assumed to both be uniform (i.e., and ). When the probability distribu-tions are skewed, it means that the user is prone to stay in a certain region or LA. The cost of location tracking by the list method becomes small and our approaches will be even more outstanding. To save space, we omit the experimental results
TABLE VIII SIGNALINGCOSTPARAMETERS
for the cases with skewed probabilities. To simplify the anal-ysis, the cost evaluations on the signaling cost and the database access cost are separate. When one cost is considered, the other cost is negligible.
1) Signaling Cost: We consider four sets of signaling cost
parameters as listed in Table VIII, where the values have been normalized based on the value of . At first, we show the effect of the number of hot regions (or the size of a hot region) on the performance. Fig. 7 shows the change of the cost ratios for each strategy as data set 1 is used and is set to 1, 2, and 4 (the size of hot regions is 32, 16, and 8, respectively). As increases, the cost of RB1 becomes smaller, while that of RB2 becomes larger. The reason is that the cost of location tracking dominates in these strategies. The search cost for the RSs is directly proportional to the number of hot regions ( ), while the search cost for the LAs is directly proportional to the size of a hot region ( ). Both of these searches are performed in RB1 , so the cost varies as increases. The current residing LA for a user is recorded
(a)
(b)
(c)
Fig. 8. The effect of the different options of the registration (data set 1).
in RB2 such that no extra search is needed. Hence, the cost of RB2 is independent of the value of . When the CMR is high, the location tracking cost becomes dominant and its cost is sensitive to the value of , so the difference of the cost ratios is large. However, when the CMR is low, the registration cost becomes dominant and its cost is independent of the value of , so the difference of the cost ratios becomes small.
Second, we observe the effect of the different options of reg-istration listed in Table VI on the performance. The experiment is performed by using data set 1 and various values of . Fig. 8 shows the change of cost ratios for the strategy with different options, as the value of CMR varies from 0.01 to ten. We found that the strategy with a different option has a different perfor-mance as the CMR changes. Also, RB1 , RB1 , and RB2 , which involve the search costs for the RSs and/or the LAs, may perform worse than the IS-41 at high CMRs. This indicates that we should adopt a different option based on the CMR for getting the best performance. That is, the strategy should dynamically follow the options with the lowest cost ratio based on the cur-rent CMR. For example, if is four, the strategy should behave
as RB1 when CMR 0.1 and behave as RB2 when CMR
Fig. 9. The best performance for different numbers of regions (data set 1).
Fig. 10. The effect of varying signaling costs (n = 2).
0.1. In both Figs. 7 and 8, we found that the performance im-provement at low CMRs is more significant than that at high CMRs. The reason is that our proposed strategy focuses on re-ducing the registration cost and there is a tradeoff between the registration cost and the location tracking cost.
Fig. 9 shows the performances of the strategy, which always adopts the best option of registration under different values of
. It can be seen that having a number of hot regions ( ) is the best at low CMRs, while a single hot region is the best at higher CMRs. At low CMRs, the minimal cost comes from RB1 , which performs well as . We arrive at the following conclusion. If a mobile user is prone to have high mobility rates, the location strategy with adequate multiple hot regions should be adopted. Otherwise, the strategy with a single hot region should be adopted.
Fig. 10 shows the effect of signaling costs on the performance of the proposed strategy. The differences between data sets 1 and 2 and between data sets 3 and 4 demonstrate the effect of the signaling cost between two RSs. The figure shows that the effect of is more significant at low CMRs. The reason is that only affects the registration cost, which is dominant at low CMRs. The differences between data sets 1 and 3 and between data sets 2 and 4 demonstrate the effect of the signaling cost between the PS and the RS. Note that and are increased/decreased together in our settings. The effect of is also significant at low CMRs. The reason is that RB1 and RB2 , which have large cost savings of registration costs, dominate in these cases.
TABLE IX
DATABASEACCESSCOSTPARAMETERS
Fig. 11. The best performance for different numbers of regions (data set 5).
Fig. 12. The effect of varying database access costs (n = 2).
2) Database Access Cost: We consider four sets of database
access cost parameters as listed in Table IX, where the values have been normalized based on the value of . We perform a similar experiment as done in Fig. 9. The result is shown in Fig. 11. We found that the proposed strategy with the best option outperforms IS-41 even when the database access cost is consid-ered. Moreover, the effect of multiple hot regions is more signif-icant than that shown in Fig. 9. This indicates that no registration in the same hot region can save more database access costs than signaling costs. As the number of hot regions increases, the cu-mulative cost savings become large.
Fig. 12 shows the effect of database access costs on the performance. The differences between data sets 5 and 6 and between data sets 7 and 8 demonstrate the effect of the database access cost of the PS. The figure shows that more cost savings are obtained as increases ( increases too). This reveals the
fact that the centralized HLR is a bottleneck of the performance. The differences between data sets 5 and 7 and between data sets 6 and 8 demonstrate the effect of the database access cost of the VLR. It is obvious that the cost ratio increases as increases.
From the experiments, it can be justified that the proposed strategy has a significant improvement over the scheme defined in the IS-41 standard. The proposed strategy will be dynamically adjusted (that is, adopting RB1 , RB1 , RB2 , or RB2 ) based on the current CMR for a user. Although we focus on the techniques of reducing the registration cost, other techniques like caching and replication should be involved to reduce the lo-cation tracking cost in the system design. Also, the search order among regions or among LAs of a region will affect the loca-tion tracking cost. A more effective order can be arranged if the time is also considered in the movement history. Each region can be associated with time periods denoting when the user may be there. Given the time, the system can make a proper decision to locate the user.
V. SUMMARY ANDDISCUSSION
In this paper, we introduce a location management scheme based on regions. The regions are derived from the user’s move-ment patterns. The proposed approach is effective if the user has a regular movement behavior. Our focus is to reduce the registra-tion cost. We present a three-level hierarchical database archi-tecture that consists of the VLR, the RS, and the profile server (PS). The RS is responsible for the registration within the re-gion. The PS acts as a distributed HLR and records the recent serving RSs of the users. We consider the registration process in two phases: first at the RS and second at the PS. The best perfor-mance can be achieved by controlling the registration frequency in each phase. From the experiments, we obtain the following criteria. The registration frequencies in both phases should be reduced at low CMRs (less than 0.1), while only the registra-tion frequency in the second phase should be reduced at middle CMRs (between 0.1 and 1). Our study also indicates that a single hot region is sufficient most of the time, except in the case of low CMRs (less than 0.1).
The movement patterns and call histories of mobile users should be properly managed to derive useful information. We name this work profile management, which includes data col-lection, data mining, and data maintenance. In the first part, the related data are collected and forwarded to the home PS for each user. Mining techniques should be explored in the second part. Five types of patterns can be considered by the following ques-tions. Who is frequently called by a user? Where do most of the incoming calls come from for a user? When does a user receive most of the calls? When and where does a user frequently make a movement? Parts of these patterns have been cleverly used in some papers. In the final part, how to refresh the movement pat-terns will be a challenge because the moving information has been hidden from the system within a region. One approxima-tion is to refer to the call history. To reduce the cost for man-aging these patterns, we can group users with similar behaviors to share the same patterns. Profile management is a problem we are currently addressing.
REFERENCES
[1] I. F. Akyildiz and J. S. M. Ho, “On location management for personal communications networks,” IEEE Commun. Mag., pp. 138–145, Sept. 1996.
[2] I. F. Akyildiz, J. McNair, J. S. M. Ho, H. Uzunalioglu, and W. Wenye, “Mobility management in next-generation wireless systems,” Proc.
IEEE, vol. 87, pp. 1347–1384, Aug. 1999.
[3] B. Awerbuch and D. Peleg, “Concurrent online tracking of mobile users,” in Proc. ACM SIGCOMM’91, Sept. 1991, pp. 221–233. [4] B. R. Badrinath, T. Imielinski, and A. Virmani, “Locating strategies for
personal communication networks,” in IEEE Globecom ’92 Workshop
Networking for Personal Communications Applications, Dec. 1992.
[5] A. Bar-Noy and I. Kessler, “Tracking mobile users in wireless commu-nications networks,” in IEEE Infocom’93, vol. 3, 1993, pp. 1232–1239. [6] , “Mobile users: To update or not to update?,” in INFOCOM’94,
June 1994, pp. 570–576.
[7] D. C. Cox, “Wireless personal communications: What is it?,” IEEE
Per-sonal Commun., pp. 20–27, Apr. 1995.
[8] EIA/TIA, “Cellular radio telecommunication intersystem operations,” Tech. Rep. IS-41 (revision C), Nov. 1995.
[9] J. S. M. Ho and I. F. Akyildiz, “Dynamic hierarchical database archi-tecture for location management in PCS networks,” IEEE/ACM Trans.
Networking, vol. 5, pp. 646–660, Oct. 1997.
[10] A. Hac and X. Zhou, “Locating strategies for personal communication networks: A novel tracking strategy,” IEEE J. Select. Areas Commun., vol. 15, pp. 1425–1436, Oct. 1997.
[11] T. Imielinski and B. R. Badrinath, “Querying in highly mobile dis-tributed environments,” in Proc. 18th VLDB, Aug. 1992, pp. 41–52. [12] R. Jain, Y. B. Lin, C. Lo, and S. Mohan, “A caching strategy to reduce
network impacts of PCS,” IEEE J. Select. Areas Commun., vol. 12, pp. 1434–1444, Oct. 1994.
[13] R. Jain and Y. B. Lin, “An auxiliary user location strategy employing forwarding pointers to reduce network impacts of PCS,” in IEEE Int.
Conf. Communications (ICC’95), 1995, pp. 740–744.
[14] G. Lee and A. L. P. Chen, “The design of location regions using user movement behaviors in PCS systems,” Multimedia Tools Applicat., vol. 15, no. 2, pp. 187–202, Nov. 2001.
[15] D. Lam, Y. Cui, D. C. Cox, and J. Widom, “A location management tech-nique to support lifelong numbering personal communication services,”
Mobile Comput. Commun. Rev., vol. 2, pp. 27–35, Jan. 1998.
[16] Z. Lei and C. Rose, “Wireless subscriber mobility management using adaptive individual location area for PCS systems,” in IEEE Int. Conf.
Communications (ICC’98), Atlanta, GA, June 1998.
[17] Y. B. Lin and A. Noerpel, “Implicit deregistration in a PCS network,”
IEEE Trans. Veh. Technol., vol. 43, pp. 1006–1010, 1994.
[18] Y. B. Lin, “Location tracking with distributed HLR’s and pointer for-warding,” in Int. Symp. Communications, 1995, pp. 31–37.
[19] , “Reducing location update cost in a PCS network,” IEEE/ACM
Trans. Networking, vol. 4, 1996.
[20] S. Mohan and R. Jain, “Two user location strategies for personal com-munications services,” IEEE Personal Commun., pp. 42–50, 1st quarter 1994.
[21] G. P. Pollini and C. L. I, “A profile-based location strategy and its perfor-mance,” IEEE J. Select. Areas Commun., vol. 15, pp. 1415–1424, Oct. 1997.
[22] R. Rezaiifar and A. M. Makowski, “From optimal search theory to se-quential paging in cellular networks,” IEEE J. Select. Areas Commun., vol. 15, pp. 1253–1264, Sept. 1997.
[23] N. Shivakumar, J. Jannink, and J. Widom, “Per-user profile replication in mobile environments: Algorithms, analysis, and simulation results,”
MONET, vol. 2, no. 2, pp. 129–140, 1997.
[24] S. Tabbane, “An alternative strategy for location tracking,” IEEE J.
Se-lect. Areas Commun., vol. 13, pp. 880–892, June 1995.
[25] J. Z. Wang, “A fully distributed location registration strategy for universal personal communication systems,” IEEE J. Select. Areas
Commun., vol. 11, pp. 850–860, Aug. 1993.
[26] H. Xie, S. Tabbance, and D. J. Goodman, “Dynamic location area management and performance analysis,” IEEE Trans. Veh. Technol., pp. 536–539, May 1993.
Shou-Chih Lo received the B.S. degree in computer science from National
Chiao Tung University, Taiwan, R.O.C., in 1993 and the Ph.D. degree in com-puter science from National Tsing Hua University, Taiwan, in 2000.
He is now with the Computer and Communication Research Center, National Tsing Hua University, Taiwan, as a Postdoctoral Fellow. His current research interests are in the area of mobile and wireless internet with emphasis on mo-bility management, interworking operation, and MAC protocols with quality of service guarantee. He also works on problems related to index and data alloca-tion on broadcast channels.
Dr. Lo received the Best Thesis Award from the Chinese Institute of Infor-mation and Computer Machinery in 2000.
Arbee L. P. Chen (S’80-M’84) received the B.S. degree in computer science
from National Chiao-Tung University, Taiwan, R.O.C., in 1977 and the Ph.D. degree in computer engineering from the University of Southern California, Los Angeles, in 1984.
He joined National Tsing Hua University (NTHU), Taiwan, as a National Sci-ence Council (NSC) sponsored Visiting Specialist in August 1990 and became a Professor with the Department of Computer Science in 1991. In August 2001, he took a leave from NTHU and became Chairman of the Department of Com-puter Science and Information Engineering at National Dong Hwa University, Hualien, Taiwan. He was a Member of Technical Staff at Bell Communications Research, NJ, from 1987 to 1990, an Adjunct Associate Professor with the De-partment of Electrical Engineering and Computer Science, Polytechnic Univer-sity, Brooklyn, NY, and a Research Scientist at Unisys, California, from 1985 to 1986. He is an editor of several international journals including the World Wide
Web: Internet and Web Information Systems, (New York: Kluwer). His current
research interests include multimedia databases, data mining, and mobile com-puting.
Dr. Chen was a recipient of the NSC Distinguished Research Award since 1996. He had organized the 1995 IEEE Data Engineering Conference and the 1999 International Conference on Database Systems for Advanced Applications (DASFAA) in Taiwan.
