See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3227566

Mobile Data Management

Article in IEEE Potentials · November 2001

DOI: 10.1109/45.969587 · Source: IEEE Xplore

CITATIONS READS
13 327

2 authors, including:

Sanjay Kumar Madria

Missouri University of Science and Technology
399 PUBLICATIONS 3,889 CITATIONS

SEE PROFILE

All content following this page was uploaded by Sanjay Kumar Madria on 27 April 2014.

The user has requested enhancement of the downloaded file.

 ireless LAN and satel-
lite services have led to

W the emergence of
mobile computing. As
a result, users are not
attached to a fixed geo-
graphical location;
Mobile data management
instead, their point of attachment to the
network changes as they move.
Assisted with low power, low-cost, and
portable computing platforms such as
laptops and personal digital assistants
(PDAs), people can now work any-

© ARTVILLE/STEVE SKINNER COMPOSITE: MKC

where, at any time.
Mobility and portability do pose
challenges to database management and
distributed computing (see Fig. 1). Sanjay Kumar Madria
Some problematic areas are how to
handle long periods of being discon-
nected, limited battery life, and variable
and
bandwidths. Also, in mobile comput-
ing, there will be more competition for
Sourav S. Bhowmick
shared data. Additionally, mobile users
will have to share their data with others.
The task of ensuring consistency of this user profile, log-in files, access rights from the closest fixed host (FH) and
shared data becomes more difficult in along with a user’s private files. At any run it on the palmtop or execute it
mobile computing due to the limitations given instant, a mobile unit communi- remotely on the fixed host. Thus, the
and restrictions of wireless communica- cates only with the base station respon- most commonly used software will be
tion channels. sible for that cell area. Ultimately, a fully replicated.
Some other interesting questions mobile unit must have unrestricted A mobile unit may have some server
include: How does mobile computing movement within G (inter-cell move- capability to perform computations
differ from distributed database com- ment) and must be able to access the locally using a local concurrency con-
puting? How does mobility affect trans- desired data from any cell. trol and recovery algorithm. Some
action processing and replication? Is When a mobile unit leaves a cell ser- mobile units may have very slow cen-
location management a database man- viced by a base station, a hand-off pro- tral processing units (CPUs) and very
agement problem? And how do we tocol is used to transfer the responsibili- little memories. Thus, they act as
replicate the location data? ty for mobile transaction and data sup- input/output (I/O) devices only and
port to the base station of the new cell depend on a fixed host.
Mobile database architecture (see Fig. 3). This hand-off involves
In the mobile computing environ- establishing a new communication link.
ment (see Fig. 2), the network consists It may also involve moving “in • Low bandwidth
of Fixed Hosts (FH), Mobile Units progress” transactions and database • Frequent disconnections
(MU), and Base Stations (BS) or states without disturbing connectivity. • High bandwidth variability
Mobile Support Stations (MSS). The entire process of the handoff • Predictable disconnections
Mobile units are connected to the wired should be transparent to a mobile unit. • Expensive
network components only through base While in motion, a mobile host • Broadcast is physically supported
stations via wireless channels. Mobile retains the network connections. The in a cell
units are battery powered, portable base stations and fixed hosts perform • Limited battery power
computers that freely move around in a the transaction and data management • Limited resources
restricted area called the “geographical functions with the help of a database • Small size and screen of laptop
region” (G). For example, in Fig. 2, G server (DBS). This permits database • Susceptible to damaging data due to
is the total area covered by all base sta- processing capabilities without affect- theft and accidents
tions. This cell size restriction is mainly ing any aspect of the generic mobile • Fast changing locations
due to the limited bandwidth of the network. Data base servers can be • Scalability
wireless communication channels. Cell either installed at base stations or can • Security
coverage is a dynamic activity that con- be a part of the fixed hosts or can be
stantly changes in response to demand. independent to both. Within this mobile Fig. 1 Constraints of mobile
To support the mobility of mobile computing environment, shared data computing
units and to exploit frequency reuse, the are stored and controlled by a number
entire G is divided into smaller areas of database servers (DBS). Modes of operations
called cells. Each cell is managed by a Base stations provide the commonly In a traditional distributed system, a
particular base station. Each base sta- used application software. This way a host is either connected to the network
tion will store information such as a mobile user can download the software or totally disconnected. In mobile com-

OCTOBER/NOVEMBER 2001 0278-6648/01/$10.00 © 2001 IEEE 11

 with the fixed network is restricted. sistent. For example, a “state tax” object
MU Selective caching of data at the host site will have different values in different
Fixed Host will minimise future network use. states of the US.
BS
DBS • Recovery protocols re-establish the The most important mobile comput-
MU
connection with the fixed network and ing issue remains transaction process-
Fixed Host resume normal operation. ing. Mobile database processing
MU • Hand-off protocols refer to the involves dealing with different types of
crossing the boundaries of a cell. State disconnects, limited bandwidth and bat-
BS DBS information pertaining to the mobile tery power and the unreliability of the
host should be transferred to the base communication link. Transaction fail-
MU
station of the new cell. ures may increase due to hand-off prob-
lems as the mobile unit moves from cell
A cell Wireless link Mobile vs. distributed to cell. A mobile unit failure creates a
Fig. 2 Architecture of MDS
computing partitioning of the network, which in
A mobile computing system is a dis- turn complicates and affects the updat-
tributed system where links between ing and routing algorithms.
Cell overlap region nodes in the network change dynamical- Also, unlike a distributed transaction,
ly. Thus, we cannot rely on a fixed net- a mobile transaction is not identified by
work structure. A single site cannot play a cell or by a remote site. It is identified
the role of co-ordinator as in a central- by the collection of cells it passes
ized system. through. A distributed transaction is
The mobile host and fixed hosts also executed concurrently on multiple
Old BS New BS
differ in computational power and processors and data sets. The execution
memory. The distributed algorithms for of the distributed transaction is co-ordi-
mobile environments should make sure nated fully by the system including con-
the bulk of the communication and currency control, replication and atomic
computation costs are borne by the stat- commit. The mobile transaction, on the
Fig. 3 Handoff between two different cells ic portion of the network. other hand, is executed sequentially
Also, many solutions for distributed through multiple base stations, and on
computing problems may not work in possibly multiple data sets, depending
puting, there are several possible modes the mobile computing arena. In a on the unit’s movement. The execution
of operations: mobile environment, a database man- of the mobile transaction is thus not
• fully connected (normal connec- agement system (DBMS) also needs to fully co-ordinated by the system.
tion); be able to recover from site, network Instead, the movement of the mobile
• totally disconnected (e.g., not a and transaction failure, as in case of dis- unit controls the execution.
failure of mobile unit); tributed systems. However, the mobility
• partially connected or weak con- factor increases the frequency of most Limited bandwidth
nection (a terminal is connected to the of these failures and complicates the Mobile computing needs to be very
rest of the network via low bandwidth). recovery. Site failures at mobile units concerned about the bandwidth con-
In addition to conserve energy, a may be due to limited battery power. sumption and variation in network
mobile computer may enter an energy The mobile unit may be in doze mode bandwidth since wireless networks
conservation mode called a doze state. (shutdown) which cannot be treated as a deliver lower variable bandwidth.
In this mode, the clock speed is reduced failure. Also, mobility may force more Bandwidth is divided among the mobile
and no user computation is performed. logging in to recover from failures. users sharing a cell. Therefore, deliver-
These disconnected modes are usual- Another important area is processing able bandwidth per user is much lower
ly predictable in mobile computing. queries. In the mobile environment, a than the raw transmission bandwidth.
Protocols can be designed to prepare the query may need to be distributed in at Certain software techniques like
system for transitions between various least two places. Part of a query may be compression and logging can be used
modes. A mobile host should be able to executed at the mobile unit and another for coping with low bandwidth. Data
operate autonomously even during total part may be at the fixed host using a compression can be used which take
disconnection. database server. less memory and communication chan-
• A disconnection protocol is execut- Another issue is location-dependent nel but takes more CPU power to
ed before the mobile host is physically query processing in a mobile environ- decompress. Logging can improve
detached from the network. The proto- ment. The same query may return dif- bandwidth usage by making large
col should ensure that enough informa- ferent results in different locations. Here requests out of many short ones and can
tion is locally available (cached) to the the replication of data has a different be combined with compression since
mobile host for its autonomous opera- meaning than in traditional distributed large blocks compress better.
tion during disconnection. It should database system (where all copies of a
inform the interested parties for the data object keep the same consistent Energy conservation
forthcoming disconnection. value). In location-dependent data man- Energy conservation is another key
• A partially-disconnection protocol agement, the same object in different issue for small palmtop units. Battery
prepares the mobile host for operation locations may have different values but power limitations must lead to new
in a mode where all communication still these values are considered as con- class of “energy efficient” data access

12 IEEE POTENTIALS
protocols and algorithms. The following tions at the time a disconnection is pre- involve updating the location of the user
strategies can be used to deal with limit- dicted. These include: in the location database as well as in
ed bandwidths and energy conservation: • Transaction process may be migrat- other replicated databases. The location
• Data can be broadcast periodically ed to a non-mobile computer if no fur- management involves searching, read-
rather than on a demand basis. There ther user interaction is needed. ing, informing and updating. If A wants
are several examples of such informa- • Remote data may be downloaded in to find the location of B, should A
tion include local traffic reports, stock advance of the predicted disconnection search the whole network or only look
market data and local sales events. The to support interactive transactions that at pre-defined locations? Should B
clients save energy by avoiding trans- should continue to execute locally on the inform any one before relocating?
mission and the unit wakes up from the mobile machine after disconnection. Management assumes that each user
doze mode only when absolutely neces- • Log records may be transferred is attached to a home location server (or
sary. Accessing broadcast data does not from the mobile computer to a non- home location register (HLR)) that
require up link channel and is “listen mobile computer. This is particularly always “knows” the unit’s current
only.” Many mobile hosts can listen to important because of the instability of address. When a user moves, the home
that broadcast, thus it supports high storage in mobile computing. Highly location server is informed about the
scalability. reliable systems use replicated logs new address. To send a message to such
• Pre-fetching can be used to down- since a mobile computer is uniquely a user, the person’s home location regis-
load the files before they are needed. vulnerable to a catastrophically failure ter is contacted first to obtain the current
• It is better to perform the execution due to user dropping the machine, data address. A special form of “address
at the fixed server rather than at the distraction by an airport security sys- embedding” is used to redirect the pack-
mobile client. Hence, for a given tem, or even the loss or theft of the ets addressed to the mobile user from
amount of energy, the trade-off is entire machine. the home location to the current loca-
between the amount of data that can be • The mobile computer may take tion. This scheme works well for the
accessed locally and the amount of data action to “declare itself down” by user who stays within his or her home
that will be processed on request removing itself from quorums for dis- area, it does not work for global moves.
remotely and delivered later. This how- tributed protocol to handle the discon- In this algorithm, when a user A
ever requires data to be partitioned nection with less overhead than in cur- calls user B, the lookup algorithm initi-
between the client and the server. rent models in which disconnection is ates a remote lookup query to the home
Another factor is the processing speed. only discovered only after it occurs. location register of B. However, this
Here again, the longer the latency that may be at a remote site. And perform-
can be tolerated in processing, the less Mobile data management ing remote queries can be slow due to
energy that is consumed. Data management in mobile comput- high network latency.
• The ability to operate disconnected ing can be described as global and local An improvement over such algo-
can be useful even when connectivity is data management. Global data manage- rithm is to maintain visitor location reg-
available. For example, disconnected ment deals with network level issues isters (VLR). The visitor location regis-
operations can extend battery life by such as location, addressing, replication, ter at a geographical area stores the pro-
avoiding wireless transmission and broadcasting, etc. Local data manage- files of users currently only visiting that
reception. ment refers to the end user level that area temporarily. If a user’s profile is
includes energy efficient data access, not found, then it queries the database in
Reliability of communication management of disconnection and the user’s home area. This technique is
Wireless connections are of lower query processing. useful in the case when a user receives
quality due to lower bandwidth, higher many calls while visiting an area since
error rates and more frequent discon- Location data management it avoids queries to the home location
nects. These factors together can increase A mobile user’s location is of prime register while at a remote site.
the communication downtime and cost importance in wireless computing. This Visitor location registers can be
due to retransmission, time-out delays is because the location of a user can be viewed as a limited replication scheme.
and error control protocol processing. regarded as a data item whose value Each user’s profile is located in its cur-
Wireless connections can also be lost changes with every move. Thus, loca- rent area when the person is not in his
due to mobility. Users may enter areas tion management is a data management or her home area.
of high interference or large concentra- problem. Primary issues here are how
tion such as conventions or other public do we know the current position of the Cache consistency
events that may result in overloading mobile unit? Where do we store the Caching of frequently accessed data
the network’s capacity. But some of location information, and who should plays an important role in mobile comput-
these problems are foreseeable. A user be responsible for determining and ing. Caching is useful during frequent
may be able to pre-announce future dis- updating the information? relocation and connection to different
connection from the network or power To locate users, distributed location database servers. Caching of frequently
down of the computer. Changing signal databases are deployed that maintain the accessed data items will reduce contention
strength in a wireless network may current location of mobile users. Thus, on the small bandwidth wireless network.
allow the system to predict an imminent location data can be treated as a piece of This will improve query response time,
disconnection. data that is updated and queried. The and help to support disconnected or weak-
Foreseeable disconnects imply that search for this data should be as effi- ly connected operations.
the system should be able to take spe- cient as any other queried data. If a mobile user has cached a portion
cial action on behalf of active transac- Writing the location variable may of the shared data, the person may

OCTOBER/NOVEMBER 2001 13
request different levels of cache consis- objects is essential in mobile computing continues its execution in the new cell.
tency. In a strongly connected mode, to increase availability and perfor- The partially executed transaction may
the user may want the current values of mance. Shared data items have different be continued at the fixed local host
the database items belonging to the synchronization constraints depending according to the instruction given by
cache. During weak connections, the on their semantics and particular use. the mobile user. Different mechanisms
user may require weak consistency These constraints should be enforced are required if the user wants to contin-
when the cached copy is a quasi-copy on an individual basis. Replicated sys- ue its transaction at a new destination.
of the database items. Each type of con- tems need to provide support for the • As the mobile hosts move from
nection may have a different degree of disconnected mode, data divergence, one cell to another, the states of the
cache consistency associated with it. application defined reconciliation pro- transaction, states of the accessed data
That is, weak connection corresponds cedures, optimistic concurrency control objects, and the location information
to “weaker” level of consistency. and so forth. also move.
Cache consistency is severely ham- Replication also permits the system • The mobile transactions are long-
pered by both the disconnection and to ensure transparency for mobile users. lived transactions due to the mobility of
mobility of clients since a server may A user who has relocated and has been both the data and users, and due to the
be unaware of the current locations and using certain files and services at the frequent disconnects.
connection status of clients. This prob- previous location wants to have the • The mobile transaction should sup-
lem can be solved by the server periodi- same environment recreated at the new port and handle concurrency, recovery,
cally broadcasting either the actual location. Mobility of users and services disconnection and mutual consistency
data, an invalidation report (reports the and its impact on data replication and of the replicated data objects.
data items which have been changed), migration still needs to be resolved. • The transaction processing models
or even control information such as There are many issues: should accommodate the limitations of
lock tables or logs. • How to manage data replication, mobile computing, such as unreliable
Broadcasting is attractive since the providing the levels of consistency, communication, limited battery life,
server need not know the location and durability and availability needed. low bandwidth communication and
connection status of its clients and the • How to locate objects of interest. reduced storage capacity.
clients need not establish an up link Should information about location also Mobile computations should mini-
connection to a server to invalidate be replicated and to what extent (is the mize aborts due to disconnection.
their caches. Also, the mobile host location dynamically changing the data Operations on shared data must ensure
saves energy since it need not transmit item)? correctness of transactions executed on
data requests, and many mobile hosts • Under what conditions do we need both stationary and mobile hosts. The
can receive the data with no extra cost. to replicate the data on a mobile site? blocking of a transaction’s executions
Depending upon what is broadcast, • How does the users’ moves affect on either the stationary or mobile hosts
an appropriate scheme can be devel- the replication scheme? How should the must be minimized to reduce communi-
oped. Given the rate of updates, the copy follow the user? In general, should cation cost and to increase concurrency.
trade-off is between the periodicity of data move closer to the user? Proper support for mobile transactions
broadcast and the divergence of the • Does the mobile environment must provide for local autonomy to
cached copies that can be tolerated. The require dynamic replication schemes? allow transactions to be processed and
more inconsistency that can be tolerated • Do we need new replication algo- committed on the mobile host despite
the less often the updates need to be rithms or can we just modify the pro- temporary disconnection.
broadcast. posed replication schemes for distrib- In optimistic concurrency control
Given a query, the mobile host may uted environment? based schemes, cached objects on
optimize energy costs by determining mobile hosts can be updated without
whether it can process the query using Mobile transaction any co-ordination. But the updates need
cached data or transmit a request for processing to be propagated and validated at the
data. Another choice could be to wait A transaction in a mobile environ- database servers. This scheme leads to
for the relevant broadcast. ment is different from a transaction in a aborts of mobile transactions unless the
However, cache coherence preserva- centralized or a distributed database in conflicts are rare.
tion under weak-connections is expen- the following ways: In pessimistic schemes in which
sive. Large communication delays • The mobile transaction might have cached objects can be locked exclusive-
increase the cost of validating cached to split its computations into sets of ly, mobile transactions can be done
objects. Unexpected failures increase operations, some that execute on the locally. The pessimistic schemes lead to
the frequency of validation since it mobile host while others execute on the unnecessary transaction blocking since
must be performed each time communi- stationary host. A mobile transaction a mobile host cannot release any cached
cation is restored. An approach that shares its states and partial results with objects while it is disconnected.
only validates on demand could reduce other transactions due to possible dis- Existing caching methods attempt to
validation frequency. But this approach connection and/or movement to another cache the entire data objects or, in some
would lessen the consistency since cell. cases, the complete file. Caching these
some old objects could be accessed • The mobile transaction requires potentially large objects over low-band-
while disconnected. computations and communications to width communication channels can
be supported by stationary hosts. result in wireless network congestion
Data replication • When the mobile user moves dur- and high communication cost. The lim-
The ability to replicate the data ing the execution of a transaction, it ited memory size of the mobile unit

14 IEEE POTENTIALS
allows only a small number of objects tems are not well equipped to handle highly mobile users,” in 1998 ACM
to be cached at any given time. continuously changing data, such as Proc. Int. Conf. Information and
location of moving objects. They Knowledge Management.
Location-dependent query address the issue of location modelling • M. Satyanarayanan, “Mobile infor-
processing by introducing the concept of dynamic mation access,” IEEE Pers. Commun.,
Processing that deals with location- attribute (whose value keeps changing), vol. 3, Feb., 1996.
dependent data can be a subject of more spatial and temporal query languages • O. Wolfson and S. Jajodia,
complex aggregate queries. For exam- and indexing dynamic attributes. “Distributed algorithms for dynamic
ple, finding the number of hotels in the replication of data,” in Proc. Symp.
area you are currently in or looking for Conclusions Principles of Database Systems, CA,
a doctor closest to your present loca- Management of data in the mobile 1992, pp. 149-163.
tion. Hence, the location information is computing environment offers new
a frequently changing piece of data. challenging problems. Existing soft- About the authors
The objective is getting the right data at ware needs to be upgraded to accom- Sanjay Kumar Madria received his
each different location to process a modate this environment. To do so, the Ph.D. in Computer Science from the
given query. The results provided critical parameters need to be under- Indian Institute of Technology in Delhi,
should satisfy the location constraints stood and defined. We have surveyed India in 1995. He is an Assistant
with respect to the point of query ori- some problems and existing solutions. Professor with the Department of
gin, where the results are received, etc. There is a need to explore these issues Computer Science at the University of
We propose building additional further and improve the existing solu- Missouri-Rolla. Prior to that he was a
capabilities into the existing database tions offered. visiting Assistant Professor with the
systems to handle location-dependent Department of Computer Science at
data and queries. Data may represent Read more about it Purdue University in West Lafayette,
the social security number (SSN) of a • B. Daniel, “Mobile computing and IN. He has also held appointments at
person or the sales tax of a city. In one databases: A survey,” IEEE Trans. Nanyang Technological University in
representation, the mapping of the data Knowledge Data Eng., 1999. Singapore and University Sains
value and the object it represents is not • B. Bruegge and B. Benninington, Malaysia in Malaysia. He has published
subjected to any location constraints. “Applications of mobile computing and more than 50 papers in the areas of
For example, the value of a person’s communications,” IEEE Pers. Web warehousing, mobile databases,
SSN remains the same no matter from Commun., vol. 3, Feb., 1996. data warehousing, nested transaction
which location it is accessed. This is • J. Cai, K.L. Tan, and B.C. Ooi, management and performance issues.
not true for sales tax data. The value of “On incremental cache coherency He guest edited WWW Journal and
the sales tax depends on where the schemes in mobile computing environ- Data and Knowledge Engineering for
query is executed. For example, sales ment,” in Proc. 1997 IEEE Int. Conf. special issues on Web data manage-
tax value of West Lafayette is gov- Data Engineering. ment and data warehousing. He was the
erned by a different set of criteria than • Y. Huang, P. Sistla, and O. Program Chair for the EC&WEB 2001
the sales tax of Boston. We can, thus, Wolfson, “Data replication for mobile conference held in Germany in
identify data whose value depends on computers,” in Proc. 1994 ACM SIG- September, 2001. He is serving as a
criteria established by the location and MOD Int. Conf. on Management of Program Chair member at various data-
data not subject to the constraints of a Data. base conferences and workshops and is
location. • T. Imielinksi and B.R. Badrinath, a reviewer for many reputable database
There is a third type of data that is “Wireless mobile computing: journals. Dr. Madria has given tutorials
sensitive to the point of query. Consider Challenges in data management,” on Web warehousing and mobile data-
a commuter who is travelling in a taxi Communications of ACM, vol. 37, no. bases at many international confer-
and initiates a query on his laptop to 10, Oct. 1994. ences.
find nearby hotels. The answer to this • P. Krishna, N.H. Vaidya, and D.K. Sourav S. Bhowmick is currently an
query depends on the location of the Pradhan, “Static and dynamic location Assistant Professor at Nanyang
origin of the query. Since the commuter management in mobile wireless net- Technological University in Singapore.
is moving he may receive different works,” Journal of Computer His current research interests include
results at a different location. Communications (special issue on XML data management, change man-
Thus, the query results should corre- Mobile Computing), vol. 19, no. 4, agement on the Web, Web warehous-
spond to the location where the result is Mar. 1996. ing, Web mining, and mobile location-
received or to the point of the origin of • Y. Lin, “Reducing location update sensitive data. He earned his Ph.D.
the query. The difference in the two cost in PCS networks,” IEEE/ACM degree from Nanyang Technological
correct answers to the query depends on Trans. Networking, vol. 5, no. 1, pp. University in 2001 and received his
the location and not on the hotel. The 25-33, 1997. Master’s degree in Computing from
answer to the query “find the cheapest • S.K. Madria, M. Mohania, B. Griffith University in Australia. He has
hotel” is not affected by the movement. Bhargava, and S. Bhowmick, “A study published 17 conference and journal
In a project called MOST, a data- on mobile data and transactions,” papers for various international confer-
base is considered that represents infor- accepted Information Science Journal, ences and international journals includ-
mation about moving objects and their 2001. ing Data and Knowledge Engineering,
location. The project’s backers argue • E. Pitoura and I. Fudos, “An effi- World Wide Web Journal, and
that existing database management sys- cient hierarchical scheme for locating Computer Journal, among others.

OCTOBER/NOVEMBER 2001 15

View publication stats