Previous Chapter Chapter 7: Conclusion

Chapter 7: Conclusion

1. Theory of Operation

Backing up and restoring sets of values in a dynamic database are simply read and write operations with respect to the database, respectively. Although the values used during restoration are typically the product of a previous back up, the two operations, reading and writing, are inherently independent, and so the following discussions of each are separate and independent.

Before starting our discussions of reading and writing, we need to understand the nature of databases. A database is a finite set of records. At any point in time there is a finite set of physical databases(1) active in the system. So, at a point in time, let D be the database defined by the union of all the physical databases active in the system.

Each record in D is a set of ordered attribute-value pairs where the attribute is the name of the field within the record and, as these databases are not in first normal form, the value is an ordered n-tuple of atomic data values (or simply values). Therefore, an ordered 3-tuple <record name, field name, atom id>, which we call a name, is needed to uniquely identify a value in D. For example, given a record R with field F where the value of F is an ordered pair, the name <R, F, 2> identifies the second value in that ordered pair. The process of reading/writing from/to a database is the process of reading/writing these values as identified by names.

The read operation is a binary function. It takes as input a database D and a name. It returns the value identified by name if name is in D, otherwise it returns ε, a null value to indicate that name was not in D. The write operation is a ternary predicate. It takes as input a database D, a name, and the value to be written. If name is in D, the value is written there and returns true, otherwise it returns false.

Reading and writing sets of values are simply a matter of performing individual reads and writes using sets of appropriate input for each.

2. Limitations

As an implementation, BURT fails to meet the full functionality described by the theoretical framework in the previous section. That framework calls for the ability to read and write atomic data values as identified by a name. BURT cannot do that in the general sense. It is limited to reading and writing the first n values of a field. For example, BURT is not capable of reading only the second value from a field that contains more than one value. This inadequacy has also manifested itself in ca_get()'s and ss_get()'s limitations.

Footnotes

(1): A physical database is a finite non-empty set of records. For further explanation see [4].

Table of Contents

Next Chapter