Physical Design

Physical Database Design (Defined): Process of producing a description of the implementation of the database
on secondary storage; it describes the base relations, file organizations, and indexes used to achieve efficient
access to the data, and any associated integrity constraints and security measures.
What does that mean for us? We will describe the plan for how to build the tables, including appropriate data
types, field sizes, attribute domains, and indexes. The plan should have enough detail that if someone else were
to use the plan to build a database, the database they build is the same as the one you are intending to create.
The conceptual design and logical design were independent of physical considerations. Now, we not only know
that we want a relational model, we have selected a database management system (DBMS) such as Access or
Oracle, and we focus on those physical considerations.
Logical vs. Physical Design: Logical database design is concerned with what to store; physical database
design is concerned with how to store it.

Although we have discussed user requirements, conceptual design, logical design, and physical design as if they
are sequential activities, they do not occur in isolation. In reality there is often feedback between physical
design, conceptual/logical design, and application designall of which are aimed at meeting the user
requirements. For example, decisions made during physical design to improve performance may include
merging tables together which in turn should be reflected in the logical model and will have an effect on the
application design.
Meeting the needs of the users is the gold standard against which we measure our success in creating a
database.

Underlying Concepts
Because physical design is related to how data are physically stored, we need to consider a few underlying
concepts about physical storage. One goal of physical design is optimal performance and storage space
utilization. Physical design includes data structures and file organization, keeping in mind that the database
software will communicate with your computers operating system. Typical concerns include:
Storage allocations for data and indexes
Record descriptions and stored sizes of the actual data
Record placement
Data compression, encryption

DB-related physical components & concepts:

Kernal: software internal to the DBMS
o Controls storage, retrieval, security
o Interacts with host services (functions provided by the
operating system, like how to access your CPU, how to use your
computers RAM)
o Data Dictionary: defines data structures used by the kernal (the data dictionary contains the
database meta-data). There is an external data dictionary (which we will write) and an internal
one that is maintained by the DBMS when we create DB structures.
o Database Engine: software that reads the data dictionary, accesses data.

Physical Design Steps (Methodology)

The process of describing the DB implementation on secondary storage can include all the following. We
will not practice all of the steps in our final project.

Involve the users! (Throughout the process)

Translate logical data model for target DBMS
o Design base relations (tables)
o Determine how to represent derived data
o Design general constraints (methods to
enforce business rules)

Design physical representation file organization

and indexes
o Analyze transactions
o Choose file organizations
o Choose indexes
o Estimate disk space requirements

Design user views

Design security mechanisms
Consider the need to introduce controlled
redundancy (denormalization)
Monitor and tune operational system *

For the final project: We will create

a data dictionary

For the final project: We will

identify one index in addition to the
PK

We will discuss these briefly but not

implement them.

* not really design done after the system is built. But typically a PLAN for this is developed.

Detail of Physical Design Methodology

User involvement
Users should be involved in ALL phases of the database development process. We clearly needed users to
define the system requirements. We went back to the users to clarify the entities, relationships, and cardinality
when creating the ERD. Occasionally we need further clarification from users during logical design and
normalization (Will you ever sort by name [Do I need to split name into last and first]?).
Similarly, we may need to back to the users during physical design, especially to clarify enterprise constraints
(business rules). We need to consider the format of data in the system. For example, data that appear to be
numbers might not be best stored as numbers. ICD-9 (or ICD-10) codes often contain a 0 at the beginning or the
end of the code, and that 0 is meaningful. 123.0 is different than 123 and 012.1 is different than 12.1 {not real
ICD codes!}. Similarly, consider social security numbers, zip codes, student ID numbersleading or trailing 0
may be important in those as well, and those leading/trailing 0 would be LOST if we stored this as a number
data type. Such data are best stored as text even though they LOOK like numbers. My rule of thumb is to ask
myself if I would be doing math with the number. If not, consider whether it needs to be stored as text data.
Translate Logical Data Model
Sohow do we go about selecting the DBMS?
Consider potential size of the DB, number of users, type of interfaces. Access is readily available as part
of MS office, for example, but it is Windows only there is no native MAC version of MS Access.
Mac users can sometimes work around this, for example, by using parallels or by logging in to a virtual
PC such as a citrix/metaframe client. But its something to consider.
Consider cost. There are some nice, free, open-source databases available
Consider trade-offs (e.g., ease of use or cost vs. lack of features). Many open-source/free DB come with
little or no support or documentation, or may lack certain desired features.
Consider organizational constraints (e.g., your organization may mandate certain DBMS)
We need to know the functionality of the target DBMS. Considerations include HOW to create tables, how the
system defines and deals with keys (not only the primary key [PK], but also foreign keys [FK] and
alternate/candidate keys [AK]), whether the system supports required (not null) constraints. Other
considerations or whether the system supports: definition of domains (allowed values), enforcement of
relational integrity constraints, and enforcement of other constraints.
An important consideration is: What are the data types allowed in the DBMS?
(We will explore Access and Oracle data types soon)

Design Tables (the book calls this design base relations)

Decide how to represent the data tables (called base relations in the formal definition) in the target DBMS.
From the logical design, we have a set of table schemas that describe
the name of the table
the name of stored attributes (the columns).
the PK and, where appropriate, AK and FK
We also might have
referential integrity constraints for FKs (what table does the foreign key link to)
assumptions

Now well add additional information. For each attribute, we will specify:
domain: consisting of data type, length, and constraints on the domain (well talk more about constraints
soon)
whether the attribute can hold nulls
default value for the attribute (optional)
any derived attributes and how they should be computed
We will create a data dictionary for the final project. More about this later in the module, but a large
component of the data dictionary is the table design. I like to organize this into Word or Excel tables, like
shown below for the Demog table in the SNDB, built in MS Access
Demog Table
Attribute Name

Description

Data Type

pt_id
name

Patient identifier.
Patient name

Text
Text

zip
Gender

Patients zip code

Patient gender

Text
Text

Race
Patient race
Primary Key: pt_id

Text

Data
Domain
Length
4
Any positive integer
20
Formatted as
last name, first name
5
99999
1
Coded in the
listgender table
1
Coded.

Allow
Null?
No (PK)
Yes
Yes
Yes
Yes

Notes about the example

You could make the last column required instead of allow null?, in which case each value would be
exactly reversed. The primary key field is always requiredif the PK is multiple columns, all the PK
columns are required.
You can indicate the primary key within the table, or below the table. You dont need to do both. If the
description is obvious from the attribute name, you can leave that blank. In many cases, though, a
description is helpful.
Attribute name is same meaning as variable name or column name. You can use any of these names
for that first column.
Domain can be descriptive (any positive integer), a pattern or example (99999), a list of specific
values, reference to a list of values, etc.

PHYSICAL DESIGN IS DBMS SPECIFIC -- Naming

Note that I specified the table design above is applicable to MS ACCESS. Access is very forgiving when it
comes to things like names. You can have spaces, you can pretty much do what you want.
Other DBMS, such as Oracle, are more strict. Most DBMS, in fact, have naming restrictions similar to those in
Oracle. Attribute names must begin with a letter, you cannot have spaces (although you CAN have underscore).
You cant use reserved words such as the name of functions (max) or datatypes (date).
So one step you might need to take at this point is to check your table and attribute names. An easy way to
fix issues with reserved words is to append some meaningful prefix or suffix, like changing date to test_date or
visit_date. Sometimes you can just make a name plural, for example, Oracle does not allow comment as a
column name, but will allow comments.
PHYSICAL DESIGN IS DBMS SPECIFIC Data Types
Although there is a reasonably small set of primitive data types (numbers, letters, images, etc), the way that
different DBMS deal with those data types, the names they use for the data types, and constraints regarding the
data types vary.
Alphanumeric data
This is a generic data type that includes letters, numbers, symbols the sort of characters we are used
to working with in a word processor.
Oracle calls this type of data CHAR for fixed length fields (always stores the full length allowed
anything not filled in with text will be filled in with spaces); it uses VARCHAR2 for variable length
fields (stores the number of characters typed). There is a VARCHAR data type for variable length fields
as well, but that is likely to be deprecated in future releases, so VARCHAR2 is preferred over
VARCHAR.
MS Access calls this type of field TEXT. It is always variable length there is no fixed length character
data type in Access.
Other systems may call this type of field string, character or other names.
With both Oracle and Access, you must specify a field length. This is the total field length for CHAR
(Oracle), and is the MAXIMUM field length for VARCHAR, VARCHAR2 (Oracle) and TEXT
(Access). In determining DB size, each character is 1 byte. (Unless you are using a 2-byte character data
set like Chinese, which requires a specific setting on your operating system and within the DBMS).
So VARCHAR2(25) [Oracle] or TEXT(25) [Access] is an alphanumeric field that can hold UP TO 25
characters, and will take up to 25 bytes of space per row.
Numeric data (numbers)
Oracle uses number (p,s) format for numbers; where p is the precision (how many digits total) and s is
the scale (how many of the p digits are after the decimal point). So a number like 9.25 would be
represented as number (3,2) there are 3 digits, 2 of which are after the decimal. Some versions of
Oracle allow you to specify numbers by type, such as integer, long, etc but that practice is discouraged
for two reasons: (1) those typed numbers take the maximum allowable space for the type usually more
space than you need, and (2) the functionality may go away in future versions.

Access uses number types, though--integer or long integer for whole numbers, and single or double for
numbers with a decimal component. Access also has a currency data type (all numbers and math done
with the number use only 2 digits, like when computing with money).
Access has an autonumber data type, which is a LONG INTEGER that is automatically generated by
the DB. You can have the system increment (count up sequentially) or generate a pseudo-random
number. I say pseudo random because its really based on a computation that is applied to the time
on the system clock.
Oracle is trickier for auto-generated numbers. You need to create a separate counter table, and retrieve
the number from that table with a trigger that pulls the number and then increments the counter (triggers
are code that run automatically).

For a nice description of Oracle data types, including the size associated with different data types, see
http://www.ss64.com/orasyntax/datatypes.html
For a nice description of MS Access data types, see
http://www.databasedev.co.uk/fields_datatypes.html

Constraints: Limits on the data

Keeping in mind enterprise constraints (business rules), determine:
required data i.e. NOT NULL
relational integrity constraints (referential integrity)
domain for the data.
The domain includes physical aspects defined with the attribute (data type, field size). It also includes
constraints imposed by the business rules. Some of these constraints can be implemented with physical
measures within the database, others might be managed by the application (user interface), still others might be
managed through policy and procedures. The domain includes ALL of these physical constraints, application
management, and policy/procedural constraints.
Lets look at some of the physical measures we can use to enforce constraints.

Not Null: The field is never allowed to be empty. Data must be entered at the time the row is created.
This is not the same as data that are needed eventually--requiring that data might be a policy measure.
Keep in mind that if you specify not null as a constraint, you cannot save the row of data if that
column is blank.

Unique: No duplicate data when you look across rows. A unique constraint can be used for a candidate
(alternate) keys (Access calls this the no duplicates property).

Primary Key: Adding the primary key constraint automatically includes not null constraint + unique
constraint (+ an index).

Foreign Key: using the foreign key constraint enforces referential integrity. Every foreign key must
match a primary key or a unique constraint on another table. The sequence of data entry is affected here
- you must enter data into the main table before it can be used as a foreign key. (Access uses
properties within the relationships window to enforce referential integrity. Oracle uses references
statement within the table constraints).

Check. A check constraint helps to enforce attribute domains.

o Boolean logic (Check SBP >=0 and SBP <=350 or Check SBP Between 0 and 350). This
would be evaluated as true or false; if false you get an error message and can't store the row of
data.
o The check constraint is called a validation rule in Access.
o The above rule would also act like a NOT NULL statement. Why? Because null data are not
between 0 and 350. How would you correct this to allow null values? By adding "...or is null"
Limits on check constraint: Cant use a comparison (such as percent of the total) or a column that varies
over time (such as sysdate [the current date on your operating system clock)]. Use triggers, instead, if
you want to enforce this sort of cross field or complex check.

Oracle example when you create the table, you could list the check constraint in the create table statement:
Create table vitals ([list of attributes and data types],
SBP Number (3) check SBP between 0 and 350,
In Access you would create the SBP column in table design view, then in the properties for this column, create
the validation rule:

Some DBMS provide more facilities than others for defining enterprise constraints. An example of a complex
constraint that might work in Oracle, but that would not work in Access:
CONSTRAINT StaffNotHandlingTooMuch
CHECK (NOT EXISTS (SELECT staffNo
FROM PropertyForRent
GROUP BY staffNo
HAVING COUNT(*) > 100))

Triggers: These are stored procedures (code) that fire automatically when data are manipulated-when there is
an insert, update, or delete statement. Triggers can call (use) other code modules. Constraints are faster than
triggers but triggers can be more complex.
In Oracle, triggers are created within PL*SQL (the internal Oracle programming language).
Access does not allow triggers on tables. If you use forms, you can emulate the action of triggers by writing
code inside the form.
More about domain constraints
Domain constraints are limits on the data values. What business rules can we implement in some physical
manner? For example, suppose our business rule says that a valid SBP must be between 0 and 350. We can do
this with a check constraint (Oracle) or validation rule (Access) as discussed above.
You might enforce business rules in the user interface instead of at the table level: e.g., use a picklist of values
and restrict users to only choosing something from the picklist. This would work nicely when the list of choices
is known. For very small lists, you might consider coding the data numerically, and using radio buttons, like:
Gender:

Male Female

where Male = 1 and Female = 2

*radio buttons can only store NUMERIC data. Each choice is given a number code, and the user is allowed to
choose only ONE of the choices like the channel selection buttons on a car radio.
Check boxes only store true/false or yes/no values (sometimes called Boolean data). In Access there is a yes/no
data type. Yes (or true) is stored as -1 and no (or false) is stored as 0. Most versions of Oracle do not allow
a Boolean data type you would use a number field of size 1 and store the values, but youd have to know what
the numbers mean. A few DBMS allow a three-valued boolean like field, equating to yes, no, and unknown.
So radio buttons are an option for pick ONE of the following. Check boxes typically imply that any or all of
the boxes may be chosen you store a yes or no for each choice.
See how the application interface design might influence some of your physical design decisions?
Some business rules can only be enforced through policy/procedure. Consider the rule FirstName must be a
valid first name. This must be enforced through policy/procedure. With physical constraints, we can only
specify that this is text, and the length of the text, but we cant determine if the text is a valid first name.

Derived Data
Examine the user requirements, ERD, logical data model and data dictionary, and produce a list of derived
attributes. Derived attribute can be stored in database or calculated every time they are needed. Document what
is to be done! Your choice may be based on: space to store the derived data, effort to keep it consistent with the
data from which it is derived; versus cost (query time) to calculate the value each time it is required. Less
expensive option might be chosen subject to performance constraints.

Code Tables (Lookup tables)

Code tables list a code and the meaning of the code, like:
ShipCode Shipper
1
UPS
2
FedEx
3
USPS ground
In a main data table, you would store the CODE. Then you can look up the value.
The SNDB listGender table is another example.
When we build the interface for out DB, we can use code tables or look up tables (single column list of
choices) to populate list boxes and combo boxes. This can simplify data entry for your users those interface
boxes can be set to display the meaning to the user, but store the code in the table.
In your final project I ask you to create a code table for ONE column somewhere in the DB. Include that code
table in your data dictionary, and list it in the domain column of where it will be used. See the gender
description in the Demog table on the next page.

At this point, you can BEGIN your final project data dictionary. For each table, include
Table name
Attribute names
Description of the attribute if not apparent from the name
Data type must be consistent with the DBMS you will use to build the database
Length of the field. Use the data type links above to see the choices for length
Indicate the Primary Key
If appropriate, also list alt key(s) and foreign key(s)
Derived attributes and how to compute
You can do this in Word, with a separate Word table for each DB table, like the example below or similar
appropriate format
Demog Table
Attribute Name

Description

Data Type

pt_id
name

Patient identifier.
Patient name

Text
Text

zip
Gender

Patients zip code

Patient gender

Text
Text

Race
Patient race
Primary Key: pt_id

Data
Domain
Length
4
Any positive integer
20
Formatted as
last name, first name
5
99999
1
Coded in the
listgender table
1
Coded.

Text

Allow
Null?
No (PK)
Yes
Yes
Yes
Yes

Or you can do this in Excel, something like:

Table
Demog

Adm_dx

Attribute
Name
pt_id

Description

Data Length

Domain

Allow Null?

Patient identifier.

Data
Type
Text

No

name

Patient name

Text

20

zip
Gender

Patients zip code

Patient gender

Text
Text

5
1

Race

Patient race

Text

Any positive
integer
Formatted as
last name,
first name
99999
Coded in the
listgender
table
Coded.

Text

No

PK

Text

60

matches
demog table
medical
diagnosis

No

PK

pt_id
diagnosis

admitting diagnosis

Key/
Index
PK

Yes

Yes
Yes

Yes