executive information systems, inc. the artificial knowledge manager and distributed ... · ·...

1

1999 Executive Information Systems, Inc.

Executive Information Systems, Inc.

The Artificial Knowledge Manager andDistributed Knowledge Management Systems

By

Joseph M. Firestone, [email protected]

Object Developers Group(ODG)February 10, 1999

EIS Web Site: http://www.dkms.com

2

Natural KnowledgeManagement Systems (NKMS)

g What is an Enterprise NKMS?g Its interaction properties are not

determined by designg They are emergentg They determine the outcomes of the

enterprise's knowledge processes.g The NKMS includes organizational

components produced by it, such ascomputers.

1999 Knowledge Management Consortium.

3

Artificial KnowledgeManagement Systems (AKMS)

g What is an AKMS?

g The AKMS supports the Knowledge andKM processes of the NKMS

g An AKMS is fully designed; while an NKMSis not

g An AKMS responds predictably to its userinputs; while an NKMS is a self-organizingcas

g Once it starts we can only interact with itand influence it.


4

AKMS Use Cases

g Two ways to look at the AKMS are interms of its use cases and itsarchitecture.

g In the Unified Modeling Language (UML)a use case is defined as "a set ofsequences of actions a system performsthat yield an observable result of valueto a particular actor." An actor is ahuman agent.


5

AKMS Use Cases (Page Two)

g When an AKMS is viewed functionally as anapplication, its users perform a set of usecases supporting various tasks within themain activities of the knowledge and KMprocesses of an NKMS.

g An AKMS doesn’t automate all NKMSactivities. Only some.

g Figure One shows the abstract relationship ofAKMS Use Cases to knowledge and KMprocesses


6

• Natural Knowledge Management Processes (NKMPs)

Figure One -- Relationships of KM Processes and Activities to AKMS Use cases

A1 A2 A3A1 A2 A3 A4

• Activitiesparticipating in

NKMPs

• AKMS Use CasesNote that they supportonly some of the NKM

Processes and Activities

NKMP3

UC3UC1

NKMP1 NKMP2

1999 Knowledge Management Consortium

7

Knowledge and KM Processes

g Knowledge Productiong Knowledge Acquisitiong Knowledge Transmissiong Representing KMg Leading KMg KM Knowledge Productiong KM Knowledge Acquisition


8

Knowledge and KM Processes

g KM Knowledge Transmissiong Changing Knowledge Process Rulesg Handling Crises in Knowledge

Processesg Allocating KM Resources and

mandating implementation for VariousKnowledge and KM Process activities

g Negotiating KM with business processrepresentatives


9

AKMS Structure

g If use cases specify the functional or activityaspect of the AKMS, the objects andcomponents of the AKMS that support theseuse cases, along with their interrelationshipsprovide its structure.

g We can begin to understand AKMS structureby visualizing a basic, abstract architecture.

g That architecture is expressed in Figure Two.


10


Figure Two -- KMC “Straw Man” AKMS Architecture

Process Control ServicesActive In-Memory Object Model

Connectivity Services

Artificial Knowledge Manager

Any ClientAny

CommBus

AnyApplication

Server Any Data Store

11

Figure Three -- The AKM, Data Stores, Application Servers, and Clients


Data Stores

Appli-cation

Servers

Appli-cation

Servers

Client Applications

DDS

DataMarts

DW ODS

PersistentObject Store

ETML

KDD

Web

Pub

Query

CTS

BPE

ROLAP

Legacy ERPBrowsers

ERPGUI

ReportingAKM GUI

ActiveKnowledge

Manager

ArtificialKnowledge

Manager

12

The AKMS and the DKMS

g A Distributed Knowledge ManagementSystem (DKMS) of an enterprise is a specifictype of AKMS designed to manage theintegration of distributed computer hardware,software, and networking objects/components into a functioning wholesupporting enterprise knowledge production,acquisition, and transmission processes.

g The DKMS, in other words supportsproducing, acquiring and transmitting theenterprise's knowledge base.


13

The AKMS and the DKMS

g Current AKMSs are, in fact, DKMSs becausedistributed systems are the currenttechnological embodiment of the AKMS.

g For the remainder of the talk I will use theterms AKMS and DKMS interchangeably.


14

AKMS: Diversity and Dynamics

g Figure Three makes clear the diversity ofcomponent types in the AKMS.

g It is because of this diversity and its rapidrate of growth in the last few years that theAKM is necessary.

g Change in the AKMS can be introducedthrough so many sources that if the AKMS isto adapt to change it needs an integrativecomponent like the AKM to play the majorrole in its integration and adaptation.


15

AKMS Architecture

g The Key Architectural Components of theAKMS are:

g The Artificial Knowledge Manager (AKM)

g Stateless Application Servers

g Application Servers that maintain State

g Object/Data Stores

g Object Request Brokers (e.g., CORBA,DCOM)

g Client Application Components

g More detail on these follows.


16

The AKM


g An AKM provides Process Control Services,an Object Model of the Artificial KnowledgeManagement System (AKMS) (the systemcorresponding to the AKMS architecture), andconnectivity to all enterprise information, datastores, and applications

g Process Control Services include:

gIn-memory proactive object statemanagement and synchronizationacross distributed objects

17

The AKM (Page Two)


gComponent management and WorkflowManagement through intelligent agents

gTransactional multithreading

gbusiness rule management andprocessing, and

gmetadata management.

g In-memory Active Object Model/PersistentObject Store is characterized by:

gEvent-driven behavior

gAKMS-wide model with sharedrepresentation

18

The AKM (Page Three)


gDeclarative business rules

gCaching along with partial instantiationof objects

gA Persistent Object Store for the AKM

gReflexive Objects, and

gSoftware Agents

g Connectivity Services should have:

gLanguage APIs: C, C++, Java, CORBA,DCOM

gDatabases: Relational, ODBC,OODBMS, hierarchical, network, flatfile, etc.

19

The AKM (Page Four)


gWrapper connectivity for applicationsoftware: custom, CORBA, or COM-based.

gApplications connectivity including all thecategories mentioned in Figure Threeabove, whether these are mainframe,server, or desktop - based.

g In the following slides I’ll expand on ProcessControl Services and the Active Object Model.

20

Process Control Services: ObjectManagement and Synchronization


g The AKMS supports a variety of datastores and application servers that allowbatch, transaction, and DSS processingto occur in the same system.

g The result of this diversity of processingactivities is to introduce frequent andrapid changes into the AKMS, its datastores, and its application servers.

21

Process Control Services: ObjectManagement and Synchronization (Page Two)


g Change in data, methods (including businessrules), and behavior is the “law of life” in theAKMS.

g The problem of managing, synchronizing andadapting to these changes in the AKMS is theDynamic Integration Problem (DIP).

g A primary function of the AKMS, and its AKMintegrative component, is to automateDynamic Integration (DI) as much aspracticable.

22

Process Control Services: ObjectManagement and Synchronization (Page Three)


g To perform dynamic integration, the AKMmust:

g look for changes in shared objects andadditions to the total pool of objects andrelationships,

g alert all system components sharing theobjects of such changes, and also

g make decisions about which changesshould be implemented in each affectedcomponent throughout the system.

23

Process Control Services: ObjectManagement and Synchronization (Page Four)


g The AKM accomplishes these tasks by usingits in-memory, shared, active object modelwith its support for event-driven behavior, acommon view of the system’s objects,declarative business rules, and caching ofdata along with use of partial instantiation ofobjects.

g In addition, the AKM relies on a persistentrepresentation of the object model. Theobjects in the object model are reflexive --aware of their present state and any changeof state.

24

Process Control Services: ObjectManagement and Synchronization (Page Five)


g The AKM accomplishes proactive monitoringand coordinating of changes in its sharedobjects through their reflexivity and capacityfor event-driven behavior, and throughsoftware agents. The capacity for event-driven behavior causes the objects to adjustin response to event-induced changes insome shared objects by makingcorresponding changes in themselves.

g A particular type of AKM event-driven objectthat is also autonomous is a software agent.

25

Process Control Services: ComponentManagement and Synchronization


g Component management is the abilityto monitor, co-ordinate, andsynchronize changes in components,and is analogous to object statemanagement at the component level.

g Component Management andSynchronization in the AKM requiresmuch the same set of capabilities asobject state management andsynchronization.

26

Process ControlServices: Work Flow Management


g Work flow refers to the automatedsystem constructed to implement a usecase, a part of a use case, or a set ofrelated use cases in a softwareapplication.

g The AKM supports management ofwork flows composed of tasksperformed by multiple applicationservers of diverse processing type.

27

Process ControlServices: Transactional Multi-threading


g Transactional multi-threading is theability to manage each thread within aprocess as a separate transaction.

g Each thread can represent an instanceof an active object.

g This form of multitasking provides theAKM with a powerful form of parallelismuseful in work flow management as wellas in object and component DI.

28

Process Control Services: Business Rule andMetadata Management and Processing


g Business Rule and MetadataManagement and Processing are bothderivative services of Object andComponent Management andSynchronization.

g Business Rules are encapsulated inobjects and components as methods,while metadata is encapsulated asattributes.

29

In-Memory Active Object Model/Persistent Object Store: Event-Driven Behavior


g The AKM provides an Active Object Model.

g Object methods in the Active Object Modelare triggered by (1) events, (2) agents, and/or (3) programmed periodic activation

g Events include user inputs, changes in objectattribute values, changes in attributesthemselves, or changes in methodsthemselves.

g Event-driven behavior is implemented in theAKMS through sequences of rules havingantecedents and consequents.

30

In-Memory Active Object Model/Persistent ObjectStore: AKM with Shared Representation


g Many of the objects in the AKM are sharedacross distributed physical platforms --either data stores, or application servers.

g In fact, the AKM may be viewed as aspecial distributed application server orbusiness process engine that maintainsstate, shares a set of reflexive objectsacross physical platforms, and managesand integrates multiple processes changingthese shared objects.

31

Figure Six -- A Distributed AKM, Shared Objects and Dynamic Integration

DynamicData Store

DataWarehouse

Data Mart

AKM Serverwith Object

Model

Communications,Change Alerts,Negotiations,Adjustments


32

In-Memory Active Object Model/Persistent Object Store: Declarative Business Rules


g Both declarative and proceduralbusiness rule networks are supportedas methods in classes and objects ofthe model.

g Declarative Rule networks are thosewhose rules fire in parallel to determinean outcome.

g Procedural Rule networks are thosewhose rules fire in sequence.

33

Figure Seven -- Declarative and Procedural Rule Networks

Inputs

Procedural Rule Network

CombinationRule

If X is A then Z is W1

If X is C then Z is W3

If X is D then Z is W4

If X is E then Z is W5

If X is B then Z is W2

TransformationRule

Declarative Rule Network

Decision


34

In-Memory Active Object Model/PersistentObject Store: Caching and Partial Instantiation


g The ability to perform partial instantiation ofobjects is particularly important to the AKM inallowing it to develop rapid queryperformance.

g In partial instantiation only those attributescalled for in a query, and only those recordsspecified are brought into the in-memoryobject model.

g In this way, the data entering the AKM fromdata stores in the AKMS can be "chunked,"and the amount of data that the AKM musthandle can be minimized.

35

Figure Eight -- The AKM and Partial Instantiation

Query

Imports only thoseattributes necessary

to answer the query andonly those objects that

can fit in memory

Update

Import

AKM with Object Model

Any Data Store


36

In-Memory Active Object Model/PersistentObject Store: The Persistent Object Store


g The AKM uses either a relational databaseor an OODBMS to store the Active ObjectModel in persistent form.

g In either case the Active Logical ObjectModel must be mapped to the physicaldata model of the database.

g The mapping is straightforward in case ofan OODBMS, because the structure of theactive object model matches the structureof the database.

37

In-Memory Active Object Model/Persistent Object Store: Reflexive Objects


g AKMs use reflexive objects.g Reflexive Objects are aware of their

present state and any change of state.g When combined with event-driven

behavior reflexive objects provide thefoundation for automatic propagation ofevents and changes in state amongthemselves.

38

In-Memory Active Object Model/Persistent Object Store: Software Agents


g A Software Agent (SA) is an object thatacts on behalf of another object (its client)and behaves to at least some degree:

g autonomously (without continuousdirection),

g socially (interacts with other agents),

g proactively (influences its environment),and

g reactively (is influenced by itsenvironment).

39

In-Memory Active Object Model/Persistent Object Store: Intelligent Software Agents


g An intelligent software agent is an SAthat:g has an in-memory knowledge base

including cognitions, evaluations,goals, and perhaps even affects;

g is rational in the sense that it makesdecisions,

gacts to attain its goals; andg learns.

40

In-Memory Active Object Model/Persistent Object Store: Static and Mobile Agents


g A static SA is one that does not move fromthe platform that creates it.

g A Mobile SA can move across a network fromone physical computer to another.

g It can do this autonomously, as it perceivesthe need for such movement.

g The "source computer" of a mobile SA is itshome agency.

g The AKM is a home agency for mobileagents.

41

In-Memory Active Object Model/Persistent Object Store: Agents and the AKM


g Mobile SAs and their agencies requirea host environment in order to execute.

g This is a distributed computingenvironment overlaying a hostdistributed computing environment.

g It provides various essential services tomobile SAs, including the ability tocreate them, and the ability to execute.

g This environment is the AKM. And, in alarger sense, the AKMS.

42

Application Servers


g The development of multi-tierdistributed processing systems wascharacterized by the appearance ofapplication servers.

g Application servers provide services toother components in a distributedprocessing system by executingbusiness logic and data logic on dataaccessed from database servers.

43

Stateless Application Servers


g "Business state is the information thatdescribes the momentary status of theorganization.”

g “To create business state, mostapplications acquire data from adatabase and then load it into memoryfor manipulations by the user."

g This is the "stateless" approachbecause, in it, a back-end database,rather than internal memory, managesstate.

44

BPEs: Application Servers That Maintain State


g "Business Process Engines manage the mostimportant business state both in a fast in-memory environment and in closecoordination with back-end databases.”

g Because of their in-memory maintenance ofstate BPEs process many user requestswithout help from a database.

g In addition, they specialize in complexbusiness rule processing, because theirability to maintain state is a specialadvantage in performing such processing.

45

Types of BPEs


g An important aspect of specifying the AKMSis specifying the current universe ofapplication servers and projecting theappearance of new types.

g Here are some criteria for defining types ofBusiness Process Engines:

g whether they are distributed acrossphysical components or not;

g whether a BPE application server dealswith a single or multiple businessprocesses;

g the business process the BPE supports.

46

Types of BPEs (Page Two)


g An AKM is just a BPE that is bothdistributed and encompasses all of anAKMS's processes.

g A multi-process BPE can fall short of beingan AKM, and instead can be restricted to acluster of related processes.

g So, there are at least three typessuggested by this criterion: a singleprocess BPE, a BPE cluster, and an AKM.

47

Knowledge and KM Process BPEs


g The third criterion for classifying BPEs is thebusiness process supported. Here is anincomplete classification of BPE applicationservers based on knowledge, KM and DataWarehousing sub-processes.

g Collaborative Planning;

g Extraction, Transformation, and Loading(ETL);

g Knowledge Discovery in Databases (KDD);

g Knowledge base/object/component modelmaintenance, and change management(The AKM);

48

Knowledge and KM Process BPEs (Page Two)


g Knowledge Publication and Delivery (KPD);

g Computer-Based Training (CBT);

g Report Production and Delivery (RPD);

g ROLAP Application Server;

g Operational Data Store (ODS);

g Forecasting/Simulation Server;

g ERP servers,

g Financial Risk Management,

g Telecommunications Service Provisioning,

g Transportation Scheduling,

g Stock Trading Servers, and

g Work Flow servers.

49

Object/Data Stores


g There are few, if any, limits on theobject/data stores in the AKMS.

g Legacy data, flat files, Relational DataBases, Object Relational Data Bases,OODBMSs, multidimensional data stores,and vertical technology databases all fitwithin the AKMS.

g The AKMS must also integrate Image,Text, Report, Video, Audio, and FileDocument Types.

50

Object Request Brokers


g ORBs

g provide an intermediate layer betweenclients and servers in a distributednetwork

g receive requests from clients and selectsservers to satisfy the requests

g can activate appropriate servers

g can translate data between clients andservers

g The AKM must support CORBA and DCOMORBs to fulfill its integrative function.

51

An AKM Standard


g The Artificial Knowledge ManagementSystem Committee (AKMSC) of theKnowledge Management Consortium(KMC) is working on an AKMStandard.

g The first meeting of the AKMSC, heldin Silver Spring, MD on January 29,agreed on the following statement ofthe scope and deliverables of thestandards effort

52

AKMSC Scope and Deliverables


g Scope of the AKM Committee:gContext: The AKMS supports

Knowledge Processes andKnowledge Management Processesranging from the individual to theenterprise levels of interaction in theenterprise context.

gStandards: AKMSC leverages activeand developing internationalstandards in specifying AKMSelements.

53

AKMSC Scope and Deliverables: Steps

g Steps in Specifying AKM System(Threaded KMC Forum on Each)follow.

g Specify AKMS Use Case Model andRelate to EKM Processes andActivities

g Specify the Artificial KnowledgeManager (AKM) Logical Component

g Specify Types of Client ApplicationComponents.


54

AKMSC Scope and Deliverables: Steps (Page Two)

g Specify Types of Application Serversg Specify Communication Buses including

Object Request Brokers (ORBs)g Specify Types of Data Storesg Specify AKMS Architectural Modelg Specify AKMS Modelg Specify Artificial Knowledge Manager

Standard


55

AKMSC Scope and Deliverables: Steps (PageThree)

g Deliverables: Proposed standards for anArtificial Knowledge ManagementSystem (AKMS), and ArtificialKnowledge Manager (AKM), expressedin a Draft Technical Report.


executive information systems, inc. the artificial knowledge manager and distributed ... · ·...

Documents