Adding indexes

To introduce the necessity of indexes, we propose to perform the following operations:

? for (x in 1 <= 2000) Person(firstname : "xx" + string(x));
? select Person.firstname = "xx20";
= bag(2628.2.201654:oid)
? select Person.firstname = "xx10";
= bag(2608.2.642448:oid)

The first operation creates 2000 person instances: as you can notice, this operation takes only a few seconds on a reasonnable computer.

The two last operations query person instance according to their firstname attribute.

As you can notice, these query operations take a while: about 3 seconds for each operation on a ULTRA Sparc: of course, this is too much.

A good idea is to affect an index on the attributes - for instance firstname, lastname and age - for which one wants to perform efficient query.

This is very simple:

1. add index specification to the class Personedit within the person.odl file as follows:

   class Person {
     string firstname;
     char lastname;
     int age;
     Address addr;
     ...
     set<Person *> children;
   
     index on firstname;
     index on lastname;
     index on age;
   };
   

2. then, use the eyedbodl tool to update the database schema:

   % eyedbodl -U bill -P {bill password, i.e. 'thekid'} -d foo -u --package=person person.odl
   Updating 'person' schema in database foo...
   Creating [NULL] hashindex 'index<type = hash, propagate = on> on Person.firstname' on class 'Person'...
   Creating [NULL] hashindex 'index<type = hash, propagate = on> on Person.lastname' on class 'Person'...
   Creating [NULL] btreeindex 'index<type = btree, propagate = on> on Person.age' on class 'Person'...
   
   Done
   

Now, you can try again to query Person instances according to its firstname, lastname or age:

% eyedboql -U bill -P thekid -d foo -w
? select Person.firstname = "xx20";
= bag(2628.2.201654:oid)
? select Person.firstname = "xx10";
= bag(2608.2.642448:oid)

and you will notice that these operations are immediate.