CS145 Assignment #3

Due Tuesday, April 29, 1997

Step 3 of Your PDA (Personal Database Application)

(a)

(15 pts.) Write an SQL database schema for your PDA, using the CREATE TABLE commands described in the handout Getting Started With Oracle. Pick suitable datatypes for each attribute. Page 286 of the text gives you the principal options regarding types. Hand in a printout of the commands you use to create your database schema (it is a good idea to keep this file for the balance of the course). Show the response of sqlplus to a request to describe each of your relation schemas. For example, to see the schema for relation foo type

     DESCRIBE foo;

(b)

(10 pts.) Execute about five INSERT commands to insert tuples into one of your relations. Show the response of sqlplus and the relation that results when you issue a SELECT * command. Again, the information on how to do this step is in Getting Started With Oracle.

(c)

(25 pts. Develop a substantial amount of data for your database and load it into your relations using the SQL load command. See The Oracle Bulk Loader for information on how to bulk-load data. To create the data, write a program in any programming language you like that creates large files of records in a format acceptable to the Oracle bulk loader, then load the data into your PDA relations. If you are using real data for your PDA, your program will need to transform the data into files of records conforming to your PDA schema. The rest of you will write a program to fabricate data: your program will generate either random or nonrandom (e.g., sequential) records conforming to your schema. Note that it is both fine and expected for your data values--strings especially--to be meaningless gibberish. The point of generating large amounts of data is so that you can experiment with a database of realistic size, rather than the small ``toy'' databases often used in classes. The data you generate and load should be on the order of:

• At least two relations with thousands of tuples
• At least one relation with hundreds of tuples

If the semantics of your application includes relations that are expected to be relatively small (e.g., schools within a university), it is fine to use some small relations, but please ensure that you have relations of the sizes prescribed above as well. When writing a program to fabricate data, there are two important points to keep in mind:

(1)

Although you have not (yet) declared keys in your relations, in many cases you probably know that an attribute or set of attributes in a relation will serve as a key. If so, be sure not to generate duplicate values for these attributes.

(2)

Your PDA almost certainly includes relations that are expected to join with each other. For example, you may have a Student relation with attribute courseNo that's expected to join with attribute number in relation Course. In generating data, be sure to generate values that actually do join--otherwise all of your interesting queries will have empty results! One way to guarantee joinability is to generate the values in one relation, then use the generated values in one relation to select joining values for the other relation. For example, you could generate course numbers first (either sequentially or randomly), then use these numbers to fill in the courseNo values in the student relation.

Turn in your program code for generating or transforming data, a small sample of the records generated for each relation (5 or so records per relation), and a script showing the loading of your data into Oracle.

Problem Set

1. (7 pts.) Exercise 3.4.1 [p. 117].

2. (7 pts.) Exercise 3.4.2 [p. 118].

3. (7 pts.) Suppose we have a class C and a subclass S. Suppose also that there are k attributes in the key for these classes, a other attributes for class C, and b additional attributes for objects in subclass S that do not belong to class C. Finally, suppose there are n objects in C and an additional m objects in S. There are three methods we discussed for converting this pair of classes to relations:
   • ODL style: one relation for each class, with all the attributes of the class, including inherited ones. Objects represented in only one relation.
   • E/R style: one relation for each class with only the key and attributes belonging to that class, with no inheritance. Entities represented in all appropriate relations.
   • Using nulls, where there is one relation with all the attributes of the class and subclass.

   Let us measure the ``performance'' of these methods by the total number of components of all the tuples in all the relations. (This measure is not necessarily the most appropriate one.) It turns out that the performance of one of these methods is always at least as good as the other two. Which? Of the remaining two methods, when (as a function of m, n, k, a, and b) should we prefer one of those two over the other?

4. (8 pts.) Exercise 3.5.4(c) [p. 126].

5. (8 pts.) Exercise 3.6.2(ii) [p. 135].

6. (5 pts.) Exercise 3.6.3(d) [p. 136].

7. (8 pts.) Exercise 3.6.4(b) [p. 136].