CACHING ESTRATEGIES FOR MOBILE

DATABASES

Heloise Manica

Universidade Federal de Santa Catarina, Florianópolis, Brasil

Murilo S. de Camargo

Universidade de Brasília, Brasília, Brasil.

Keywords: Mobile Computing, Mobile Da

tabases, Data Management, Cache replacement, Cache consistency.

Abstract: Caching remote data in local storage of a mobile client has been considered an effective solution to improve

ystem performance for data management in mobile computing applications. In this paper, we propose a

taxonomy for cache management in mobile database systems. The aim is to provide a unifying framework

for the problem of caching in mobile computing, then a comparative review of the work done in this area up

to now. Such a framework, with the associated analysis of the existing approaches, provides a basis for

identifying strengths and weaknesses of individual methodologies, as well as general guidelines for future

improvements and extensions.

1 INTRODUCTION

In a mobile computing environment, the clients are

mobile units (MUs) that communicate with data

servers through a wireless link, accessing

information at anytime and anywhere. Several new

applications such as traveler information systems,

sales in shopping center and train station (Hara,

2002) have been motivating researches on query

processing in mobile databases systems (MDB).

The data request is made through equipments as

ersonal digital assistants (PDA), notebooks, etc.

that have limited local resources. Moreover, the

wireless network is vulnerable to frequent

disconnections, low-quality communication and

scarce bandwidth.

An effective solution to deal with these problems

is the data caching techni

que that stores data copies

frequently used in the clients. The cache

management considers two dimensions: consistency

and replacement policies. Cache invalidation aims to

keep data consistency between the client’s cache and

the server. The cache replacement policy determines

which data should be removed from the cache when

there is no more space to accommodate a new item.

In a mobile computing environment the solutions

roposed for traditional distributed systems cannot

be applied because they generate high network

traffic and too much power consumption.

In this paper we present and classify different

approaches

for database cache management in

mobile computing. A contribution of this paper is

the taxonomy proposed for the problem and a

comparative review of the work done in this area.

The remaining of this paper is organized as

fol

lows. In sections 2, 3 and 4 we propose and

describe a taxonomy for cache management in

mobile computing. Finally, in section 5 we discuss

critical issues related to cache management in

mobile databases.

2 TAXONOMY

Figure 1 illustrates the taxonomy for the cache

management problem in mobile database proposed

in this paper. We begin our classification

considering the client’s cache content. This way,

three main categories can be identified: physical data

storage, logical data storage and other storage forms.

545

Manica H. and S. de Camargo M. (2004).

CACHING ESTRATEGIES FOR MOBILE DATABASES.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 545-551

DOI: 10.5220/0002616205450551

 SciTePress

Figure 1: Taxonomy for cache management problem in mobile databases

In the physical data storage model the MU cache

contents are copies (tuple or page) of data items

from the server. Otherwise, in the logical data

storage, arbitrary query answers are stored in the

client’s cache.

Different from the physical model, the data is

retrieved from the server using queries. This requires

more processing capability at the server, but only the

required data is transmitted over the wireless link.

The remaining of this section describes the cache

invalidation and replacement in the physical data

storage model. Next two sections will describe the

logical data storage model and other storage forms.

2.1 Cache invalidation

Another classification for cache invalidation is

found in (Tan et al., 2001). Our taxonomy considers

that the cached data can become invalid either

because of data updates in the server (temporal

invalidation) or due to the client movement to

another location area (spatial invalidation).

2.1.1 Temporal invalidation

In (Barbara & Imielinsk, 1994) the authors

presented the first models introducing cache

inconsistency techniques for mobile computing.

They proposed some interesting strategies based

on invalidation reports (IR) that encouraged the

appearing of new ones [(Hu & Lee, 1998), (Yen et

al., 2000), (Kahol et al., 2001), (Cao, 2002)].

Next, we introduce the temporal invalidation

parameters considered in our taxonomy proposal.

Architecture. Cellular networks are composed of a

fixed component (where servers are located), by

several cells covered by mobile support stations -

(MSS) and by mobile units (MUs). The

communication channel is divided into downlink

(server to client) and uplink (clients to server).

Another architecture is the ad-hoc network where

computing devices are able to change information

directly, without MSS help.

Server type. The server can be either statefull or

stateless. A statefull server knows what MUs reside

in its cell as well as their caches states. A stateless

server has no information about MUs.

Invalidation method. The invalidation method

refers to the way in which the server is going to keep

the client up-to-date. Two types are commonly used:

broadcast or on-demand. In the broadcast, the server

sends IRs to clients by the downlink channel. The

clients “hear” this channel and filter the needed data,

without using the uplink channel. In the on-demand

method, the client asks the server to check their

caches validity using the uplink channel. The IRs

can be sent through broadcast synchronously or

asynchronously. In the asynchronous method, the

IRs are sent immediately after changes on data items

have occurred. In the synchronous method IRs are

broadcasted periodically.

ICEIS 2004 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

546

Kind of information. The information sent in the IR

can be values of the data that were changed since the

last report (state based) or just information about the

modified items, e.g. their identifiers (history based).

Handoff. Handoff is a process that occurs when a

MU crosses the boundary from a cell to another. In a

handoff, the data cache can be kept or completely

invalidated.

Some strategies may be more specific and

consider characteristics such as the connection time,

the cache consistency or the granularity level.

Next we briefly describe some relevant temporal

invalidation strategies proposed in the literature.

Table1 describes the invalidation parameters of each

strategy.

Barbara and Imielinsk (1994) proposed diferent

techniques named Broadcasting Timestamps (TS),

Adaptive Invalidation Reports (AIR), Amnestic

Terminals (AT), Signatures (SIG) and Quasi-copies.

In the TS strategy the server broadcasts IRs with

the timestamps of the data items that have changed

in the last w seconds. If the cache item has a

timestamp smaller than the one in the IR, then it

must be updated.

The method AIR proposes to extend the window

size w for items requested frequently by MU that

remains disconnected for a long time. They also

mention that TS should have the window size w

variable depending on the data item.

In AT strategy the server informs the data items

identifiers that have changed since the last IR. The

MU compares the items in its cache with those in the

IR. If the item is in the report, then the MU drops it

from its cache, else it considers the cache as valid.

Signatures are checksums computed over the data

items values. This technique compares data files and

checks their differences. The MUs subscribe to the

items of their interests. The items that are not in the

subscribed sets are considered equal to the ones in

the IR that are being broadcasted.

The quasi-copies technique allows different

values from the server in a controlled mode. For

instance, a MU stores product prices, it is acceptable

to use values that are not updated since they differ

less than 0.5 % of the value stored in the server.

Some data can have their values often changed,

such as the weather forecast. Others can have their

values sporadically updated as a client phone.

Considering this fact, the proposal in (Yen et al.,

2000) associates an absolute validity interval (AVI)

for each cached item. The cached item is invalidated

if the access time is greater than the last update time

by its AVI.

In the lazy pull-based model (Chan et al., 1998)

each client is responsible for invalidating its own

cache items. The update reports are sent to clients on

demand and the cached items are only validated

when they are accessed. First, the query is sent to the

server that verifies the validity of the client’s cache.

If necessary, the server sends back updated data

items or those that are not in the client's cache.

Different from TS and AT, the strategy

Asynchronous and Statefull (AS) is based on

asynchronous IR and statefull servers (Kahol et al.,

2001). Each MU maintains its own Home Location

Cache (HLC) to deal with disconnections.

Invalidation messages from servers are stored in the

HLC while the MU is disconnected. When the MU

is reconnected, the invalidation messages stored are

delivered to it.

The Bit-Sequence (BS) invalidation strategy

(Jing et al., 1997) uses a bit-sequence (or bit-vector)

to refer to data items in the report. That is, each bit

represents a data item in the database. The value 1

(one) means that the given item was updated in the

server, and 0 (zero) means that it did not have any

update. This strategy can use grouping methods to

decrease the report size.

In the Local Optimal Strategy (LOP), (Hara,

2002), a MU is able to access data items from other

MU cache. When a MU requests a data item, it

verifies if the data is in its local cache. If it is not the

case, the MU obtains the data item from the

broadcast channel or from other MU cache.

Global Optimal Strategy (GOP). The bases of

GOP (Hara, 2002) strategy are the same as the

previous LOP. The main difference is that GOP

strategy replaces its cache considering the share of

cached items among all connected MU.

Stable Group Optimal (SOP). The application

framework of SOP strategy (Hara, 2002) is the same

as in LOP and GOP. Additionally, SOP proposes

that data items are cooperatively cached in stable

groups of MUs. The invalidation is done considering

the data cache of all MUs connected to the group.

The Counter-based (CB) model (Cao, 2002)

maintains a counter to every data item. Using

counters, the server is able to know which data items

are more frequently accessed (hot data items). The

server keeps the identification of each client

(statefull server). When a handoff takes place, the

client should notify its old position to the new MSS,

so that it can recover information about the client.

The model Scalable Asynchronous Cache

Consistency Scheme (SACCS), (Wang et al., 2003),

requires that MSS identify its database objects that

are valid in the MU caches, instead of recognize all

the objects of all the MUs. It introduces three main

characteristics: the use of a flag at server and MU’s

cache; use of an identifier for each entry in the MU’s

cache after its invalidation; and all valid entries of

MU’s cache are configured with an uncertain state

when the MU is reconnected to the system.

CACHING STRATEGIES FOR MOBILE DATABASES

547

Table1: Characteristics of temporal cache invalidation strategies

Strategies

Characteristics

Archite

cture

Server

Type

Invalidation

Method

Inform

ation

Type

Hand-Off

Disco

nnecti

Time

Update

Time And

Access

Frequency

Consistency Granularity

TS, AIR AND

cellular stateless

broadcast

synchronous

history

based

invalid the

cache

yes n.v. n.v.

mentioned

as future

work

SIG cellular stateless

broadcast

synchronous

state

based

invalid the

cache

yes

mentioned

as future

work

n.v.

mentioned

as future

work

Quasi-Copies cellular stateless

broadcast

synchronous

history

based

invalid the

cache

n.v. n.v.

relax the

consistency

n.v.

IAVI cellular stateless

broadcast

periodic

history

based

n.v. n.v. yes

considers a

reasonable

average

n.v.

Lazy Pull-

Based

point-

to-point

n.v. on-demand

state

based

n.v. n.v. yes n.v.

considers

tree data

granularity

levels

AS cellular statefull

broadcast

asynchronous

call-back

history

based

keeps the

cache

yes n.v. n.v. n.v.

BS cellular stateless

broadcast

periodic

history

based

n.v. n.v. n.v. n.v.

considers a

rough bit

granularity

LOP, GOP

SOP

ad-hoc stateless

broadcast

synchronous

state

based

n.v. n.v. yes n.v. n.v.

CB cellular statefull broadcast

state

based

keeps the

cache

n.v. n.v. n.v. n.v.

SACCS cellular

stateless

and

statefull

broadcast

state

based

mark the data

items with a

unstable state

n.v. n.v. n.v. n.v.

n.v. = no verified

2.1.2 Spatial Invalidation

Spatial invalidation occurs when data values stored

in cache become invalid because of the client

movement to a new location area.

The maintenance of a valid cache when the

clients move is called location-dependent cache

invalidation and a data item can have different value

depending on its location (Zheng et al., 2002).

The main factor considered in the spatial

invalidation is the valid scope (or valid area). The

valid scope of an item value is defined as the set of

cells in which the item value is valid. There are

different forms to relate the data value with its valid

area.

The system can store points that represent the

geographical region in which the data is valid. It can

store either all the points or in an approximate form,

only a part of the geographical region. Other form is

to use vectors to save information about the region

that data is valid. In another form, the system divides

the database in logic sections according to the region

in which the data is valid.

Next, we present some important spatial

invalidation strategies proposed in the literature.

The strategy Polygonal Endpoints - PE (Zheng et

al., 2002) stores all endpoints of the polygon

representing the valid scope of a data item stored.

When the number of endpoints is large, this

technique will consume a large portion of the

wireless bandwidth and space for caching the valid

scope in the client. The advantage is the complete

knowledge of the valid scopes.

v’

=valid area circle

Figure 2: Possible valid areas. From (Zheng et al. 2002)

Zheng et al. (2002) proposed the utilization of an

approximate circle (AC) inserted inside the original

polygon (v’ in figure 2). Thus, the valid area will be

the approximate area defined by the center and

radius of the circle. A problem occurs when the

polygon shape is thin and long. In this case, the

approximating error will be high and the cache can

consider valid data as invalid if the query is outside

ICEIS 2004 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

548

the circle. Caching-Efficiency-Based (CEB) also

proposed in (Zheng et al., 2002), is a generic method

for balancing the overhead and the precision of the

valid scopes to be attached.

The method Bit Vector with Compression BVC

(Xu et al., 1999) considers that each cell has an

identification (ID) and it uses a bit vector to record

scope information. The bit vector length is equal to

the number of cells in the system and all data in

cache is associated to a bit in the vector. This way,

the value 1 (one) means that the data item is valid in

the cell and 0 (zero) that it is invalid.

Grouped Bit Vector with Compression (GBVC),

(Xu et al., 1999), proposes to store information

about cells that are adjacent or near MU current

location. The model proposes the division of the

wide geographical area of the system into groups

and intra-groups. The cell ID consists of two parts:

group ID and subgroups ID.

Another model also proposed in (Xu et al., 1999)

is the Implicit Scope Information (ISI). This model

divides the database into multiple logic sections.

Data items with the same valid area are placed at the

same section. The data item in cache will have the

format {D

, SDN

and SN

}, where D

is the value of

the data item, SDN

is the section number, and SN

the data number inside the section (scope number).

2.1.3 Spatial and temporal invalidation

In the literature, there is little work considering both

the spatial and temporal invalidation. The authors in

(Xu et al., 2003) present a performance study of the

strategies BVC, GBVC and ISI in a scenario where

temporal and spatial updates coexist.

2.2 Cache replacement policies

In location-dependent data services (LDD), the

cache replacement policy must consider other

factors besides the access probability such as:

movement, direction, speed, etc. The factors

considered in our taxonomy are valid scope, distance

and direction.

Valid Scope Area. Valid scope is the geometric area

in which the data value is valid. A common way to

perform location-dependent cache invalidation is to

attach the valid scopes to the data values returned to

the client (Xu et al., 1999).

Distance. In LDD, the server answers queries

according to the client’s location. When the valid

scope of a data value is far away from the current

client’s location, this data will have a lower chance

to become useful.

Distance in Urban Areas. The computation of the

distance between client’s location and the data valid

area can change according to the kind of application.

In a rural zone for example, we can use the

Euclidean distance

()( )

yyxx −+−

. However,

in an urban area this formula is not suitable because

the MU can move through the streets with buildings

or other obstacles.

Direction. The direction can be used first to

eliminate from the cache the data that are in the

opposite direction of the client’s movement.

Next, we describe some proposed policies for

cache replacement.

The Probability Area (PA), (Zheng et al., 2002)

defines the data items to be replaced according to a

cost function defined as the product of the access

probability of a data item and its attached valid

scope. The cost function of a value j of an item i is

ji,ji,

)A(v Pi C

⋅

, where Pi is the access probability of

the item i and A(v’

i,j

) is the valid scope for a value j

of an item i. When the data replacement is carried

out, this policy selects the data with the least cost

(s).

In the Probability Area Inverse Distance (PAID)

policy (Zheng et al., 2002), the cost function of a

value j of an item i is given by C

i,j

= Pi . A(v’

i,j

) /

D(v’

i,j

) where Pi and A(v’

i,j

) is defined in the same

way as above, and D(v’

i,j

) is the distance between

the current location and the valid area v’

i,j

. When

data replacement is carried out, PAID ejects the data

value (s) with the least cost (s).

To consider the movement direction, the authors

in (Zheng et al., 2002) take extensions of the model

PAID: PAID-U (Probability Area Inverse Distance -

Unidirectional) and PAID-D (Probability Area

Inverse Distance - Directional). In PAID-D, the

distance is calculated considering the client’s current

direction of movement. PAID-D keeps the data that

are in the direction of the client’s movement. On the

other hand, in PAID-U the distance is computed

regardless of the current direction of the client’s

movement.

The Manhattan Distance (MD) policy (Jung et

al., 2002) is suitable for location dependent queries

in urban area. The distances in urban zones are given

by |(x1-x2)|+|( y1-y2)|. The proposed algorithm

computes the Cache Replacement Score (CRS)

based on the MD computation. It chooses the

victims according to CRS by the current location of

the MU. Thus, the victims who are farthest from the

MU will be replaced first.

3 LOGICAL DATA STORAGE

The semantic cache model (SC) is an attractive

approach for mobile computing. The idea is to

maintain in the client’s cache both the semantic

CACHING STRATEGIES FOR MOBILE DATABASES

549

descriptions and associated answers for previous

queries. The SC utilizes the semantic information to

organize and to manage the client’s cache.

The query processing uses the semantic

descriptions to determine which data is available in

the cache and which ones will have to be requested

to the server. The semantic description is also

utilized in the definition of the cache replacement

policy, not requiring any additional attached

information to each tuple as in traditional cache

management systems.

In a basic version of this approach, the semantic

segment is represented by the set (S

, S

, and

), where S

and S

define the relation and the

involved attributes. S

denotes the selection

condition and S

represents the result (pointer for

the result pages).

A segment S contains all or part of the result of a

query Q. Thus, Q is divided into two parts (query

trimming). The first one is part of the query result

that is in S, called probe query. The second one is

the part that could not be found in S, called reminder

query.

After the query splitting by the first segment, the

next candidate segment will divide again the

remaining query. This process continues until there

are no more candidate segments or the reminder

query finishes. At the end, if the reminder query is

not empty it is sent to the database server to be

computed. When the server returns the answer, the

whole result is composed by all probe and reminder

queries.

The affinity refers to the kind of relationship

between the data items in the cache. This

relationship can be temporal (temporal locality) or

semantic (semantic locality). Temporal locality is

the property that items that have been referenced

recently are likely to be referenced again in the near

future. Semantic Locality is the property that if an

item has been referenced, other items with the same

semantic function (for instance, the nearest) are also

likely to be referenced.

In the following, we introduce some models for

replacement in semantic cache for mobile

computing.

Dynamic Least Recent Used (D-LRU), (Ren &

Dunham, 1998), is an adaptation of the traditional

LRU approach. Considering that the segments can

be shrunk or enlarged, the algorithm D-LRU only

updates the value of the timestamp when a new

segment is created or when a new element is added.

If it is shrunk, an old timestamp is kept.

The two-level LRU model (Ren & Dunham,

1999) uses an additional structure called clusters that

are groups of semantically related segments. The

first level LRU policy is carried out at a cluster

level. The cluster with the oldest timestamp is

selected as candidate to be further examined. Next,

for all the segments in this cluster, a second level

LRU is run and replaces the items that are least

recently accessed until the cache space is enough to

hold the new query.

Furthest Away Replacement (FAR), (Ren &

Dunham, 2000), is a solution for SC replacement in

LDD. This policy classifies the segments in two sets:

the first one with segments that are in the direction

of the movement, and the second with segments that

are out of the direction. The victims are always

selected from out-direction set. When the out-

direction set is empty, the most distant segments of

the in-direction set are replaced.

4 OTHER WAYS OF STORAGE

An example of other ways of storage data is the

summary database. The main goal of the summary

database model (Madria & Roddick, 1998) is to

increase the data availability through the

construction of a summary of the main database

(complete in the MSS) that will be stored in the

client’s cache.

Thus, the mobile client can process a query using

summarized data in cache returning approximate

results while MSS returns an exact result. To

compute the database summary, it uses the hierarchy

concept, generally defined by database

administrator.

This model is able to provide several

approximated levels of answers for queries carried

out by the MU, using data stored locally in the client

or remotely in the server. The MU can stay

disconnected using a portion of the local database. In

long time connections, the local data can be updated,

avoiding a high degree of outdated data.

5 FINAL REMARKS

In this paper, we have discussed cache invalidation

and replacement issues in the context of mobile

computing. In summary, we have presented several

solutions proposed in the literature and we have also

proposed a taxonomy for this problem.

Regarding physical storage, we have presented

the main areas in cache management: query

invalidation caused by both temporal- and location-

dependent updates, and cache replacement

considering the valid scope for the data items. For

the logical storage, we have presented the semantic

cache model and finally, as other storages formats,

we have presented the summary database model.

ICEIS 2004 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

550

This paper provides a basis for identifying

strengths and weaknesses of individual

methodologies, as well as general guidelines for

future improvements and extensions. The semantic

cache model for LDD opens up a new dimension of

research in mobile computing. Some improvement

can still be made on the proposed cache replacement

schemes.

Since mobile clients may have different

movement patterns, adaptive techniques can be

developed. Considering that clients can move, but

also stay temporarily fixed, we will continue our

research work looking for a good proposal on

semantic cache replacement policy that considers

user’s behavior to decide the best way to replace the

data in the client’s cache.

Furthermore, an important and challenging area

that is in need of good solutions is the data

management in independent or ad-hoc network. In

(Chiasserini et al., 2003) is treated the cache

placement problem, but with static topology. There

are few proposals for cache management for this

kind of mobile architecture (Tassiulas & Su, 1997).

REFERENCES

Barbara, D. and Imielinski, T., 1994. Sleepers and

Workaholics: Caching Strategies in Mobile

Environments. In

Proceedings of ACM SIGMOD. May

1994.

Chiasserini, C. F.; Nuggehalli, P. and Srinivasan, V., 2003

Energy-Efficient Caching Strategies in Ad-Hoc

Wireless Networks. In

Proceedings of MobiHoc’03.

Cao, G., 2002. A Scalable Low-latency Cache Invalidation

Strategy for Mobile Environments. In

IEEE

Transactions on Knowledge and Data Engineering

Chan, B. Y.; Si, A. and Leong, H. V., 1998. Cache

Management for Mobile Databases: Design and

Evaluation. In

Proceedings of the 14th International

Conference on Data Engineering.

Gray, C. G. and Cheriton, D. R., 1989. Leases: An

Efficient Fault-Tolerant Mechanism for Distributed

File Cache Consistency. In

Proceedings of SOSP’89.

Hara, T., 2002. Cooperative Caching by Mobile Clients in

Push-based Information Systems. In

Proceedings of

CIKM’02

, November 4-9, Mc Lean, Virginia, USA.

Hu, Q.L. and Lee, D.L., 1998. Cache Algorithms Based on

Adaptive Invalidation Reports for Mobile

Environments. In C

luster Computing, 1 (1): 39-48,

February 1998.

Jing, J.; Elmagarmid, A. K.; Helal, A. and Alonso, R.,

1997. Bit-sequences: A new cache invalidation

method in mobile environments. In

ACM/MONET,

2(2): 115-127.

Jung, Y. Y.; Lee, J. and Kim, K., 2002. Broadcasting and

Caching Policies for Location-Dependent Queries in

Urban Areas.

WMC’2002. Georgia, USA.

Kahol, S. K.; Gupta, S. K. S. and Srimani, P. K., 2001. A

strategy to Manage Cache Consistency in a Distributed

Mobile Wireless Environment.

IEEE Transactions on

Parallel and Distributed Systems (TPDS’2001)

Madria, S. K. and Roddick, M. M., 1998. A Query

Processing Model for Mobile Computing using

Concept Hierarchies and Summary Databases.

The 5th

International Conference of Foundations of Data

Organization

, Kobe Japan, November 12-13, 1998.

Ren, Q. and Dunham, M. H., 1998. Semantic Caching and

Query Processing.

Technical Report 98-CSE-04.

Department of Computer Science and Engineering

Southern Methodist University, Dallas, May, 1998.

Ren, Q. and Dunham, M. H., 1999. Using Clustering for

Effective Management of a Semantic Cache in Mobile

Computing. In

ACM MobiDE, Seattle – USA, 1999.

Ren, Q. and Dunham, M. H., 2000. Using Semantic

Caching to Manage Location Dependent Data in

Mobile Computing. In

Proceedings of Mobicom’2000.

Tan, K. L.; Cai, J. and Ooi, B. C., 2001. An Evaluation of

Cache Invalidation Strategies in Wireless

Environments.

IEEE (TPDS), 12(08): 789-807.

Tassiulas, L. and Su, C. J., 1997. Optimal memory

management strategies for a mobile user in broadcast

data delivery system. IEEE Journal on Selected Areas

in Communications (JSAC), 15 (7): 1226-1238.

Wang, Z.; Das, S.; Che, H. and Kumar, M., 2003. SACCS:

Scalable Asynchronous Cache Consistency Scheme

for Mobile Environments. In

(ICDCSW'03) - 23rd

International Conference on Distributed Computing

Systems Workshops

, pp. 797.

Xu, J.; Tang, X. and Lee, D. L., 2003. Performance

Analysis of Location-Dependent Cache Invalidation

Schemes for Mobile Environments.

TKDE 15 (2) 474-

488, 2003.

Xu, J.; Tang, X.; Lee, D. L. and Hu, Q., 1999. Cache

Coherency in Location-Dependent Information

Services for Mobile Environment. In

Proc. the 1st Int.

Conf. on Mobile Data Access (MDA'99),

Hong Kong,

Springer-Verlag LNCS, vol. 1748, pp. 182-193.

Yen, J.; Chan, E.; Lam, K. Y. and Leung, H. W., 2000. An

Adaptive AVI-Based Cache Invalidation Scheme for

Mobile Computing Systems

. In 11th International

Workshop on Database and Expert Systems

Applications - DEXA'2000

Zheng, B. ; Xu, J. and Lee, D. L., 2002. Cache

Invalidation and Replacement Strategies for Location-

Dependent Data in Mobile Environments. In

IEEE

Trans. on Computers, Special Issue on Database

Management and Mobile Computing

, 51(10): 1141-

1153, October 2002.

CACHING STRATEGIES FOR MOBILE DATABASES

551