1. Introduction

Data warehousing is very popular in these years because it can provide a relevant integrated data source to a variety of users [1,4]. The data in the integrated data source come from one or several underlying databases and are well organized for user queries. A data collector is responsible for collecting the necessary information and then passes it to the data warehouse. In a data warehouse, view creation is very important to meeting the requirements of group users. Two kinds of views are commonly used. One is called materialized view, which copies data from underlying databases to a data warehouse according to given view definitions; the other one is called virtual view, which generates data from other materialized views.

Object-oriented representation could easily depict complex relations among objects. It can also show interwoven composition of attributes within objects. Besides, it has the advantage of inheritance, encapsulation, and polymorphism. Thus, the object-oriented concept has been embedded into different techniques, including databases and data warehouses. For example, Chen et al. introduced a data warehouse model suitable in object-oriented environments with a single data source [2,3]. That proposed model maintained the original structures in the source database to store the materialized views in the object-oriented data warehouse. Zhuge and Garica-Molina designed algorithms for view maintenance in a data warehouse of graph structure [12].

Many useful view maintenance techniques for object-oriented databases were proposed as well [7-9]. Trujillo et al. implemented the object-oriented modeling of data warehouses by the Unified Modelling Language [11]. Suri and Sharma embedded the object– oriented technology to some existing applications [10]. Pahwa and Chhabre introduced an approach to transfer a relational schema from into an object-oriented data warehouse [6].

In this paper, we extend the uncompressed data model to manage views derived from multiple-source environments. Meta-objects are presented and a class-integration procedure is designed to help the execution of view creation and deletion in an object-oriented data warehousing with multiple data sources.

2. Preliminary

Class is a basic group concept in object-oriented representation. A class may include some attributes and methods. The value in an attribute may be atomic or come from another class which has been defined. An instance can be declared based on a class and will own the attributes and methods in the class. If a class is a descendent of another class, then the former can inherit the attributes and methods from the latter.

After classes and instances are generated, views can then be defined to act as virtual classes for increasing the modeling and schema restructuring capability [5]. A view is usually defined by a query sentence. Objects satisfying the condition of a view are sent from the underlying databases to the data warehouse. The number of attributes in the view is equal to that given in the query sentence. A view in a data warehouse can be defined to retrieve the objects from more than one data source, with the relationships between these data sources being determined by the conditions in the query sentence. Two kinds of condition sentences may be used here. One is the independent condition sentence, in which variables can be retrieved in a single source database. The other is the dependent condition sentence, in which variables must be retrieved from more than one source database. Restated, when a condition sentence can be checked from a single source database, it is an independent condition sentence; otherwise, it is a dependent condition sentence. An object-oriented data warehouse can then be specified by the given classes, instances and view definitions.

3. Meta-Object and Class Integration

A meta-object is used in this paper to keep the class identifiers and instance identifiers used for a view definition of a data warehouse for increasing the processing efficiency of view management. It can be defined by a triple {Meta-mv, mc, mi}, where mv is the identifier of a view, mc is the set of classes used in mv, and mi is the set of instances kept in the data warehouse for mv. With the meta-object, an object-oriented data warehouse W can thus be formally defined as a quadruple {C, V, I, M}, where C is a set of classes, V is a set of view definitions, I is a set of instances generated according to C and V, and M is a set of meta-objects generated according to V.

Since the paper handles a data warehouse formed from multiple databases, the classes and instances may come from different databases. Given a class name from an underlying database in a data warehouse, the class name may also exist in some other underlying databases or may have been used in other views. A procedure called the Class-Integration Procedure is then designed to handle the integration of the classes with the same names in the multiple data sources. It uses the meta-objects to speed up the integration process. The procedure is stated as follows.

4. The Class-Integration Procedure

Input: A data warehouse W(C, V, I, M) and a class b.c, where b is a source databaseand c is a class in the database b.

Output: A revised data warehouse W’(C', V, I’, M) which integrates the class b.c withthe other classes in W.

Step 1. Search the meta-objects M for the class b.c. If b.c can be found in M (meaning it is used by some views in V), set W' = W and exit the procedure; otherwise, do Step 2.

Step 2. Send the request to the data collector to retrieve the class definition of b.c.

Step 3. Receive the class definition of b.c from the data collector. If the class name c (without b) exists in C of the data warehouse W (meaning the classes of the same name c in other source databases are used by some views in V), do the next step, otherwise, create the class c in C of the data warehouse W and exit the procedure.

Step 4. Check whether the set of attributes of the class b.c. is a subset of the class c in C, If it is, exit the procedure, otherwise, do the next step.

Step 5. Modify the class c in C to include the attributes in b.c. That is, New attributes of c in C = (old attributes of c in C) ∪ (attributes of b.c) New methods of c in C = (old methods of c in C) ∪ (methods of b.c).

Step 6. Modify all instances in I inheriting from the class c according to the new

After Step 6, the structure of the class c in the underlying database b has been integrated with the class c in C of the data warehouse W. The class c can thus be correctly used in the warehouse W.

5. The Algorithm of View Creation under Multiple Data Sources

Based on the discussion above, a view-creation algorithm under multiple data sources is designed here. The select-from-where syntax for a new warehouse view (WV) is stated as follows:

Create Warehouse View WV(wva1, wva2, …, wvan) as
Select b s 1 . a 1 , b s 2 . a 2 ,..., b s n . a n
From b f 1 . c 1 , b f 2 . c 2 ,..., b f k . c k
Where w1, w2, …, wm

Here, wva_i represents the i-th attribute in the view W, b s 1 . a 1 represents the i-th attribute which comes from a class in the source database b s i (if the attributes and the classes of the attributes exist in all the source databases of the view, the parameter b s i can be omitted), b f i . c i denotes the i-th class from the source database b f i and w_i denotes the i-th condition. The implementation algorithm for processing the above statement is proposed as follows.

6. An Example for View Creation

An example is given below to demonstrate the above view-creation algorithm. Assume a data warehouse is formed from two underlying object-oriented data sources shown in Figures 1 and 2, respectively.

Figure 1: The classes in the first data source (DS₁).

Figure 2: The classes in the second data source (DS₂).

The two databases have the same class names, but lightly different attributes and methods in the two classes StudInfo and Name. Assume two instances are created by referring to the class Dept in DS₁. One is called CS with attribute values (001, Computer Science) and the other is called IM with attribute values (002, Information Management). Similarly, assume two instances A1 and B1 respectively with attribute values (001, CS, 1) and (102, IM, 2) are created by referring to the class Classes, two instances WCC and TPH with attribute values (Chen, Wei, Chou) and (Hong, Tzung, Pei) respectively are created by referring to the class Name, and two instances ST01 and ST02 respectively with attribute values (863201, WCC, A1) and (853001, TPH, B1) are created by referring to the class StudInfo in DS1. For DS2, assume three instances are created by referring to the class Dept in DS2. One is called IE with attribute values (101, Industrial Engineering), another is called BM with attribute values (102, Business Management) and the other is called IM with attribute values (002, Information Management). Similarly, assume three instances B1, C1 and D1 with attribute values (102, IM, 2), (101, IE, 1) and (103, BM, 3) respectively are created by referring to the class Classes, three instances WCY, WTT and TMW with attribute values (Wang, Chen, Yang, Simon) , (Wang, Tzi, Ting, Julia) and (Tsai, Ming, Wen, Joe) respectively are created by referring to the class Name, three instances ST03, ST04 and ST05 with attribute values (873201, WCY, C1, image), (873204, WTT, B1, image) and (853202, TMW, D1, image) respectively are created by referring to the class StudInfo. Assume two view definitions, FreshMan and BothClassList are given in Figure 3.

Figure 3: Two view definitions in the example.

Two meta-objects, Meta-FreshMan and Meta-BothClassList, are created in the data warehouse. For the meta-object Meta-FreshMan, Meta-mv = Meta-FreshMan, mc = (DS1.StudInfo, DS1.Name, DS1. Classes, DS2.StudInfo, DS2.Name, DS2.Classes), and mi = (DS1.ST01, DS1.WCC, DS1.A1, DS2.ST04, DS1.WCY, DS1.C1). The above metaobjects are represented by a graph as shown in Figure 4.

Figure 4: A graphical representation of the meta-objects in the example.

Only eight instances, including ST01, ST04, WCC, WTT, A1, B1, C1 and IM satisfy the conditions of the view definitions. These eight instances are thus sent from the source database to the warehouse and are thus reformatted and saved in the object-oriented data warehouse. Also assume the following new view in Figure 5 is to be defined in the data warehouse:

Figure 5: A new view definition.

The view-creation algorithm processes the new view definition as follows.

Step 1. Since the classes DS1.StudInfo and DS2.Studinfo in the new view definition have been selected by the previous views in W, the algorithm executes Step 2.

Step 2. Create a new meta-object Meta-SecondStud in W.

Step 3. The attribute mc in the meta-object Meta-SecondStud is set as DS1.StudInfo and DS2.StudInfo.

Step 4. Since the two source databases DS1 and DS2 are used in the new view, A is {DS1, DS2}.

Step 5. Form the following query statements:

Query Q_DS1 for DS1
Select StudClass.DeptOf.DeptName, StudClass.ClassID, StudID
From StudInfo
Where StudClass.Grade = 2

Query QDS2 for DS2
Select StudClass.DeptOf.DeptName, StudClass.ClassID, StudID
From StudInfo

Note that in this step, the condition DS1.Dept.DeptName= DS2. Dept.DeptName is not put in the above queries since it is not an independent condition.

Step 6. For the query statement QDS1, do the following substeps:

Step 6(1). Set the counter m = 1.

Step 6(2). Read a₁ = StudClass.DeptOf.DeptName.

Step 6(3). Since in a1, the attribute names StudClass.DeptOf and StudClass are of type class, B is thus {StudClass.DeptOf, StudClass}.

Step 6(4a). Since the class of DeptOf is Dept and the class of StudClass is Classes, both of them are thus checked by the Class-Integration procedure. Since both of them have been used in the data warehouse W, they are not processed.

Step 6(4b). Add the items DS1.Dept and DS1.Class into the attribute mc in the meta-object Meta-SecondStud.

Step 6(4c). Form the following two query statements as follows:

Query Q DS1 11 :
Select tid
From DS1.Classes
Where Grade = 2;

Query Q DS1 12 :
Select tid
From DS1.Dept

Step 6(5). Set m = m + 1; m is thus 2.

Step 6(6). Repeat Steps 6-2 to 6-4 to find the other query statements as follows:

Query Q DS2 21 :
Select tid
From DS1.Classes
Where Grade = 2;
Query Q DS2 11 :
Select tid
From DS2.Classes.
Query Q DS2 12 :
Select tid
From DS2.Dept
Query Q DS2 21 :
Select tid
From DS2.Classes

Also, the items DS2.Dept and DS2.Class are added into the attribute mc in the meta-object Meta-SecondStud.

Step 7. Send the query statements Q_DS1, Q_DS2, Q DS1 11 , Q DS1 12  and  Q DS2 11 (The query statements Q DS1 21  and  Q DS2 21 are the same as Q DS1 11  and  Q DS2 11 and are thus removed) to the data collector.

Step 8. In the source database DS1, the instance ST02 satisfies Q_DS1, instance B1 satisfies Q DS1 11 and instances CS and IM satisfy Q DS1 12 In the source database DS2, the instances ST03, ST04 and ST05 satisfy Q_DS2, instances B1, C1 and D1 satisfy Q DS2 11 instances CS, IE and BM satisfy Q DS2 12 The eleven instance identifiers {ST02, ST03, ST04, ST05, B1, C1, D1, CS, IM, IE, BM} (Since the instances B1 in DS1 and B1 in DS2 inherit from the same class Classes, only one instance identifier is sent back from the data collector), are thus sent back to the warehouse W.

Step 9. Check the eleven instance identifiers in W. D is thus {ST02, ST03, ST05, B1, C1, D1, IE, BM}.

Step 10. Request the data collector to get the instances in D.

Step 11. The instances ST02, ST03, ST05, B1, C1, D1, IE and BM with their contents are thus sent back to the warehouse W. Denote them as P.

Step 12. Since only the instances ST02 and ST03 satisfy the dependent condition sentence DS1.Dept.DeptName = DS2.Dept.DeptName in WV, G is thus {ST02, ST03}.

Step 13. Add G to I of W.

Step 14. Since the instances ST02, B1 and IM in DS1 and the instances ST03, B1 and IM in DS2 are retrieved by view SecondStud, the items DS1.ST02, DS1.B1, DS1.IM, DS2.ST03, DS2.B1 and DS2.IM are thus added into the attribute mi of meta-object Meta-SecondStud.

Step 15. The view definition SecondStud is added to the warehouse W.

After the view-creation procedure is executed, all of the instances used in the view SecondStud are stored in the data warehouse. Also, the meta-object Meta-SecondStud is automatically created for later management and maintenance of the view SecondStud. The graphical representation of the warehouse after the view SecondStud has been inserted is shown in Figure 6, where a circle represents a class, a rectangle represents an atomic type, and an ellipse represents a set of attributes.

Figure 6: The graphical representation of the warehouse after the view has been inserted.

7. The Algorithm of View Deletion under Multiple Data Sources

When a view is no longer needed, it may be deleted from the objectoriented data warehouse. The statement for deleting a warehouse view is stated as follows:

8. An Example for View Deletion

Continuing the above example, assume the following statement is given to delete the view FreshMan in the data warehouse:

8.1 Delete Warehouse View FreshMan.

The view-deletion algorithm processes the statement as follows.

Step 1. Since the view FreshMan has existed in the data warehouse W, the algorithm executes Step 2.

Step 2. Set the counter j = 1.

Step 3. Read the first class DS1.StudInfo from the mc part of the meta-object Meta-FreshMan.

Step 4. Since the class StudInfo is used by the view SecondStud, all of the instances of the class StudInfo are kept.

Step 5. Set j = j + 1. j is thus 2.

Step 6. Repeat Steps 3 to 5 for the remaining five classes DS1.Name, DS1.Classes, DS2.StudInfo, DS2.Name and DS2.Classes. Since the class Name is not referred to by views in the data warehouse W except by view FreshMan, the class Name and its instances WCC and WCY are thus removed from C and I of the warehouse W.

Step 7. Set the counter j = 1 again.

Step 8. Read the first instance DS1.ST01 from the mi part of the meta-object Meta-FreshMan.

Step 9. Since the instance ST01 is not referred to by other views, the instances ST01 is thus removed from I.

Step 10. Set j = j + 1. j is thus 2

Step 11. Repeat Steps 8 to 10 for the remaining five instances DS1. WCC, DS1.A1, DS2.ST04, DS2.WCY and DS2.C1. Since the instances ST04, A1 and C1 are not referred to by views in the warehouse W except by view FreshMan (the instances WCC and WCY were removed in Step 6), they are thus removed from I.

Step 12. The view definition FreshMan and the meta-object Meta- FreshMan are removed from V and M.

After Step 12, the view FreshMan and its meta-object are removed from the data warehouse. Moreover, the instances ST01, ST04, A1, C1 and all the instances inheriting from the class Name are deleted. A graphical representation of the warehouse after the view FreshMan has been deleted is shown in Figure 7.

Figure 7: The graphical representation of the warehouse after the view has been deleted.

9. Conclusion

In this paper, we have extended our previous data model from a single source to multiple-source environments. The meta-objects have been presented to make the creation and deletion of views in an objectoriented data warehousing easy and efficient. A class-integration procedure has also been designed to handle the integration of the classes with the same names in the multiple data sources. Moreover, two implementation algorithms for view creation and view deletion have been proposed. In the future, we will attempt to use ontology to flexibly find out similar classes, instead of the classes with the same names, in class integration from multiple data sources.

Competing Interests

The authors declare that they have no competing interests.