TOWARDS A COMPLETE DATA MANAGEMENT

FRAMEWORK BASED ON INTELLIGENT AGENTS

Iulian Alexandru Negroiu, Octavian Paul Rotaru, Mircea Petrescu

University “Politehnica” Bucharest Romania

Keywords: Data Management, Data Processing, Agent Based Framework, Intelligent Agent.

Abstract: Applications are more and more complex and the volume of information has an exponentially growth

nowadays. In these conditions a huge amount of information needs to be processed while the processing

time and power should be kept to a minimum. The increasing amount of data transferred over the Internet

and other networks, which are open to a big number of clients, was reflected in the growth of the distributed

information systems. Also, there are a multitude of servers distributed among remote locations, which are

serving the same purposes. In such cases, traditional models of distributed computing, caching, concurrency

control, e.g. are inappropriate in overcoming the actual efficiency problems and in supporting the

development of complex applications. We believe that intelligent and autonomous agents can solve this

problem. This paper opens the research for creating a complete data management intelligent agent based

framework. The possible areas that can be handled using agents are identified and discussed together with

the required agents and agencies, while attempting to provide a bird’s eye architectural view of the proposed

framework.

1 INTRODUCTION

The increasing complexity of the applications and

information systems leads to a growth in the need of

using distributed models and complex multi-server

solutions in the development of computing

applications. Such software systems may use both

wireless and wired networks, may have time

constraints to be met and may also require to access

low memory devices.

In these technical conditions there is a need to

find a complete approach for managing data in the

entire business process of an application. We believe

that a framework based on intelligent agents can

constitute a suitable solution for the above stated

problem. We aim to identify the data management

areas that can be handled using agents and the agents

and agencies that are required to cover the identified

areas.

The framework addresses some of the main

performance issues caused by the load of data,

distribution of the information systems or network

traffic. Other very important aspect of the

framework is the flexibility to changes and uses of

data information systems that are supported.

Changing of data while offline or online, or by

external systems can no longer be a tremendous

problem and may require minimum or even no work

at all in adapting the system at runtime. Specialized

intelligent agents that compose the framework can

perform actions like constraint checking, object

locking, and queries.

2 THE FRAMEWORK

A list of well-defined intelligent agents that perform

different actions having different scope, can work

together in order to compose an agents’ based

framework. The framework itself can be seen as an

abstract agent that resides in the system, mainly in

the logical layer, but having tentacles in data and

presentation layer as well, and that reacts to different

changes of the system having as the main goal the

complete management of the data and information.

The framework will work the same as an external

system that integrates perfectly into the actual

environment. The minimum required agents that

should be part of the proposed framework, in order

to cover the management of data in an application,

are:

339

Alexandru Negroiu I., Paul Rotaru O. and Petrescu M. (2006).

TOWARDS A COMPLETE DATA MANAGEMENT FRAMEWORK BASED ON INTELLIGENT AGENTS.

In Proceedings of the Eighth International Conference on Enterprise Information Systems - AIDSS, pages 339-342

DOI: 10.5220/0002460103390342

 SciTePress

o Constraint checking agents (Constraint

Checking Agency in Figure 1);

o Intelligent query agents (Intelligent Query

Agency in Figure 1);

o Statistics collect agents (Statistics Collection

Agency in Figure 1);

o Statistics process agents (Statistics Processing

Agency in Figure 1);

o Caching update agents (Caching Maintenance

Agency in Figure 1);

o Caching decision agents (Caching Management

Agency in Figure 1);

o Resource Locking Agents (Locking Agency in

Figure 1);

o Mediation agents for multi source information

access (Mediation Agency in Figure 1);

o Dispatcher, coordination and supervisor agents

(Coordination Agency in Figure 1).

Figure 1: The architecture of the proposed data

management agent-based framework

3 FRAMEWORK’S AGENCIES

All the agencies presented above in Figure 1 will be

discussed in this section in order to clearly define

their purpose and provided services, justify their

presence in the framework and present their

interactions with other agents.

3.1 Constraint Checking Agents

Integrity constraints are valuable tools for enforcing

consistency of the data in a database. In order to

ensure integrity and consistency over the data

spanning multiple distributed information systems

one has to check the global constraints applied in

the specified situation.

A simple modification of data on one site can

cause the violation of a global constraint. The check

for such violation involves accessing related data

from multiple sites. In consequence, one local

constraint checker can generate sub-constraint

checkers on the remote data by using multiple

remote agents. The development of a framework or

agent that checks the global integrity constraints of

a system is analyzed in (Madiraju, 2004).

Having an agent-based system that implements

the constraint checking mechanism will allow the

decentralization of the remote checking by having

an intelligent decision based solution in a

concurrent environment.

An agent based constraint-checking solution for

federative database cannot be separated from the

transaction processing, and therefore it will require

the implementation of the Two Phase Commit

(2PC) (Ozsu, 1999) algorithm. 2PC is suitable to be

implemented using agents since anyway its logic is

distributed between participants, which will execute

the local sub-transaction and vote if to commit or

rollback, and a coordination process, which will

take the global commit or rollback decision based

on the votes of the participants. It must be noted that

the concept of heterogeneous two phase commit and

support for it took nearly a decade to get

implemented by DBMS vendors. (Breitbart, 2000)

mentions several early efforts in this area.

Another important application of intelligent

agents is the verification of constraints using partial

information. This is mainly relevant in distributed

environments where accessing remote located data

is usually very costly. Even in some centralized

databases the verification of the constraints using

partial information can lead to performance

improvements, if the verification is done without

using the data stored into the database, and an

intelligent agent can also handle it. Information

regarding the theory of constraint verification using

partial information can be found in (Gupta, 1994).

3.2 Intelligent Query Agents

Another important part of the framework consists in

querying the data from the informational systems.

The huge volume of information as well as

communication bandwidth limitations, time

constraints, multiplicity of data formats and maybe

the most important, the need for openness and

intuitive human interfaces makes the querying layer

of the data management system to a crucial

component of any information system.

This layer should handle the translation of the

requests in queries and sub queries, spread among

the data sources in accordance with the data needed,

while giving the impression of a single, unified,

homogeneous data source.

ICEIS 2006 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS

340

Researches in the field have come up with agents’

models like AQUIRE (Das, 2002) that makes use of

combinations of planning and traditional database

query optimization techniques and agent-oriented

retrieval of data from distributed data sources.

3.3 Statistic Agents

Gathering and processing statistics may become an

important part of an agent-based framework like the

one we propose. Intelligent query agents, caching

agents, resource locking agents etc. all of them need

a decision support mechanism in order to complete

their own goals with the expected level of

performance. In the decision mechanism of an

agent, both reactivity to changes in the environment

and commitment to long-term courses of actions

(behaviours) must be present.

In order to preserve and improve the performance

of an agent, gathering and processing statistics

become mandatory. Agents are very well suited for

such a job, being able to reside in the application

and to gather statistics on changes in the

environment.

3.4 Caching Agents

A significant portion of an application’s resources is

consumed by I/O operations, which usually are data

accesses. The data access operations are usually the

bottlenecks of any software system, especially if

they require network transport as well. It is

therefore required to implement the data access

modules and/or components as efficient as possible.

Caching is a technique that can drastically improve

the performance of any database application. Due to

caching multiple read operations for the same data

are avoided. Caching decisions and operations can

be outsourced to agents, but these agents will

continuously require collaborating and

communicating with the intelligent query agents.

In case of a distributed database the caching

agents can also be distributed at every node and the

data can be also cached in files on the disk and not

necessary into the memory. A cached data validity

verification agent is also required in order to check

if the cached data is still valid or not. “Spying” all

the data modification requests handled can do the

validity verification. The caching agents will

perform all the activities and make all the decisions

shown in Figure 2. The caching update agents will

handle the activities of putting and getting data from

cache, while the caching decision agents will decide

if caching is required or not for the retrieved data

and if the data contained by the cache is still valid

or not.

The type of cache to be used, Primed Cache or

Demand Cache (Nock, 2004), is more an

architectural decision that a decision that can be

taken by an agent. The framework should provide

agents able to handle both on demand and primed

caching techniques, as well as combined caching

strategies, in order to cover the requirements of all

kinds of applications.

Figure 2: Caching Activity Flow.

3.5 Resource Locking Agents

Multi-user applications needs take care of resource

locking in order to avoid the situation in which more

than one user attempts to update the same

simultaneously. Files, database objects, connections,

etc. are all resources that requires to be locked by a

user whenever requiring their usage. Intelligent

agents can handle both the identification of the

critical resources that requires locking and the

locking and unlocking mechanisms.

An important problem that needs to be tackled by

the framework is the granularity of the locks, which

influences heavily the overall performance of the

system. Choosing the unit of locking is a trade-off

between concurrency and performance. The

concurrency is low when the locking unit is big,

while the performance can be affected in case the

locking unit is too small due to the computational

overhead introduced by the big number of locks.

TOWARDS A COMPLETE DATA MANAGEMENT FRAMEWORK BASED ON INTELLIGENT AGENTS

341

3.6 Mediation Agents

Distributed heterogeneous information systems are

built on different technology, may have different

access points and may understand different

languages. In this conditions there is a need of a

fully and easy to customize mechanism that can

interact and understand multiple information

systems and transport methods. The mediation

agents are necessary for maintaining a well-defined

and logical communication with the information

systems. They can improve performance and can

deal with asynchronous communication issues as

well.

3.7 Supervisor Agents

The framework we propose contains a big number of

agents all of them needing to communicate and all of

them having a well defined role in the system. The

agents should form a team and communicate one

with another in the easiest manner while still

maintaining the purpose of creation and giving

outputs in each ones specific area.

The goal of maintaining the distribution of roles

while still giving the impression of one consistent

module can be undertaken by a supervisor and

coordinator agent. The existence of such an agent is

not a must, the agents of the framework being able

to communicate between them. Yet, one supervisor

agent will be well noted in the process of

customization and completion of the framework

with new agents, as well as in the work of

integration with the actual application, acting as the

project manager of the agents’ “team”.

We can see the agents of our framework as having

roles in a big team. Agents can act and coordinate

with other agents depending on their roles. In this

idea, the coordination between agents is mainly

obtained through common tactics, strategies and

observations of actions of team members, rather than

explicit communication (Coradeschi, 1998).

4 CONCLUSIONS

Having a better performance and easier to customize

systems is an important goal in application

development and design. Intelligent agents can offer

this two advantages and much more. A data manager

framework is an important part of a system, having

the role of managing the entire process of data

processing and acting as a link between the user

logic and the data systems. Such a framework should

take decisions regarding the when and what data to

be cached, regarding resource locking, how to query

and what information system to be queried etc.

This paper presents a bird’s eye architectural

view of our proposed data management agent based

framework. The starting point is the identification of

the areas that are possible candidates for agent-based

implementation. The paper also discusses all the

identified areas in order to justify their appurtenance

to the framework and evaluate the consequences.

We believe that an agent-based implementation

of such a framework is not only possible but also

very useful in regards to performance. Apart from

that, the separation of concerns provided by the

intelligent agents increases the overall performance

of the application. Furthermore, in case of

distributed applications collaborative intelligent

agents seam to be the logical choice.

REFERENCES

Praveen Madiraju, and Rajshekhar Sunderraman. 2004. A

Mobile Agent Approach for Global Database

Constraint Checking. SAC. Nicosia, Cyprus

Subrata Das, Kurt Shuster, and Curt Wu. 2002.

ACQUIRE: Agent-based Complex Query and

Information Retreival Engine. AAMAS. Bologna,

Italy.

M. Tamer Ozsu, Patrick Valduriez. 1999. Principles of

Distributed Database Systems 2

edition. Prentice Hall

Inc.

Breitbart, Y. 1990. Multidatabase Interoperability.

Sigmod Record.

Ashish Gupta, Yehoshua Sagiv, Jeffrey D. Ullman,

Jennifer Wildom. 1994. Constraint Checking with

Partial Information. 13

ACM SIGACT-SIGMOD-

SIGART Symposium on Principles of Database

Systems. Minneapolis. pg. 45 - 55.

Clifton Nock. 2004. Data Access Patterns: Database

Interactions in Object-Oriented Applications. Addison-

Wesley, Pearson Education.

Silvia Coradeschi and Lars Karlsson. 1998. A role-Based

Decision-Mechanism for Teams of Reactive and

Coordinating Agents. RoboCup-97: The First Robot

World Cup Soccer Games and Conferences. Springer

Verlag Lecture Notes in Artificial Intelligence (LNAI)

Stan Franklin and Art Graesser. 1996. Is it an Agent, or

just a program?: A Taxonomy for Autonomous Agents.

International Workshop on Agent Theories,

Architectures and Languages, Springer-Verlag.

Dr. Bruce C. Gabrielson, Mr. Leonid Kunin. 1999.

Enhacing Near Real-time Security Using an Intelligent

Agent Approach. IA-COP Information Assurance

Workshop, Advanced Information Technology Joint

Program Office, McLean, VA.

ICEIS 2006 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS

342