DATA CACHING ON MOBILE DEVICES

The Experimental MyMIDP Caching Framework

Hagen H¨opfner, Sebastian Wendland and Essam Mansour

International University in Germany, Campus 3, 76646 Bruchsal, Germany

Keywords:

Mobile information systems, Data caching, MIDP, Database connectivity, MySQL.

Abstract:

Data caching is an appropriate technique for reducing wireless data transmissions in mobile information sys-

tems. The literature describes numerous caching approaches on a theoretical level. Semantic, preemptive, or

context aware caches are discussed but not implemented for mobile devices even evaluated in real applications.

The problem here is the complexity of data management tools. Software for mobile devices must consider the

limited footprint of the used hardware as well as the restrictions of the application programming interfaces. In

this poster paper we describe the caching framework of our MyMIDP database driver. The framework pro-

vides interfaces that allow to implement the caching approaches discussed in the literature and to test them on

MIDP 2.0 enabled mobile devices in a MySQL environment. Hence, we provide researchers with a necessary

tool for prooﬁng the efﬁciency of their caches.

1 INTRODUCTION

Nowadays data access is almost possible everywhere

and at any time. Mobile information systems (mIS)

use wireless links for data communication. Current

technologies like UMTS or wireless LAN provide

high transmission rates, and ﬂat rate contracts reduce

the monetary costs. However, wireless data trans-

mission is energy intensive, and using it reduces the

up-time of mobile devices drastically. One solution

to this problem is to cache data once it has been re-

ceived. The literature on mIS focuses on semantic

caching. At this, query results are completely stored

and indexed using the corresponding query (Keller

and Basu, 1996; Lee et al., 1999; Ren and Dunham,

1999). This allows for analyzing the cache content.

In the best case new queries can be answered with-

out communication with the server. To a certain de-

gree it is also possible to supplement cached data to

answer a query (Godfrey and Gryz, 1999; H¨opfner

and Sattler, 2003; Ren and Dunham, 2003). In addi-

tion to this semantic caches even more enhanced ap-

proaches have been researched in the context of mIS.

Preemptive caches, a technology that is also known as

“data hoarding”, (Peissig, 2004; Kubach and Rother-

mel, 2001; Liu et al., 1996) use query rewriting tech-

niques in order to cache data before it is explicitly

used. This increases the probability of re-usability of

the cache content. Context aware caches (Ren and

Dunham, 2000; Zheng et al., 2002) take the context

of the usage of the system into account when decid-

ing upon caching and/or replacing data.

However, almost all of these ideas were discussed

but not implemented for mobile devices or even eval-

uated in real applications. The challenge addressed in

this paper was to provide researchers with a tool for

prooﬁng the efﬁciency of their caches while consid-

ering the limited footprint of mobile devices and the

restrictions of the available application programming

interfaces. Taking this into account, we developed a

caching framework for implementing various caching

approaches. It is an extension to our MyMIDP

(H¨opfner et al., 2009a; H¨opfner et al., 2009b) driver

that allows for directly accessing MySQL databases

from MIDP 2.0 enabled mobile devices.

The remainder of this poster paper is structured as

follows: Section 2 presents the caching extension of

the MyMIDP driver. Section 3 illustrates the usage

of the caching framework. Section 4 concludes this

short paper and gives an outlook on future research.

2 THE MyMIDP CACHE

EXTENSION

The MyMIDP cache extension transparently extends

the MyMIDP driver with a caching system. It pro-

256

Höpfner H., Wendland S. and Mansour E. (2009).

DATA CACHING ON MOBILE DEVICES - The Experimental MyMIDP Caching Framework.

In Proceedings of the 4th International Conference on Software and Data Technologies, pages 256-259

DOI: 10.5220/0002281302560259

 SciTePress

start

initialize connection

(and cache)

create statement

statement

cached?

retrieve from

database

retrieve from

cache

store in cache

internal

postprocessing

process

result set

close connection

(and cache)

end

yes

extended ’database querying’ step

Figure 1: Cache framework general workﬂow.

vides local data caching for database query result

sets, thus reducing the mobile device’s dependency

on unreliable and low bandwidth connections for data

transfer and, therefore, saving energy.

Here, transparency means that an application does

not notice the difference between a cached and a not

cached connection, or the difference between a result

set from the database and one from the cache. As the

workﬂowin Figure 1 shows, most of the caching logic

is hidden in the normal “database querying” step and

is thus not visible to the user.

Overall, the caching workﬂow works as follows:

If the result set of a SQL statement is found in the

cache, it is loaded, processed and returned to the

user without the need of any communication with the

database server. Otherwise, the statement is executed

and its result set cached. On default the driver uses a

per-statement caching approach with a ten entry cache

and a ﬁrst-in, ﬁrst-out (FIFO) replacement strategy.

All cached data is kept in MIDP’s record man-

agement system (RMS), a nonvolatile collection of

records of binary data. A database result set is seri-

alized into a set of records (one record for each row

in the result set) and written to a record store. A cen-

tral cache handler is responsible for tracking the re-

sult sets and can retrieve them as required. There are

two reasons for this approach. First, the MIDP stan-

dard does not deﬁne a ﬁle-access API. Second, mobile

devices only have a few hundred kilobytes of main

memory, prohibiting an in-memory cache.

3 CACHE CUSTOMIZATION

The driver defaults to a simple per-statement caching

with a FIFO queue. However,the frameworkprovides

two extension points for application programmers to

change this default behavior. Utilizing them allows

to implement more enhanced caching approaches as

discussed in Section 1.

The ﬁrst extension point provides a way to cus-

tomize the caching strategy – the way the driver de-

cides what to store in the cache. The second one deals

with cache replacement – the way the cache is kept

up-to-date. Both extension points are implemented as

Java interfaces (cf. Listings 1 and 2) and are set dur-

ing the driver’s initialization phase.

Listing 1: FIFO Cache Replacement.

public interface

Cac heSt ra tegy

{

public

String p ro ce ss Que ry( String query )

throws

SQ LE xcept io n;

public

String c ac he Mi ss( String query )

throws

SQ LE xcept io n;

public

Ca ch edRes ul tS et po st Pro ce ss Qu ery( String

query , Ca ch edRes ul tS et resu lt )

throws

SQ LE xcept io n;

}

Listing 2: FIFO Cache Replacement.

public interface

Ca ch eRepl ac em entSt rat egy

{

public

Vector s es si on Sta rt( En um er ation

ca ch eCont en t) ;

public

Ca ch eEntr y ne wE nt ry( E nu me rat io n

cacheContent , Ca cheEn tr y ne wE nt ry);

public boolean

se ss ionEn d() ;

}

3.1 Cache Strategy

The cache strategy controls the way the driver de-

cides what to store in the cache. This is accomplished

via three lifecycle methods which are called during

the execution of an SQL statement. These meth-

ods roughly correspond to the stages in the workﬂow

shown in Figure 1.

Before a statement is executed, the driver scans

through the cache index to check, whether the state-

DATA CACHING ON MOBILE DEVICES - The Experimental MyMIDP Caching Framework

257

ment is already cached or not. This is preceded by a

call to the

processQuery

method, which determines

the cache index key from the SQL statement.

If the search for the key comes up empty, the

cacheMiss

method is called. The task of this method

is to modify the original SQL statement into the SQL

statement required

by the caching strategy, which is

then send to the MySQL server instead of the origi-

nal statement. The produces result set is then added

to the cache, using the cache key provided by the

processQuery

method.

After the driver ﬁnished creating a result set, either

from cache or from the remote database, it is post-

processed by the

postProcessQuery

method. The

task of this method is to create a result set that is iden-

tical to the one which would have been created with-

out the use of caching, thus ensuring transparency.

The new result set is then passed to the application.

A Preemptive Cache Example. To show the capa-

bilities of this interface, lets have a look at how the

included preemptivecaching strategy is implemented.

It uses the idea of expanding the column selections

of the SQL statement to include all columns, leaving

the other clauses untouched (i.e. a

SELECT a, b, c

FROM xyz

becomes a

SELECT * FROM xyz

First, the

processQuery

method uses the

FROM

(and following

WHERE

GROUP BY

...) clause as cache

index key. It simply cuts the front of SQL statement

off and returns the part beginning at the

FROM

iden-

tiﬁer (i.e. a

SELECT a, b, c FROM xyz

becomes

xyz

). When the entry is not found, the

cacheMiss

method simply takes the result of the

processQuery

method and adds a

SELECT *

to its beginning, thus

creating a valid SQL query which is then send to

the database server. Finally, the

postProcessQuery

method creates a new result set based on the origi-

nal query. For each column deﬁnition in the original

SQL statement the corresponding column from the

database result set is selected and added to the new

one. Afterwards, eventual renaming is applied where

necessarry (as deﬁned by SQL

clauses). Finally,

the new result set is returned to the client.

Of course, this kind of caching can fail quite eas-

ily, for example with the use of SQL functions or

joins. However, the application developer knows the

properties of the strategy used in his application and

can thus write the SQL statements in a more robust

way. Also, there is always the option of not us-

ing caching for some statements. Furthermore, most

problems can be avoided with a more intelligent SQL

analysis. However, please note that the current ver-

sion of MIDP does not implement any regular expres-

e.g. for enabling preemptive caching

sion engine, a developer must hence implement his

own, a comprehensive undertaking in it’s own.

3.2 Cache Replacement Strategy

The cache replacement strategy controls the actual

content of the cache. This is done via the three

methods listed in Listing 2. The

sessionStart

and

sessionEnd

methods are both only called once dur-

ing the drivers lifecycle, namely during the initial-

ization and termination phases, respectively. The

newEntry

method on the other hand is called every

time the cache content is to be modiﬁed.

During the initialization phase of the driver,

the cache content index is read from the RMS

by the cache manager and is then passed to the

sessionStart

method of the cache replacement

strategy. This method identiﬁes the obsolete cache

entries and returns them in order to be deleted. Simi-

larly, the

sessionEnd

method is called during driver

shutdown, indicating whether or not the cache should

be cleared or not. Overall, the job of these two meth-

ods is to move the cache into a valid state. For in-

stance, because the default FIFO strategy indicates a

cache ﬂush during driver shutdown, it also marks the

entire cache content for deletion during initialization.

The task of the

newEntry

method, on the other

hand, is to manage the cache content. It is called

whenever a new entry is to be added to the cache and

identiﬁes at most one cache entry which is to be re-

moved from the cache. The default FIFO strategy, for

instance, does return the oldest cache entry once the

cache reached a predeﬁned size of ten entries.

4 CONCLUSIONS AND

OUTLOOK

This paper presented our caching framework for mo-

bile devices. It is an extension for the MyMIDP

database driver for MIDP 2.0 compatible devices. As

mentioned in Section 1 we had to consider device and

API limitations during the design and development

process. The given goals were reached. Our current

development version of the MyMIDP driver (incl. the

described cache implementations) has a footprint of

27kB, only. The caching framework provides ﬂexible

and transparent data caching for database result set.

For testing purposes we implemented only one

straight forward caching strategy. However, more en-

hanced approaches like semantic, preemptive or con-

text aware caches can also be implemented with our

framework. Therefore, future areas of research in-

clude also the implementation of a MIDP 2.0 com-

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

258

patible SQL analysis library that is necessary for an-

alyzing queries. Also, given the current rapid growth

in mobile device capabilities, a memory based cache

will sooner or later move into the area of feasibility.

The ﬁnal goal is to provide a database framework

similar to the capabilities of the standard Java imple-

mentation (i.e. Java SE).

Note: The GPL lizensed MyMIDP source code

as well as our proof-of-concept MySQL client for

mobile phones are available online at http://it.i-

u.de/dbis/myMIDP.

REFERENCES

Godfrey, P. and Gryz, J. (1999). Answering queries by se-

mantic caches. In Bench-Capon, T., Soda, G., and

Tjoa, A. M., editors, Database and Expert Systems

Applications: Proceedings of the 10th International

Conference, DEXA’99, volume 1677 of LNCS, pages

485–498, Heidelberg. Springer-Verlag.

H¨opfner, H. and Sattler, K.-U. (2003). Towards Trie-Based

Query Caching in Mobile DBS. In K¨onig-Ries, B.,

Klein, M., and Obreiter, P., editors, Post-Proceedings

of the Workshop Scalability, Persistence, Transactions

- Database Mechanisms for Mobile Applications,

number P-43 in LNI, pages 106–121. Gesellschaft f¨ur

Informatik, K¨ollen Druck+Verlag GmbH.

H¨opfner, H., Schad, J., Wendland, S., and Mansour, E.

(2009a). MyMIDP: An JDBC driver for accessing

MySQL from mobile devices. In Chen, Q., Cuz-

zocrea, A., Hara, T., Hunt, E., and Popescu, M., ed-

itors, Proceedings of the The First International Con-

ference on Advances in Databases, Knowledge, and

Data Applications (DBKDA 2009), March 1-6, 2009,

Cancun, Mexico, pages 74–80. IEEE Computer Soci-

ety.

H¨opfner, H., Schad, J., Wendland, S., and Mansour, E.

(2009b). MyMIDP and MyMIDP-Client: Direct Ac-

cess to MySQL Databases from Cell Phones (Demo).

In Freytag, J.-C., Ruf, T., Lehner, W., and Vossen,

G., editors, Proceedings of the 13. Conference on

Business, Technology, and Web, March 2-6, 2009,

M¨unster, Germany, volume P-144 of Lecture Notes

in Informatics (LNI) - Proceedings, pages 604–607,

Bonn, Germany. Gesellschaft f¨ur Informatik, K¨ollen

Druck+Verlag GmbH.

Keller, A. M. and Basu, J. (1996). A predicate-based

caching scheme for client-server database architec-

tures. The VLDB Journal, 5(1):35–47.

Kubach, U. and Rothermel, K. (2001). Exploiting loca-

tion information for infostation-based hoarding. In

Rose, C., editor, Proceedings of the 7th Annual Inter-

national Conference on Mobile Computing and Net-

working, pages 15–27. SIGMOBILE, ACM Press.

Lee, K. C. K., Leong, H. V., and Si, A. (1999). Semantic

query caching in a mobile environment. ACM SIG-

MOBILE Mobile Computing and Communications

Review, 3(2):28–36.

Liu, G., Marlevi, A., and Maguire, G. (1996). A mo-

bile virtual-distributed system architecture for sup-

porting wireless mobile computing and communica-

tions. Wireless Networks, 2(1):77–86.

Peissig, J. (2004). guidePort – An Information and Guid-

ance System. In Kyamakya, K., editor, WPNC 04

Proceedings, number 0.1 in Hannoversche Beitr¨age

zur Nachrichtentechnik, pages 1–17, Aachen. NICCI-

MON, IEEE, VDI, Shaker Verlag GmbH.

Ren, Q. and Dunham, M. H. (1999). Using clustering for

effective management of a semantic cache in mobile

computing. In Banerjee, S., Chrysanthis, P. K., and

Pitoura, E., editors, Proceedings of the 1st ACM Inter-

national Workshop on Data Engineering for Wireless

and Mobile Access, pages 94–101, New York, NY,

USA. ACM Press.

Ren, Q. and Dunham, M. H. (2000). Using semantic

caching to manage location dependent data in mobile

computing. In MobiCom ’00: Proceedings of the 6th

annual international conference on Mobile computing

and networking, pages 210–221. ACM.

Ren, Q. and Dunham, M. H. (2003). Semantic Caching and

Query Processing. Transactions on Knowledge and

Data Engineering, 15(1):192–210.

Zheng, B., Xu, J., and Lee, D. (2002). Cache Invalidation

and Replacement Strategies for Location-Dependent

Data in Mobile Environments. IEEE Transactions on

Computers, pages 1141–1153.

DATA CACHING ON MOBILE DEVICES - The Experimental MyMIDP Caching Framework

259