CMSC 424
Database Design
Section 401
Dr. David Kuijt
Contact Info
Professor: David Kuijt
Office: AVW 3205
Phone: 5-0534
Office Hours: T/Th 3:15-4:00
(or by appointment)
TA: Debbie Heisler
Office: AVW 3270
Phone: 405-7027
Office Hours: TBA
Basic Information
• Required text:
Korth & Silberschatz Database System Concepts, Fourth Edition,
McGraw Hill 2001.
• Warnings:
– Late homework or projects are not acceptable. Hand in what you
have finished. Exceptions will be made only for emergencies or
medical reasons with a doctor's note.
– No makeup exams. Exceptions as above.
– Cheating will result in an immediate grade of XF ("failure through
academic dishonesty" -- this goes on your permanent transcript),
and may result in suspension or expulsion from University. This is
your only warning. Don't do it.
We live in a database world. The simplest acts are tied to
databases. The last time I called out for delivery pizza, it
involved at least four enormous databases. What were
• Pizza Hut knew what I had ordered before -- they asked if I
wanted the same pizza as last time. They probably store
lots more information than that -- perhaps all my old order
information. They could use this information to make
corporate decisions (quantities of materials to order;
forecasting pizza trends) as well as a local aid.
Motivation (2)
• Every delivery food place around here uses caller ID.
That's a relatively simple database, just giving names and
incoming telephone number, but it helps them avoid some
types of fraudulent orders and errors when writing down
names, addresses, and stuff.
• I used a telephone. When you use a telephone, all the
details about the call are stored in a database. Call length,
what number you called, time of call, billing information,
and so on. Cell phone databases are even more complex.
Motivation (3)
• I paid by credit card. Huge databases are involved.
• Every time somebody runs your credit card through a
swipe reader or types in the number, they're checking
information in a database.
Is this a valid credit account?
Does it have enough money to cover the bill?
Is the credit card stolen?
Debit the merchant account, credit the consumer account
• Whether the transaction is accepted or rejected, all the
details are recorded in a database somewhere.
More Motivation
• Every time you go to an ATM, use a credit card, buy something with a
UPC bar code at a supermarket or department store, go to a movie,
concert, or Caps game, register for classes, or get a parking ticket in
Lot 4, you are working with databases.
• Everything in your wallet that isn't a photograph is an entry in a
database somewhere
medical plan cards
credit cards
student ID
driver's license
membership cards in clubs or interest groups
everything. Even the currency!
Final Motivation
• Databases are all around us.
– Knowledge is power.
– Databases give us power (the ability to do things we couldn’t
otherwise do)
– they give other people power over us, and knowledge about what
we are doing.
– This class is about Databases.
So what is a Database?
What is a Database?
At the simplest level, a database architecture has two
(1) Data.
– Usually a whole lot of it
– Representing multiple types of different objects
– Each type may be related to itself and to other types in multiple
(2) A set of methods to access and manipulate the data.
• For any reasonable-size database the data may be quite
• It is an attempt to record or model all the aspects of the real
world that are important to one specific purpose -telephone calls, for example, or credit card accounts.
• Lots of different objects need to be stored as data, and they
need to be stored in such a way as to reflect the ways that
the objects can interact with each other.
Database Example
• For example, consider a local telephone system database.
Types of data stored include:
– Account information:
• customers: individuals, groups, companies that have leased numbers;
• billing addresses, payment history, calling plans and billing contracts;
– Hardware information:
network structure (call routing),
hardware age, reliability, and maintenance information,
system load tracking,
network billing pattern (what numbers are long distance from what
other numbers, and what ones are local)
Database Example (2)
• Local telephone system database continued. Additional
types of data would include:
– Call information:
• start and end time,
• telephone number that initiated the call;
• telephone number(s) that received the call
• All that information could be stored in files with much less
fuss and bother. Why use a database? Why not just store
the information in flat files?
Why Not Flat Files?
• Why use a database? Why not just use a flat file?
• Databases have a number of advantages over flat files.
– Data Access. The set of programs that provide access to a database
allow much more complex and flexible queries to the database
with greater efficiency and convenience.
– Reduced duplication and better control over data consistency. Data
redundancy is bad. Which item to change in an update? How do
you know that you've found all the copies? Data inconsistency
(disagreement between various copies of the same data) is a
serious problem.
– Integrity Constraints can be enforced inside a database -- telephone
numbers all 10 digits; phone numbers in Maryland all have the first
three numbers being 410 or 301.
Why Not Flat Files? (2)
– Uniform access and control of data using a standard language
– Data Independence. We want the data to be independent of the
representation chosen for it within the system. Tying the data to a
given representation is what caused the Y2K fuss -- only two digits
were used for a "year" field.
– Concurrency control. Multiple users on a single database is a big
– Recovery.
– Security. Different users of the database may need different levels
of access to information.
– Centralized Control
– Platform independence (portability). Since the internal file
structure and access program details are hidden from the user, it is
much easier to use the database on multiple platforms.
Data Abstraction
• Most users don't need to understand all the details of the
implementation and data design of a complex database. To
make a database convenient to use, the system provides
users with an abstract view of the data, limiting the
information available to them. There are usually three
levels of data abstraction.
– Physical Level
– Conceptual Level
– View Level
Data Abstraction (2)
• Physical level. The actual implementation details of lowlevel data structures are described at this level.
• Conceptual level. This level describes all the different data
types that exist by defining a relatively small number of
simple structures, including all the relationships that these
data types have with each other. Implementation of these
objects might be complex, but it is hidden from the user at
this level. Database administrators are usually the only
ones who have access at this level.
Data Abstraction (3)
• View level. There may be multiple different views, each of
which represents a simpler subset of the functions and data
available at the conceptual level. Different user types may
require different parts of the database (for example, a bank
account database might be accessed by cashiers, account
holders, credit card companies, and the bank's payroll
manager. Each of them can only access a small part of the
full database of bank account information). Creating a
number of restricted views makes the database more useful
for the individual user types, giving each type access
according to the needs of that type.
Data Abstraction (4)
• Definition: a Schema is a
specification of a particular
database using a particular data
The three levels of data abstraction
are often referred to as:
• External Schema(s) (for the
view level(s)).
• Conceptual Schema (for the
conceptual level)
• Internal Schema (for the
physical level)
Database as Model
• A model represents a perception of a real system
• Models help us manage or understand the real world
system they represent.
• When modeling a system we select aspects and
characteristics we want to represent; we abstract them to
form a simple(r) system
– examples: a map, an airplane flight simulator, computer weather
analysis program
• A database is a model of reality
Data Models underlying the Database
• The data model is a collection of conceptual tools for
describing data and its attributes
– data objects
– interrelationships of the data
– data semantics and consistency constraints
• There are two well-established data models used in
database design
– Entity-Relationship (E-R) model
– Relational model
– older methods included the Network and Hierarchical data models
• Each was tied closely to the underlying implementation, which made
it more difficult to model data and to modify or update the database.
As a result they aren’t much used any more
Entity-Relationship Model
• Diagram based model
• Two primitives
– Entities -- each represents a unique real-world object
– Relationships -- each represents an association among several
– Each are associated in sets of the same type (for example, one
entity set might be customer, representing the set of all entities that
represent customers at a given bank)
• Third important notion: Attributes
– Entities are associated with a set of attributes
Diagram-based Model
• Entity: a distinguishable object we want to model
– e.g., room CSI 3120, Celine Dion, Elizabeth I of England
• Entities have attributes (single-valued properties)
– e.g., a person has a name, SSN#, gender, …
– if an attribute has more than a single value, we should model it as
an Entity
• Entity Set: a set of entities of the same type
• Entity Sets may overlap
– CSI 3120 is a member of CLASSROOMs and also a member of
• Relationship is an association among entities
– David Kuijt teaches-in CSI 3120
• Relationship Set is a collection of relationships of the same
• Relationships may also have attributes
– e.g., the relationship teach-in has an attribute “weekday” and
another attribute “time” to store the day and time in which a given
Entity of the set FACULTY teaches in a given Entity of the type
Example Database Design (1)
• Application: library database. Authors have written books
about various subjects; different libraries in the system
may carry these books.
• Entities (with attributes in parentheses):
Authors (SS#, name, tel, birthdate)
Books (ISDN, title)
Subjects (sname)
Libraries (lname)
• Relations [associating entities in square brackets]:
– Wrote-on [Authors, Subjects]
– Carry [Libraries, Subjects]
– Index [Subjects, Books]
Diagram of Initial Database Design
Poor Initial Design
• Our first design is a poor model of the real-world system
we are examining. Problems in our first design:
– no relationship associating authors and books
– no relationship associating libraries and books
– common queries will be complex and difficult:
Q: what libraries carry books by a given author?
Q: what books has a given author written?
Q: who is the author of a given book?
Q: how many copies of a given book exist at each library?
Q: what edition of a book does the library have?
Example Database Design (2)
• Application: library database as before
• Entities (with attributes in parentheses):
Authors (SS#, name, tel, birthdate)
Books (ISDN, title)
Subjects (sname)
Libraries (lname)
• Relations [associating entities in square brackets]
(attributes in parentheses):
– Wrote [Authors, Books]
– In-stock [Libraries, Books] (quantity, edition)
– Index [Subjects, Books]
Diagram of Improved Database Design
• Fundamental concept for databases
• Must be able to uniquely identify things within a database
(in the E-R model, Entities and Relationships)
– Avoid duplication of results in a search; identify data redundancy
in other operations
– Halt search on positive results
– Quick lookup in underlying data structures used at the Physical
Level of abstraction
• Examples of possible keys
– Student ID number (SS#) is used as a key for most UMD databases
having to do with students
Entity Keys
• Superkey: set of attributes whose values uniquely identify the entity
• candidate key: a minimal superkey (a minimal subset of a superkey
whose values still uniquely identify the entity)
• primary key: if there is more than one possible candidate key, one is
chosen as the primary one used for most entity-identification purposes
• weak entity: has no primary key; instead it depends upon another
strong entity’s primary key to exist
– e.g., CHILDren of EMPLOYEEs are weak; the primary key of
EMPLOYEE in addition to the attributes of the CHILD are used
for identification
– weak entities are “existent dependent” on a strong entity -- when
the strong entity gets deleted, so does the weak one
Relationship Keys
• Depend upon the entity mapping of the relationship
– one-one: the primary key of any of the entities can be used to
uniquely distinguish a given relationship between two unique
– one-many: the primary key of the “many” entity, plus possibly a
subset of the attributes of the relationship, will uniquely identify a
given relationship
• e.g., MOTHER gave-birth-to CHILD; to identify a specific gavebirth-to relationship requires the primary key of MOTHER and
possibly the (date) and (time) attributes of gave-birth-to
– many-many: the union of the primary keys of the entities
associated, plus possibly a subset of the attributes of the
relationship, will uniquely identify a given relationship
• e.g., PERSON married PERSON; SS# of both and possibly date
Special Cases
• Relationships may associate different entities of the same
• Ternary versions of the above
• M-N relationships: many-one mappings are often more
useful in practice than many-many mappings.
– DUMMY Entities can be used to convert an M-N mapping
relationship to a pair of relationships, one M-1 and one N-1.
(ISA Hierarchy)
• This is a way to represent entity complexity
• specialization: top-down refinement of entities with
distinct attributes
– Entity type BANK ACCOUNT might be subdivided into related
but different types CHECKING ACCT and SAVINGS ACCT
• generalization: bottom-up abstraction of common attributes
– Course types DATABASE, SYSTEM, and NETWORK all have
common attribute (project). From them we can abstract a new
– other common course attributes are included (e.g., course number)
ISA Hierarchy Example:
Top-down Refinement
• Account entity with
attributes balance and
• additional complexity: we
want to represent two
subtypes of account
– Savings Account with
attribute Interest Rate
– Checking Account with
attribute Overdraft Limit
ISA Hierarchy Example:
Bottom-up Abstraction
• Three related entities with
similar attribute project
• we abstract a new type of
super entity Practical
Course and link the three
entities as subtypes
• other shared attributes
(e.g., course number) are
also promoted to the upper
level entity
(Part-of Hierarchy)
• This is a way to represent relationship complexity
– relationships among relationships are not supported by the E-R
– often we want to model lower-level relationships differently
• Groups of entities and relationships can be abstracted into
higher level entities
Part-of Hierarchy Example
• Entities driver, car, tires, doors,
engine, seats, piston, valves
• Relationship drives is
insufficient to model the
complexity of this system
• Part-of relationships allow
abstraction into higher level
entities (piston and valves as
parts of engine; engine, tires,
doors, seats aggregated into
Mapping an E-R Schema to Tables
• Motivation - translating E-R database designs into
Relational designs
– Both models are abstract, logical representations of a real-world
– Both models employ similar design principles
– Converting an E-R diagram to tables is the way we translate an ER schema to a Relational schema.
– Later on we’ll examine how to convert a Relational schema to an
E-R schema
Mapping an E-R Schema to Tables (2)
– Strong Entity E with primary key PK and attributes A, B, …
==> E(PK, A, B, …)
– Weak Entity F with (non-primary) key WK and attributes C, D, …
depending upon E above for primary key
==> F(PK, WK, C, D, …)
– Relationship R with attributes L, M, … and associating Entities E
(with primary key PK), E2 (PK2), E3 (PK3), …
==> R(PK, PK2, PK3, …, L, M, …)
– Relationships between weak entities and the strong one on which they are
dependent usually do not require representation because it is usually a
many-one relationship with no attributes on the relationship (they are on
the weak entity) and so the resulting table R(PK, WK) is a subset of the
weak entity itself.
Table Details
• The whole table represents a single Entity Set or
Relationship Set.
• Each entry (row) in the table corresponds to a single
instance (member in that set)
• For an Entity Set each column in the table represents an
attribute in the E-R diagram
• For a Relationship Set each column in the table represents
either an attribute of the Relationship or one of the parts of
the primary key of the Entity Sets it associates
Mapping an E-R Schema to Tables (3)
– ISA relationships: choose either to
• Represent the super class entity, then represent each subclass with the
primary key of the super class and its own attribute set. This is very
similar to the way weak entities are treated.
• Or, map the subclasses to separate relations and ignore the whole
super class. This is good when the subclasses partition the whole
superclasses between them (the subclasses are disjoint and the union
of the subclasses covers the whole super class).
– Aggregate (part-of) relationship
• Translation is straightforward -- just treat the aggregate as an entity
and use the methods defined above.
– With last week’s lecture, this covers the material of chapter 2.
Relational Database Model
• Most popular logical data model
• Relations (also called tables) represent both Entity Sets
and Relationship Sets.
– Attributes form the columns of the table (column and attribute are
– Each row represents a single entity or relationship (called a row or
• Each instance of an attribute takes values from a specific
set called the domain of the column (the domain defines
the type)
Relational Database Model (cont)
• A relation schema is made up of the name and attributes of
a relation with their underlying domains
• A database schema is a set of all relation schemas.
• The notions of keys, primary keys, superkeys are all as
previously described
Query Languages
• a language in which a user requests information from the
– a higher level language than standard programming languages
• query languages may be procedural or non-procedural
– procedural languages specify a series of operations on the database
to generate the desired result
– non-procedural languages do not specify how the information is
– most commercial relational database systems offer a query
language that includes procedural and non-procedural elements
Relational Algebra
• procedural query language
• set of operators that map one or more relations into another
• closed algebraic system
– best feature - operations on operations
– form relational algebraic expressions
• two types of operations: set-theoretic and database specific
Relational Algebra Operations
• database specific:
(horizontal) selection ()
(vertical) projection ()
outer join
• set operators
cartesian (cross) product
Example Relations
ename salary
Shirley 35K
Christos 37K
Robin 22K
DEPT dept
Database Specific Operators
• (horizontal) selection ()
– picks a subset of the rows
• (vertical) projection ()
– picks a subset of the columns
• join
– creates a new relation (table) out of two
• equijoin (based upon equality of attributes)
• natural join (equijoin plus projection to eliminate duplicated columns)
Set Operators
• union
– both relations must be union-compatible -- same degree and same
• set difference
– both relations must be union-compatible as above
• intersection
– same deal
• cartesian (cross) product
– note similarity to join operation; join can be defined as a cross
product followed by a selection criteria
More Operators
• rename ()
– results of operations in the relational algebra do not have names
– it is often useful to be able to name such results for use in further
expressions later on
– conceptually similar to an assignment operator in most
programming languages
• semijoin
– very useful in practical implementation of large queries
– semijoin of R and S is equivalent to the join of R and S projected
onto the attributes of R.
Still More Operators: Outer Join
• outer join is an extension of the join operation to deal with
missing information
– three types: left outer join, right outer join, and full outer join
– left outer join computes the natural join, then takes all tuples
(rows) in the left relation that did not match on the join attribute
and includes them in the result, with all attributes of the right
relation padded with null values
– right outer join is the same, except non-matching tuples in the right
relation are included in the result padded with null values
– full outer join includes all non-matching tuples of both relations
appropriately padded
– see examples in text, p108-109
Still More Operators
• division
– R/S: given R(A,B) and S(B), then a given tuple t is in R/S if for all
s in S there exists an r in R such that r.B=s.B and t.A=r.A.
– So tuple t with attribute t.A is in the result if and only if R
contained tuples (t.A, B1), (t.A, B2), (t.A, B3), … for every possible
value Bi contained in S.
– Note that S must be defined on a subset of the attributes of R for
the operation to be meaningful.
A Short Interlude: Integrity
• the preceding slides covered chapter three up to section 3.3
• before attacking chapter 4 (SQL), we’re going to make a
brief excursion up to chapter 6, touching sections 6.1 - 6.4
• Integrity constraints attempt to enforce data consistency
and prevent accidental damage to the database during
• We’ve already seen two forms of integrity constraints:
– key declarations (stipulating that certain attributes form a candidate
key for a given entity set)
– mapping form of a relationship (one-one, one-many, many-many)
Integrity Constraints
• Domain Constraints
– simplest form of integrity constraint
– type declarations are one such domain constraint (e.g., integer,
floating point, double-precision, fixed length character string).
– domains can be further restricted (e.g., check clause in SQL can
ensure that hourly wages are  4.00 dollars)
– easily tested whenever a new data item is entered into the database
– extensions like date or currency can be easily supported on a
strongly typed programming language
– Null values can be useful for values to be filled in later, but some
attributes may need to be specified as “not Null” (e.g., primary
keys cannot have a null value)
Integrity Constraints (2)
• Key Constraints
keys must have unique values
primary key -- a candidate key declared primary
unique key -- a candidate key
foreign key -- a set of attributes that are a primary key for some
other relations
• foreign keys are an important concept because we need to treat
foreign keys differently from other attributes (for example, protecting
their uniqueness and insuring referential integrity) even though they
aren’t a primary key in the current relation
Referential Integrity
– We often want to be able to ensure that an attribute value in a tuple
of a relation appears in at least one tuple of another relation. For
» EMP(eno, ename, salary)
» DEPT(dno, dname, floor)
» WORKS-IN(eno, dno, hours)
– note that eno is a foreign key in WORKS-IN
– We want the following to be true:
• eno(WORKS-IN)  eno(EMP)
• dno(WORKS-IN)  dno(DEPT)
(every eno is a real employee)
(every dno is a real department)
– SQL allows the declaration of domain/key/referential integrity
constraints with the clause check in its DDL
Referential Integrity: SQL DDL Example
Create table customer
char(20) not null,
primary key
Create table branch
primary key
char(15) not null,
(assets 0))
Create table account
char(10) not null,
primary key
foreign key (branch-name) references branch,
(balance 0))
Create table depositor
char(20) not null,
char(10) not null,
primary key
foreign key (cust-name) references customer,
foreign key (account-no) references account)
Referential Integrity and Database Modifications
• Database modifications may violate referential integrity
• Insertion: inserting a value into the referencing relation
that is not in the referenced relation
• Deletion: deleting the last example of a given value in the
referenced relation and leaving that value in the
referencing one
– proper handling may lead to cascading deletions
• Update to the referencing relation (constraints as Insertion)
• Update to the referenced relation (constraints as Deletion)
• An assertion is an arbitrary expression that the database
must always satisfy
– e.g., student GPA > 2.8, or sum(all-charges) < credit-line
– Domain constraints and referential integrity constraints are special
forms of assertion that are easy to test
– SQL supports assertions as follows:
create assertion <assertion-name> check <predicate>
– When an assertion is made the system checks it for validity. If it is
validated, every future modification of the database is checked
against the assertion and allowed only if it is not violated.
– This can be very expensive if assertions are complex or numerous
• A trigger is a statement that the system executes automatically as a side
effect of an update to the database.
• A trigger has two parts:
– condition under which it is executed
– actions to be taken if it is executed
• Example: instead of having an assertion “balance 0” for a checking
account, use a trigger on negative balances that sets the balance to zero
and creates a new loan for the amount of the overdraft
• Triggers make the system reactive
• Triggers are also called active rules
• Like Assertions, Triggers can be very expensive.
Trigger Example
define trigger overdraft on update of account T
(if new T.balance < 0 then (insert into borrow values
(, T.account-number,
T-customer-name, - new T.balance)
update deposit S
set S.balance = 0
where S.account-number = T.account-number))
(note: SQL syntax given here is slightly different from that in the text, p235)
SQL (Structured Query Language)
(Astrahan, Gray, Lindsay, Selinger, …)
• Most common and influential commercial query language; well
established as the industry standard query language for relational
• Developed (as “Sequel”) at the IBM Research Lab in San Jose in the
early 70s
• Four basic commands
– select
– insert
– delete
– update
• Result of each query is a relation
SQL Example
emp e
e.age > 30;
• e is a tuple variable ranging over the emp relation
• a tuple variable followed by a “.” and an attribute is an indexed tuple
variable and specifies the corresponding attribute of the tuple, very
similarly to in many programming languages
• what follows the keyword select is the target list
• what follows from is the tuple variable list and consists of a list of
relations and variable names
• what follows where is the qualification clause; an arbitrary boolean
• Basic format of the select command
select [distinct] target_list
from tuple_variable_list
where qualification
[order by target_list_subset];
• Semantics
– evaluate qualification: select the subset of the cartesian product of the
ranges of the tuple variables that satisfy the qualification
– evaluate target list: eliminate columns that are not in the target list
– prepare the result as a relation with columns according to the target list
– if distinct is used, eliminate duplicate tuples
– if order by is used, sort the result accordingly
SQL: some example queries
• We will give a number of simple query examples using the following
relational schema:
sailors(sid, sname, rating)
boats(bid, bname, colour)
reserve(sid, bid, date)
(1) Find the names of sailors who have reserved boat #2
sailors s, reserve r
s.sid=r.sid and
SQL: example queries (2)
(2) Find the names of sailors who have reserved a red boat
sailors s, reserve r, boats b
s.sid=r.sid and and b.colour=“red”
(3) Find the colours of all boats reserved by Pat
sailors s, reserve r, boats b
s.sname=“Pat” and s.sid=r.sid and
SQL: example queries (3)
(4) Find the names of sailors who have reserved at least one boat
sailors s, reserve r
(5) Find the names of sailors who have reserved a red or a green boat
sailors s, reserve r, boats b
s.sid=r.sid and and
(b.colour=“red” or b.colour=“green”)
SQL: example queries (4)
(6) Find the names of sailors who have reserved a red and a green boat
sailors s, reserve r, boats b, reserve r2, boats b2
s.sid=r.sid and and b.colour=“red”
and s.sid=r2.sid and and
Note: in the above query if sailor Pat has reserved one green boat and two
red ones, the name Pat will appear twice in the results. To avoid that,
use the keyword distinct in the select line, as in:
select distinct s.sname
• Basic format of the select command
select [distinct] target_list
from tuple_variable_list
where qualification
[order by target_list_subset];
• Simple query examples use this relational schema:
sailors(sid, sname, rating)
boats(bid, bname, colour)
reserve(sid, bid, date)
SQL: target list
• “*” is an abbreviation for all attributes in the from list
sailors s
order by s.rating
• Each item in the target list can be as general as attribute_name =
expression, where the expression is any arithmetic or string expression
over indexed tuple variables and constants. It can also contain some
built-in functions like sqrt, sin, mod, etc. as well as aggregates (coming
up later)
SQL: target list expression example
With rating an integer from 1 to 10, this query gives a rating bonus to
sailors who sailed two different boats on the same day.
s.sid, s.sname, rating=s.rating+2
sailors s, reserve r, reserve r2
s.sid=r.sid and s.sid=r2.sid and and !=
What’s wrong with the above?
– What happens if s.rating = 9 before this query?
– Domain constraints might take care of this, but we need to be
Qualifications: each item in a qualification (where clause) can be as
general as expression=expression
name1 = s1.sname, name2 = s2.sname
sailors s1, sailors s2
2*s1.rating = s2.rating-1
Further elaboration:
• tuple variables can be implicit if the system can figure out which
relation each attribute belongs to
• table names can be used as tuple variables
Example: find names, ages, and departments of employees who are over
40 and work on the first floor.
ename, age, emp.dname
emp, dept
age>40 and floor=1 and emp.dname=dept.dname
SQL provides set operators: union, intersect, and minus
Example: find the names of employees who work in the toy department
and make at most 60K
Note that it is usually possible to phrase a single query in multiple ways.
The previous query could have also been written:
Or also (even simpler):
dname=“toy” and sal60K
Writing a query in different ways will usually change how efficient the
query is -- the above query is very likely to be faster than the example
using intersects, and that one is likely to be faster than the one using
SQL also provides set operators: contains (a set being a
superset of another) and exists (a set not being empty).
Both return Boolean results, so may be negated (using
Example: find the names of employees who manage all the departments
on the first floor.
select mgr
dept d1
where (select d2.dname
from dept d2
where d1.mgr=d2.mgr)
(select dname
from dept
where floor=1)
SQL allows nested queries using the keyword in
Example: find the names of employees who work on the first floor.
select ename
from emp
wheredname in
(select dname
from dept
where floor = 1)
The same query in flat form is
select dname
from emp, dept
where emp.dname=dept.dname and floor=1
The connective in tests for set membership. Similar connectives are:
– not in (set non membership)
– op any (op relationship with some tuple in the set)
– op all (op relationship with all tuples in the set)
where op is one of (=, !=, <, >, <=, >=)
Example: find the names of employees who make more than everybody on the
first floor.
select ename
from emp
wheresal > all
(select sal
from emp, dept
where emp.dname=dept.dname and floor = 1)
Scoping of variables works exactly as in Pascal or C
Example: find the names of students who take a course from their advisor.
select sname
from student
wheres# in
(select s#
from enroll
where c# in
(select c#
from class
where prof=student.advisor))
Recap: SQL
Four basic commands
SQL Insert
• Insert command format:
insert into relation_name values (value_list)
insert into relation_name select_statement
• Semantics of insert
– format one: add the tuple corresponding to value_list into
– format two: execute the select statement, then add all the resulting
tuples into relation_name
insert into student values (1, “Carey”, “CS”, “Stonebraker”)
SQL Insert
Example: relation
register(S#, name, paid)
in which registered students are recorded. After the end of
registration week, we execute:
insert into student
select r.s#,
from register r
where r.paid=“yes”
SQL Delete
• Delete command format:
delete relation_name where qualification
• Semantics of delete: execute the corresponding select
(or “*”)
and then remove the resulting tuples from relation_name
SQL Delete
Example: with the following schema
student(s#, name, major, advisor)
enroll(s#, c#, grade)
course(c#, dept)
The following command expels CS majors who received a grade of
less than 2.5 in a CS course:
major=“CS” and s# in
(select s#
from enroll, course
where enroll.s#=student.s# and grade<2.5
and enroll.c#=course.c# and dept=“CS”)
SQL Update
• Update format
• Semantics of update: it is equivalent to executing:
– insert into del_temp
select *
from relation_name
where qualification
SQL Update
• Semantics of update (cont): … then executing
– insert into app_temp
select ext_target_list
from relation_name
where qualification
Ext_target_list is identical to
target_list in the original update
command, but augmented with
for all attributes of the range of
tuple_variable that don’t appear
in target_list.
– delete the tuples in del_temp from relation_name
– add the tuples in app_temp to relation_name
SQL Update
Example: give a 10% grade raise to every CS major in CS564
c#=“CS564” and s# in
(select s#
from student
where major=“CS”)
SQL Update
Which is equivalent to:
insert into del_temp
select s#, c#, grade
from enroll
where c#=“CS564” and s# in
(select s#
from student
where major=“CS”)
insert into app_temp
select s#, c#, grade=1.1*grade
from enroll
where c#=“CS564” and s# in
(select s#
from student
where major=“CS”)
SQL Aggregates
Aggregate functions are functions that take a collection of
values as input and return a single value. SQL supports
five built-in aggregate functions:
– average: avg
– minimum: min
– maximum: max
– total: sum
– cardinality: count
using distinct to aggregate only unique values is often important with
avg, sum, and count
SQL Aggregates
Example: find the number of students
select num_of_students = count(s#)
from student
why do we not need to use distinct in this example?
Example: find the number of employee records
select count (*)
from emp
if an employee appears more than once in the emp relation, for example
if he had switched jobs or had two jobs, then this command would
count that employee once for each record
SQL Aggregates
• Qualified Aggregates:
Example: find the average age of employees in the toy department
select avg(age)
from emp
where dname=“toy”
SQL: Group By clause
• Group aggregates: groups of tuples are computed using the
group by clause
– the attributes given in the clause are used to form groups
– typles with the same value on all attributes in the clause are placed
in one group
Example: in each department, find the minimum age of employees who make
more than 50K
group by
dname, min(age)
SQL: Having clause
• Sometimes it is useful to state a condition that applies to
groups in group by rather than to tuples. We do that in
SQL with the having clause. SQL applies predicates of
having after groups have been formed.
Example: find the average salary for employees under 30 for each
department having more than 10 such employees
group by
dname, avg(sal)
SQL: Multiple Group Bys
Example: using relation emp(ss#, ename, dept, cat, sal)
Count the employees and average monthly salary for each
employee category in each department
group by
dept, cat, count(*), avg(sal)/12
dept, cat
SQL: Multiple Group Bys
Select … from emp
group by cat
Select … from emp
group by dept
SQL: Multiple Group By
Select … from emp
group by dname, cat
note that some dname/cat groups are empty.
SQL: Examples on Having
Find the average salary of employees under 30 for each
department with more than 10 such employees
group by
dname, avg(sal)
(employee age under 30)
(group by department)
(group size > 10)
SQL: Examples on Having
Find the average salary of employees under 30 for each
department with more than 10 employees
group by
e.dname, avg(e.sal)
emp e
emp ee
(employee age under 30)
(group by department)
(number of employees in group)
(… from the same dept as e)
(why is this query different from the previous one?)
SQL: Examples on Having
Find categories of employees whose average salary exceeds
that of programmers
group by
cat, avg(sal)
avg(sal)> (select avg(sal)
from emp
where cat=“programmer”)
SQL: Examples on Having
Find all departments with at least two clerks
group by
count(*) >= 2
SQL: Examples
Find the names of sailors with the highest rating
rating = (select max(rating)
SQL: Examples
For each boat, find the number of sailors of rating >7 who
have reserved this boat
group by
bid, bname, count(s.sid)
sailors s, boats b, reserve r
s.sid=r.sid and and rating>7
SQL: Examples
For each red boat, find the number of sailors who have
reserved this boat
group by
bid, bname, count(s.sid)
sailors s, boats b, reserve r
s.sid=r.sid and
SQL: Examples
Difference between the last two queries?
– First one gave a qualification on the tuples
(take all tuples of the multijoin
discard tuples that do not fulfill ratings>7
then group them by boat id
then find the cardinality of each group)
– Second one gave a qualification for the groups
(take all tuples of the multijoin
group them by boat id
discard groups representing boats that are non-red
find the cardinality of remaining groups)
And Now, For Something
Completely Different...
• The recent SQL material largely covers chapter 4, at least
sections 4.1 through 4.6 and some of 4.9.
• Earlier we examined Relational Algebra, covering sections
3.1 through 3.3
• Now we leave chapter 4 and head back to examine sections
3.6 and 3.7, covering Relational Calculi
– based upon predicate calculus
– non-procedural query languages (descriptive rather than
– we will examine two relational calculi: tuple calculus and domain
Tuple Calculus
Query: {t | P(t)}
P: is a predicate associated with some relation R
t: is a tuple variable ranging over the relation R
t[A]: is the value of attribute A in tuple t
– students in CMSC 424
• {t | t  enroll  t[course#] = CMSC424}
– students in CMSC 424 conforming with the CMSC-420
• {t | t  enroll   s  enroll  t[course#] = CMSC424 
s[course#] = CMSC420  t[ss#] = s[ss#]}
Tuple Calculus
• Quantifiers and free variables
– ,  quantify the variables following them, binding them to some
value. (in the previous slide, s was bound by )
– A tuple variable that is not quantified by  or  is called a free
variable. (in the previous slide, t was a free variable)
• Atoms
– R(t)
– t[x]  s[y]
where t is a tuple variable
where t,s are tuple variables and
  {, , , , , }
Tuple Calculus
• Formulas
– an Atom is a Formula
– If P and Q are Formulas, then so are (P), P, PQ, PQ, and PQ
– If P(t) is a Formula, then so are t P(t) and t P(t)
• Equivalences
(P  Q)  P   Q
(P  Q)  P   Q
t P(t)   (t ( P(t)))
 t P(t)   ( t ( P(t)))
Tuple Calculus
• Safety
– Math is too powerful; we can easily phrase expressions that
describe infinite sets
{t | t  enroll}
– These expressions are called unsafe
– When we are dealing with finite sets, unsafe expressions happen in
expressions that involve negation ()
– We can avoid this problem by using an entirely positive (nonnegated) scope as the first operand of any conjunction where we
use negation. The first operand establishes the scope and the
second one filters the established scope.
{t | t  enroll  t[course#]  CMSC-420}
Domain Calculus
• Another form of relational calculus
• Uses domain variables that take values from an attribute’s
domain, rather than values representing an entire tuple
• Closely related to tuple calculus
• Domain Calculus serves as the theoretical basis for the
query language QBE, just as the relational algebra we
examined earlier forms the basis for SQL
• Expressions are of the form:
{< x1, x2, x3, ..., xn> | P( x1, x2, x3, ..., xn) }
Domain Calculus
• Atoms
– < x1, x2, x3, ..., xn>  R
– x y
where x,y are domain variables and
  {, , , , , }
– x c
where c is a constant
• Formulas
– an atom is a formula
– If P and Q are formulas, then so are (P), P, PQ, PQ, and PQ
– If P(x) is a formula and x is a domain variable, then x P(x) and
x P(x) are also formulas
Domain Calculus
• Queries are of the form:
{< x1, x2, x3, ..., xn> | P( x1, x2, x3, ..., xn) }
• Examples
{<ss#, course#, semester> | Enroll(ss#, course#, semester)}
{<x, y, z> | Enroll(x, y, z)  y = CMSC-424}
Reductions of Relational Algebra and Calculi
• Relational Algebra (sections 3.2-3.5), Tuple Calculus
(section 3.6), and Domain Calculus (section 3.7) can be
reduced to each other: they have equivalent expressive
power. For every expression in one, we can compute an
equivalent expression in the others.
Functional Dependencies
• Important concept in differentiating good database designs
from bad ones
• FD is a generalization of the notion of keys
• An FD is a set of attributes whose values uniquely
determine the values of the remaining attributes.
Emp(eno, ename, sal)
key FDs:
eno => ename
Dept(dno, dname, floor)
eno => sal
Works-in(eno, dno, hours)
(eno,dno) => hours
dno => dname
dno => floor
Functional Dependencies
• If   R and   R, then  =>  holds in the extension
r(R) of R iff for any pair t1 and t2 tuples of r(R) such that
t1()=t2() , then it is also true that t1() = t2()
• We can use FDs as
– constraints we wish to enforce (e.g., keys)
– for checking to see if the FDs are satisfied within the database
A => B satisfied?
A => C satisfied?
C => A satisfied?
AB => D satisfied?
Functional Dependencies
• Trivial dependencies:
 => 
 =>  if   
• Closure
– we need to consider all FDs
– some are implied by others; e.g., FDs are transitive; if A=>B and
B=>C, then A=>C
– Given F = set of FDs, we want to find F’ (the closure of all FDs
logically implied by F)
Armstrong’s Axioms
• Reflexivity
• Augmentation
• Transitivity
if    then  => 
if  =>  then   
if  =>  and  =>  then  => 
Armstrong’s Axioms can be used to derive three additional useful rules:
• Union rule
if  =>  and  =>  then  => 
• Decomposition rule
if  =>  then  =>  and  => 
• Pseudotransitivity rule if  =>  and  =>  then  => 
FD Example
R(A, B, C, G, H, I)
F=( A=>B
CG => H
CG => I
B => H )
F+=( A => H
CG => HI
AG => I
AG => H )
transitivity: A => B => H
union rule: CG => H, CG => I
augmentation: A=> C, AG => CG => I
augmentation: A=> C, AG => CG => H
Closure of Attribute Sets
• Useful to test if an attribute set is a superkey
• the closure + of a set of attributes  under F is the set of all attributes
that are functionally determined by 
• there is an algorithm to compute the closure
R(A, B, C, G, H, I)
F=(A=>B, A=>C, CG => H, CG => I, B => H )
Example: Algorithm to compute (AG)+
starts with
A=>B expands
A=>C expands
CG=>H expands
CG=>I expands
B=>H causes no more expansion
Uses of Attribute Closure
• Testing for superkey
– to test if  is a superkey, compute + and determine if + contains all
attributes of R
• Testing functional dependencies
– to check if a functional dependency  =>  holds, (is in F+) just check to
see if   +
– in other words, we compute + using attribute closure, and then check if
the result contains .
– Simple, cheap, and useful test.
Relational Database Design
• A major goal in designing a database is to have a schema
– makes queries simpler (easy to phrase)
– avoids redundancies and update anomalies (about which more
Schema and Query Simplicity (1)
Example Schema 1:
EMP(eno, ename, sal, dno)
DEPT(dno, dname, floor, mgr)
Query 1: find all employees that make more than their manager
select e.ename from EMP e, EMP m, DEPT d
where e.dno = m.dno and d.mgr=m.eno and e.sal>m.sal
Query 2: for each department, find the maximum salary
select d.dname, max(e.sal) from EMP e, DEPT d
where e.dno = d.dno group by d.dno
Q1 requires two joins; Q2 requires a join and a group-by.
Schema and Query Simplicity (2)
Example Schema 2: (a single relation)
ED(eno, ename, sal, dno, dname, floor, mgr)
Query 1: find all employees that make more than their manager
select e.ename from ED e, ED m
where e.mgr=m.eno and e.sal>m.sal
Query 2: for each department, find the maximum salary
select d.dname, max(sal) from ED e
group by dno
Q1 requires one join; Q2 requires just a group-by.
Schema and Query Simplicity (3)
• How did we get simpler queries?
• Schema 2 was a more complicated relation with more information; in
essence ED was EMP and DEPT from Schema 1 with the join precomputed
• Should we just precompute the joins and store bigger relations?
• Taken to the extreme, we could compute the universal relation with all
attributes inside it and null values for those values that make no sense
• Why wouldn’t we want to do that?
• Problems with too-complex relations: repetition of information (data
redundancy) and inability to represent certain information (update
DB Design: Redundancy and Anomalies
• Redundancy (repetition of information)
– each department is repeated for each employee in it
– great risk of inconsistencies -- suppose the department is moved to
a new floor?
– A simple update (change in mgr name, department floor, etc) in
Schema 1 becomes multiple updates in Schema 2
• Anomalies (inability to represent some types of
– departments can’t exist without employees. A department cannot
exist until the first employee is inserted, and it can no longer exist
when the last employee is deleted from the ED relation
DB Design: Dealing with Anomalies
• So complex relations make for simpler queries, but have
the disadvantages of data redundancy and creation of
anomalies. How do we balance the two objectives? We
– simple queries
– no anomalies; minimize data redundancy
• If we start with Schema 2 and discover anomalies we can
decompose the relation(s) until the problems go away.
This process is called normalization.
Objectives of DB Design
• no redundancy
– for space efficiency and to reduce the potential for inconsistencies
• update integrity
– avoid update anomalies
• linguistic efficiency
– simpler queries are much better for the application programmer and for
the query optimizer
• good performance
– smaller relations imply more joins (bad)
Lossy Decompositions
• Not all decompositions are reversible (lossless)
Shipment(S#, P#, J#) decomposed into SP(S#, P#) and
SJ(S#, J#)
Lossy Decompositions
Shipment(S#, P#, J#) decomposed into SP(S#, P#) and
SJ(P#, J#)
If we join SP and SJ again into SP-PJ(S#, P#, P#, J#) we get:
from the joined tuples we cannot
deduce the original form of the data.
this is called the connection trap
and the decomposition is lossy
Example of Lossy Join Decomposition
• Lossy-join decompositions result in information loss
• Example: decomposition of R=(A,B) into R1=(A) and R2=(B)
R= (A, B)
 1
 2
 1
R1 X R2 = (A,
Decomposition Continued
• Decompose the relation schema
• All attributes of an original schema (R) must appear in the
decomposition (R1, R2)
• Lossless (reversible) join decomposition: for all possible relations r on
schema R, the decomposition into (R1, R2) is lossless if
r = R1(r)
R2 (r)
• The decomposition of R into R1 and R2 is lossless if and only if at least
one of the following dependencies is in F+:
R1  R2  R1
R1  R2  R2
Lossless Join Decomposition and
Functional Dependencies
• So FDs can help determine whether a decomposition is lossless
• R is a relation schema and F its FDs. Then a decomposition
R = R1  R2
is lossless if at least one of the following dependencies holds
R1  R2  R1
R1  R2  R2
• either of the above FDs guarantees uniqueness in the mapping (and
therefore that the decomposition is lossless)
Dependency Preservation
• Dependencies are preserved in a decomposition if we do
not need to join in order to enforce FDs -- all FDs remain
intra-relational and do not become inter-relational
• To check if a decomposition is dependency preserving, we
need to examine all FDs in F+
• There is an algorithm for testing dependency preservation
(requires the computation of F+)
Goals of Normalization
• Decide whether a particular relation R is in “good” form
• if it is not in “good” form, decompose it into a set of
relations (R1, R2, R3, …, Rn) such that:
– each relation is in “good” form
– the decomposition is a lossless-join decomposition, based upon
functional dependencies
• Types of FDs in R(A, B, C, D) with (A, B) a candidate
– trivial:
– partial:
AB ==> A
A ==> C
(C depends upon a part of the key)
TEACH(student, teacher, subject)
student, subject ==> teacher (students not allowed in the same subject
of two different teachers)
teacher ==> subject
– transitive: A ==> C ==> D
(each teacher teaches only one subject)
ED(eno, ename, sal, dno, dname, floor, mgr)
eno ==> dno ==> mgr
Normalization using FDs
• When we decompose a relation schema R with a set of
functional dependencies F into R1, R2, R3, …, Rn we want:
– lossless-join decomposition: otherwise the decomposition results in
loss of information relative to the original schema R
– no redundancy: the relations Ri should be in either BCNF (BoysCodd Normal Form) or 3NF (Third Normal Form) (about which
more in a slide or two)
– Dependency preservation: let Fi be the set of dependencies in F+
that include only attributes in Ri:
• preferably the decomposition should be dependency perserving. That is,
F1  F2  F3  …  Fn = F+
• Otherwise checking updates for violation of FDs may require
computing joins, which is expensive
The Normal Forms
• 1NF: every attribute has an atomic value
• 2NF: 1NF and no partial dependencies
• 3NF: 2NF and no transitive dependencies.
Equivalently (text definition): if for each FD X==> Y either
• it is trivial, or
• X is a superkey, or
• Y-X is a proper subset of a candidate key (each attribute in Y that isn’t
in X is contained in some candidate key)
• BCNF: if for each FD X==> Y either
• it is trivial, or
• X is a superkey
Distinguishing Examples
• 1NF but not 2NF:
SUPPLY(sno, pno, jno, scity, jcity, qty)
– (sno, pno, jno) is the candidate key
– sno ==> scity, jno ==> jcity are both partial dependencies
• 2NF but not 3NF:
ED( eno, ename, sal, dno, dname, floor, mgr)
– transitive FD: eno ==> dno ==> dname
• 3NF but not BCNF:
TEACH(student, teacher, subject)
– student, subject ==> teacher
– teacher ==> subject
Boyce-Codd Normal Form
BCNF is perhaps the most useful Normal Form for database
A relation schema R is in BCNF with respect to a set F of
functional dependancies if for all functional dependancies
in F+ of the form X==> Y where XR, YR; at least one
of the following holds:
– X ==>Y is trivial (that is, Y  X)
– X is a superkey for R
BCNF Example
• R = (A, B, C)
• F = (A==> B,
B==> C)
• R is not in BCNF
• Decomposition R1 = (A, B), R2 = (B, C)
– R1 and R2 are in BCNF
– Lossless-join decomposition
– Dependency preserving
Third Normal Form: Motivation
• There are some situations where
– BCNF is not dependency preserving, and
– efficient checking for FD violation on updates is important
– In these cases BCNF is too severe and a looser Normal Form
would be useful
• Solution: define a weaker Normal Form, called Third
Normal Form, where
– FDs can be checked on individual relations without performing a
join (no inter-relational FDs)
– There is always a lossless-join, dependency-preserving
Third Normal Form
• A relation schema R is in 3NF with respect to a set F of
functional dependancies if for all functional dependancies
in F+ of the form X==> Y where XR, YR; at least one
of the following holds:
– X ==>Y is trivial (that is, Y  X)
– X is a superkey for R
– Each attribute A in X==>Y is contained in a candidate key for R
(note: possibly in different candidate keys)
• A relation in BCNF is also in 3NF
• 3NF is a minimal relaxation of BCNF to ensure
dependency preservation
3NF Example
• R = (J, K, L)
• F = (JK==> L,
L==> K)
• Two candidate keys: JK and JL
• R is in 3NF
– JK==>L
– L==>K
JK is a superkey
K is contained in a candidate key
• BCNF decomposition has R1 = (J, L), R2 = (J, K)
– testing for JK==>L requires a join
• There is some redundancy in this schema
Testing for 3NF
• Optimization: need to check only FDs in F, need not check
all FDs in F+
• Use attribute closure to check, for each dependency
X==>Y, if X is a superkey
• If X is not a superkey, we have to verify if each attribute in
Y is contained in a candidate key of R
– This test is rather more expensive, since it involves finding
candidate keys
– Testing for 3NF has been shown to be NP-hard
– Interestingly, decomposition into 3NF can be done in polynomial
time (testing for 3NF is harder than decomposing into 3NF!)
Comparison of BCNF and 3NF
• It is always possible to decompose a relation into relations
in 3NF such that:
– the decomposition is lossless
– the dependencies are preserved
• It is always possible to decompose a relation into relations
in BCNF such that:
– the decomposition is lossless
– but it may not be possible to preserve dependencies
BCNF and 3NF Comparison (cont.)
Example of problems due to redundancy in 3NF
– R = (J, K, L)
F = (JK==> L, L==> K)
A schema that is in 3NF but not BCNF has the problems of:
– repetition of information (e.g., the relationship between l1 and k1)
– need to use null values (e.g., to represent the relationship between
l2 and k2 when there is no corresponding value for attribute J)
Design Goals
• Goal for a relational database design is:
– Lossless Join
– Dependency Preservation
• If we cannot achieve this, we accept one of
– lack of dependency preservation (or use of more expensive inter-relational
methods to preserve dependencies)
– data redundancy due to use of 3NF
• Interestingly, SQL does not provide a direct way of specifying
functional dependencies other than superkeys
– can specify FDs using assertions, but they are expensive to test
– Even if we have a dependency preserving decomposition, using SQL we
cannot efficiently test an FD whose left hand side is not a key
BCNF and Over-normalization
• Goal is to obtain schemas that are:
– Lossless Join
– Dependency Preserving
• but sometimes we have to look at the meaning, too
TEACH(student, teacher, subject)
student, subject ==> teacher
(students not allowed in the
same subject of two teachers)
teacher ==> subject
(each teacher teaches one subject)
• This 3NF has anomalies:
– Insertion: cannot insert a teacher until we have a student taking his subject
– Deletion: if we delete the last student of a teacher, we lose the subject he
BCNF and Over-normalization (2)
• What is the problem? Schema overload. We are trying to capture two
– 1) subject X can be taught by teacher Y
– 2) student Z takes subject W from teacher V
It makes no sense to say we lose the subject he teaches when he does
not have a student. Who is he teaching the subject to?
• Normalizing this schema to BCNF cannot preserve dependencies, so
we better stay with the 3NF TEACH and another (BCNF) relation
SUBJECT-TAUGHT (teacher, subject) to capture the meaning of the
real-world environment more effectively.
Getting Physical: Storage and File
Structure (Chapter 11)
• Up until now we have examined database design from a
high-level conceptual view, passing over actual
implementation and underlying hardware.
– Appropriate focus for database users
– But hardware does have an influence on implementation, and
implementation does have an influence on what conceptual designs
will be more efficient and useful
• Now we get physical -- examine physical storage media to
give a background for later focus on implementation of the
data models and languages already described
Chapter 11
At this point we are focussing on the following sections
• 11.1 Overview of Physical Storage Media
• 11.2 Magnetic Disks
• 11.3 RAID (very briefly)
• 11.4 Tertiary Storage
• 11.5 Storage Access
• 11.6 File Organization
• 11.7 Organization of Records in Files
• 11.8 Data-Dictionary Storage
Classification of Physical Storage Media
• Media are classified according to three characteristics:
– speed of access
– cost per unit of data
– reliability
• data loss on power failure or system crash
• physical failure of the storage device
• We can also differentiate storage as either
– volatile storage
– non-volative storage
Physical Storage Media Overview (11.1)
• Typical media available are:
Main memory
Flash memory
Mag disk
Optical storage (CD or DVD)
Tape storage
Physical Storage Media -- Cache and
Main Memory
• Cache
– fastest and most costly form of storage
– volatile
– managed by computer system hardware
• Main memory
– fast access (10s to 100s of nanoseconds)
– generally too small or expensive to hold the entire database
• current capacities commonly used are up to a few Gigabites
• capacities have gone up and per-byte costs have decreased steadily,
roughly a factor of 2 every 2-3 years
– volatile
Physical Storage Media -- Flash Memory
• Also known as EEPROM -- Electrically Erasable Programmable ReadOnly Memory
• non-volatile
• reading data is comparable to main memory speeds
• writing is more complex
– can’t overwrite a single location -- a whole bank of memory must be
erased to permit writing within that bank.
– erasing is only supported a limited number of times -- 10,000 to one
million erase cycles
– writes are slow (a few microseconds), and erases are slower
• cost comparable to main memory
• widely used in computer systems embedded in other devices, such as
digital cameras and hand-held computers
Physical Storage Media -- Magnetic Disk
data is stored on a spinning disk and read/written magnetically
primary medium for long-term storage of data
typically stores entire database
data must be moved from disk to main memory for access, and written back
for storage
– much slower access than main memory (about which more later)
direct access -- possible to read data on disk in any order, unlike magnetic tape
capacities up to 100 gig
– much larger capacity and cheaper cost/byte than main memory or flash memory
– capacity doubles every two or three years
survives power failures and system crashes
– disk failure can destroy data, but this is more rare than system crashes
Physical Storage Media -- Optical
• Non-volatile; data is read optically from a spinning disk using a laser
• CD-ROM (640 MB) and DVD (4.7 to 17 GB) most popular forms
• Write-once, Read-many (WORM) optical disks used for archival
storage (CD-R and DCD-R)
• Multiple-write versions also available (CD-RW, DVD-RW, and DVDRAM)
• Reads and writes are slower than with magnetic disk
• Juke-box systems available for storing large volumes of data
– large numbers of removable disks
– several drives
– mechanism for automatic loading/unloading of disks
Physical Storage Media -- Tape Storage
• Non-volatile
• used primarily for backup (to recover from disk failure) and for
archival data
• sequential access -- much slower than disk
• very high capacity (40-300 GB tapes available)
• tape can be removed from drive; storage costs much cheaper than disk,
but drives are expensive; data is read optically from a spinning disk
using a laser
• Juke-box systems available for storing large volumes of data
– e.g., remote sensing data, possibly hundreds of terabytes (1012 bytes) or
even a petabyte (1015 bytes)
Storage Hierarchy
Primary storage: fastest media but
– cache
– main memory
secondary storage: next level in
hierarchy; moderately fast access
time, non-volatile
– also called on-line storage
– flash memory, magnetic disks
tertiary storage: lowest level in
hierarchy; slower access time, nonvolatile
– also called off-line storage
– optical storage, magnetic tape
Magnetic Disks (11.2)
Read-write head
– positioned very close to the platter
surface (almost touching it)
– Reads or writes magnetically
coded information
Surface of platter divided into
circular tracks
– over 16,000 tracks per platter on
typical hard disks
Each track is divided into sectors
– a sector is the smallest unit of data
that can be read or written
– sector size is typically 512 bytes
– typically 200 (on inner tracks) to
400 (outer tracks) sectors per track
Magnetic Disks (cont)
To read/write a sector
– disk arm swings to position head
on the right track
– platter spins continually; data is
read/written as sector passes under
Head-disk assemblies
– multiple disk platters on a single
spindle (typically 2 to 4)
– one head per platter, mounted on a
common arm
Cylinder i consists of ith track of all
the platters
Magnetic Disks (cont)
• Earlier generation disks were susceptible to head crashes
– disk spins constantly at 60, 120, even 250 revolutions per second
– head is very close to the surface; if it touches the surface it can
scrape the recording medium off the surface, wiping out data and
causing the removed medium to fly around, causing more head
– newer disks have less friable material; less subject to head crashes
Magnetic Disks (cont)
Disk controller -- interfaces
between the computer system and
the disk drive
– accepts high-level commands to
read or write a sector
– initiates actions such as moving
the disk arm to the right track and
actually reading or writing the data
– computes and attaches checksums
to each sector to verify that data is
read back correctly
– Ensures successful writing by
reading back sector after writing it
– Performs remapping of bad sectors
Multiple disks are connected to a
computer system through a
– controllers functionality
(checksum, bad sector remapping)
often carried out by individual
disks, reducing load on controller
Two disk interface standards are
ATA (AT attachment) and SCSI
(Small Computer System
Disk Performance Measures
• Access time -- the time it takes from when a read or write request is
issued to when data transfer begins. Consists of:
– Seek time -- time it takes to reposition the arm over the correct track
• average seek time is 1/2 the worst case seek time
• 4 to 10 milliseconds on typical disks
– Rotational latency -- time it takes for the sector to appear under the head
• average latency is 1/2 the worst case latency
• 4 to 11 milliseconds on typical disk (5400 to 15000 rpm)
• Data Transfer Rate -- rate of retrieval from or storage to disk
– 4 to 8 MB per second is typical
– Multiple disks may share a controller, so the rate that controller can handle
is also important. E.G., ATA-5: 66MB/sec, SCSI-3: 40MB/sec, Fiber
Channel: 256 MB/s
Disk Performance Measures (cont.)
• Mean time to failure (MTTF) - the average time the disk is expected to
run continuously without any failure.
– Typically 3-5 years
– Sounds good, but if you have 1500 disks then 300 per year will fail, or
about 1 per day
– MTTF decreases as the disk ages
• RAID (Redundant Arrays of Independent Disks) (11.3)
– disk organization techniques that manage a large number of disks,
providing a view of a single disk of
• high capacity and high speed by using multiple disks in parallel
• high reliability by storing data redundantly, so that data can be recovered even
if a disk fails
• MTTdata loss can be as high as 500,000 to 1,000,000 hours on a RAID
Optimization of Disk-Block Access:
• Requests for disk I/O are generated both by the file system and by the
virtual memory manager
• Each request specifies the address on the disk to be referenced in the
form of a block number
– a block is a contiguous sequence of sectors from a single track on one
– block sizes range from 512 bytes to several K (4 -- 16K is typical)
– smaller blocks mean more transfers from disk; larger blocks makes for
more wasted space due to partially filled blocks
– block is the standard unit of data transfer between disk to main memory
• Since disk access speed is much slower than main memory access,
methods for optimizing disk-block access are important
Optimization of Disk-Block Access:
• Disk-arm Scheduling: requests for several blocks may be
speeded up by requesting them in the order they will pass
under the head.
– If the blocks are on different cylinders, it is advantageous to ask for
them in an order that minimizes disk-arm movement
– Elevator algorithm -- move the disk arm in one direction until all
requests from that direction are satisfied, then reverse and repeat
– Sequential access is 1-2 orders of magnitude faster; random access
is about 2 orders of magnitude slower
Optimization of Disk-Block Access:
• Non-volatile write buffers
– store written data in a RAM buffer rather than on disk
– write the buffer whenever it becomes full or when no other disk
requests are pending
– buffer must be non-volatile to protect from power failure
• called non-volatile random-access memory (NV-RAM)
• typically implemented with battery-backed-up RAM
– dramatic speedup on writes; with a reasonable-sized buffer write
latency essentially disappears
– why can’t we do the same for reads? (hints: ESP, clustering)
Optimization of Disk-Block Access:
• File organization (Clustering): reduce access time by
organizing blocks on disk in a way that corresponds
closely to the way we expect them to be accessed
– sequential files should be kept organized sequentially
– hierarchical files should be organized with mothers next to
– for joining tables (relations) put the joining tuples next to each
– over time fragmentation can become an issue
• restoration of disk structure (copy and rewrite, reordered) controls
Optimization of Disk-Block Access:
• Log-based file system
– does not update in-place, rather writes updates to a log disk
• essentially, a disk functioning as a non-volatile RAM write buffer
– all access in the log disk is sequential, eliminating seek time
– eventually updates must be propogated to the original blocks
• as with NV-RAM write buffers, this can occur at a time when no disk
requests are pending
• the updates can be ordered to minimize arm movement
– this can generate a high degree of fragmentation on files that
require constant updates
• fragmentation increases seek time for sequential reading of files
Storage Access (11.5)
• Basic concepts (some already familiar):
– block-based. A block is a contiguous sequence of sectors from a single
track; blocks are units of both storage allocation and data transfer
– a file is a sequence of records stored in fixed-size blocks (pages) on the
– each block (page) has a unique address called BID
– optimization is done by reducing I/O, seek time, etc.
– database systems seek to minimize the number of block transfers between
the disk and memory. We can reduce the number of disk accesses by
keeping as many blocks as possible in main memory.
– Buffer - portion of main memory used to store copies of disk blocks
– buffer manager - subsystem responsible for allocating buffer space in
main memory and handling block transfer between buffer and disk
Buffer Management
• The buffer pool is the part of the main memory alocated for
temporarily storing disk blocks read from disk and made available to
the CPU
• The buffer manager is the subsystem responsible for the allocation and
the management of the buffer space (transparent to users)
On a process (user) request for a block (page) the buffer manager:
checks to see if the page is already in the buffer pool
if so, passes the address to the process
if not, it loads the page from disk and then passes the address to the process
loading a page might require clearing (writing out) a page to make space
Very similar to the way virtual memory managers work, although it can do a
lot better (why?)
Buffer Replacement Strategies
• Most operating systems use a LRU replacement scheme. In database
environments, MRU is better for some common operations (e.g., join)
– LRU strategy: replace the least recently used block
– MRU strategy: replace the most recently used block
• Sometimes it is useful to fasten or pin blocks to keep them available
during an operation and not let the replacement strategy touch them
– pinned block is thus a block that is not allowed to be written back to disk
• There are situations where it is necessary to write back a block to disk
even though the buffer space it occupies is not yet needed. This write
is called the forced output of a block; useful in recovery situations
• Toss-immediate strategy: free the space occupied by a block as soon as
the final tuple of that block has been processed
Buffer Replacement Strategies
• Most recently used (MRU) strategy: system must pin the block
currently being processed. After the final tuple of that block has been
processed the block is unpinned and becomes the most recently used
block. This is essentially “toss-immediate” with pinning, and works
very well with joins.
• The buffer manager can often use other information (design or
statistical) to predict the probability that a request will reference a
particular page
– e.g., the data dictionary is frequently accessed -- keep the data dictionary
blocks in main memory buffer
– if several pages are available for overwrite; choose the one that has the
lowest number of recent access requests to replace
Buffer Management (cont)
• Existing OS affect DBMS operations by:
read ahead, write behind
wrong replacement strategies
Unix is not good for DBMS to run on top
Most commercial systems implement their own I/O on a raw disk
• Variations of buffer allocation
common buffer pool for all relations
separate buffer pool for each relation
as above but with relations borrowing space from each other
prioritized buffers for very frequently accessed blocks, e.g. data
Buffer Management (cont)
• For each buffer the manager keeps the following:
– which disk and which block it is in
– whether the block is dirty (has been modified) or not (why?)
– information for the replacement strategy
• last time block was accessed
• whether it is pinned
• possible statistical information (access frequency etc.)
Buffer Management and Disk-block
Access Optimization (end)
• Disk-block access methods must take care of some
information within each block, as well as information
about each block:
– allocate records (tuples) within blocks
– support record addressing by address and by value
– support auxiliary (secondary indexing) file structures for more
efficient processing
• These concerns are linked in to our next topic:
file organization.
File Organization
The database is stored logically as a collection of files.
Each file is a sequence of records
A record is a sequence of fields
Easy so far, but as we just finished discussing, anything
stored on disk is stored in blocks, which are a physical
constraint unrelated to the storage system used for files.
• So how do we organize a file into blocks and records?
– formatting fields within a record
– formatting records within a block
– assigning records to blocks.
Fixed-Length Records
• Simplest approach. We know
the length of each record, and
they are all the same
– store record i starting from
byte n * (i - 1), where n is
record length
– record access is simple
Fixed-Length Records
• Problems:
– records may cross blocks
• normal modification: don’t
permit records to cross block
– deletion of record i leaves a
gap, which requires some way
of dealing with the empty
– E.G., record 2 (A-215) is
deleted from the example
block on the right
Fixed-Length Records: Deletion
• One simple fix is to shift all the
records down to fill the gap, as
shown on the right.
• This involves a lot of work, so
it might be slow
Fixed-Length Records: more Deletion
• Another fix would be to shift
the last record (record n) to the
deleted position I
• Much less work
• Not useful if the records are
stored in order (sorted)
Fixed-Length Records: still more
• Another possibility is to not
move records at all
• Maintain a header at the
beginning of the file
• Store a link to the list of
addresses of deleted records
• Use each deleted record to store
the link to the next deleted
• Essentially a linked list, often
called a free list
Variable Length Records
• Can occur in a database system in several ways
– storage of multiple record types in a file
– record types that allow variable lengths for one or more fields
– record types that allow repeating fields
• Several ways to store variable length records
– attach an “end of record” symbol (delimiter) to mark the end of
each record
– store the length of the record at the beginning of each record
– store header information at the beginning of the file with the
location and length of each record
– these techniques can be applied at the file, block, or record level
Variable Length Records
– delimit each record within the file
– store a length field at the beginning of each record
– store header information at the beginning of the file with the location and length of
each record within the file
– delimit each record within the block
– store a length field at the beginning of each record
– store header information at the beginning of the block with the location and length
of each record inside the block
– delimit each field within the record
– store a length field at the beginning of each field
– store header information at the beginning of the record with the location and length
of each field
Variable Length Records
Two more techniques for storing variable-length records
– use fixed-length fields
• reserve space -- if there is a maximum space, just reserve that, and
mark unused space with a special null (end-of-record) symbol
• wasteful if the maximum record length is much larger than the
average record length
– list representation
• represent variable-length records by lists of fixed-length records,
chained together by pointers
• useful for variable-length records caused by repeating same-length
• we don’t want a single field of the variable-length record to cross the
boundary of two fixed-length records in its representation, so this can
also be wasteful of space
Organizing Records in a Block
• Two major ways we can organize records within a block
(disk page)
– fixed-packed (contiguous storage)
– slotted page structure (indexed heap)
1) fixed-packed -- records are stored contiguously
highly inflexible
records may span over block boundary
fragmentation with deletions and insertions
external pointers may prevent internal block reorganization -records are pinned to their address in the block
Organizing Records in a Block
2) slotted page structure
– an initial block header storing block address and offset is used to
reference the record
– records are indexed within a block
– insertions and deletions are easy (there are no assumptions about
contiguity of records and record-address startpoints to deal with)
– records may be rearranged within the block without concerns about
external pointers
– records are not pinned within the block
Organizing Records in a Files
• Given a set of records, how do we organize them in a file?
Three possible methods are:
– 1. Heap -- no order at all. A record can be placed anywhere in the
file where there is space
– 2. Sequential -- records are stored in a sorted order according to the
value of a search key
– 3. Hashing -- a hash function computed on some attribute of each
record is used to specify in which block of the file the record
should be placed
– Records of each relation are often stored in separate files.
Sometimes it is useful to use a clustering file organization, where
records of several relations might be stored in a single file.
Heap File Organization
• Heap -- no order at all. A record can be placed anywhere
in the file where there is space
– easy insert, easy delete
– lack of any structure makes queries (including finding a particular
record) very difficult
– not usually useful for anything except very small relations
Sequential File Organization
• Sequential -- records are stored
in a sorted order according to
the value of a search key
– designed for efficient queries
in sorted order
– very suitable for applications
that require sequential
processing of the entire file
– difficult to maintain sorted
order with insert/delete
– deletions can use a free list
(pointer chain) to mark empty
space as previously described
Sequential File Organization
• Insertions use the following
– locate the location to be
– if there is space there, insert
with no more work
– otherwise insert the record in
an overflow block
– in either case the pointer chain
must be updated
• Every so often we need to
reorganize the whole file to
restore sequential order
Clustering File Organization
• Simple file structure stores each
relation in a separate file
– tuples can be represented as
fixed-length records
– easy to implement
– well-suited to small databases
• Large databases often attempt
to store many relations in one
file using a clustering file
• E.G., relations customer and
depositor shown to the right:
Clustering File Organization
• depositor relation stores the
different accounts that a
particular customer has
• customer relation stores the
address information of a given
• Both relations use customername as a key
• Some common queries on the
two relations join them based
on the customer-name attribute
Clustering File Organization
• Storing the two relations
together, sorted on customername, allows the join to be
computed much more quickly
• There is a price to pay -- some
operations are now more
expensive (slower)
• for example, consider
• sequential pass through the
customer relation is now hard
Clustering File Organization
• To allow sequential access
through all tuples of the
customer relation, we chain
together all the tuples of that
relation using pointers
• clustering results in variablesize records
• Careful use of clustering can
produce significant
performance gains
Data Dictionary
The data dictionary (also called system catalog) stores
metadata -- data about data. For example:
• Information about relations
names of relations
names and types of attributes of each relation
names and definitions of views
integrity constraints
• User and account information, including passwords
Data Dictionary
• Statistical and descriptive data
– number of tuples in each relation
• Physical file organization information
– how relation is stored (sequential, hash, clustered, etc.)
– physical location of relation -- operating system file name or disk
addresses of blocks containing records of the relation
• Information about indices
– (about which more after the midterm, when we cover chapter 12)
Data Dictionary
• In effect, the data dictionary is a mini database. The data
within can either be stored as:
– special-purpose data structures and code to access it, or
– a set of relations, using the existing database structures and code to
access it (most common solution)
Data Dictionary
Example of a possible system catalog representation:
Relation-metadata =
Attribute-metadata =
User-metadata =
Index-metadata =
View-metadata =
(relation-name, number-of-attributes,
storage-organization, location)
(attribute-name, relation-name, domain-type,
position, length)
(user-name, encrypted-password, group)
(index-name, relation-name, index-type,
(view-name, definition)
Reading and Review
Upcoming dates:
– Mar 11: HW#2 is due
– Mar 11 and 13: midterm review
– Mar 18: midterm
Text Sections we’ve covered:
Chapter 1: Introduction
Chapter 2: E-R data model
Chapter 3 (except 3.4 & 3.5): Relational model
Chapter 4: SQL
Chapter 6: Integrity and Security
Chapter 7 up to and including 7.7: Relational-Database Design
Chapter 11 (except 11.3 and 11.9): Storage and File Structure
Reading and Review
Chapter 1: Introduction
1.1: Applications and motivation
1.2: Database systems vs. file systems
1.3: Views
1.4: Data models
1.5: Database languages
1.6: Database users and administrators
1.7-1.10: other stuff
1.11: Summary
Reading and Review
Chapter 2: Entity-Relationship Model
2.1: Basic concepts: entities (entity-sets), relationship (set)s, and attributes
2.2: Constraints and mapping
2.3: Keys
2.4: Design issues
2.5: E-R Diagram
2.6: Weak entity sets
2.7: Extended E-R features (specialization, aggregation)
2.8: Design of an E-R schema
2.9: Reduction of an E-R schema to tables
2.10: (ignore)
2.11: Summary
Reading and Review
Chapter 3: Relational Model
3.1: Structure of relational databases
3.2: The relational algebra
3.3: Extended relational-algebra operations
3.4, 3.5: (ignore)
3.6: Tuple relational calculus
3.7: Domain relational calculus
3.8: Summary
Reading and Review
Chapter 4: SQL
4.1: Background
4.2: Basic structure
4.3: Set operations
4.4: Aggregate functions
4.5: Null values
4.6: Nested subqueries
4.7, 4.8: (ignore)
4.9: Modification of the database
4.10: (ignore)
4.11-4.14: (really ignore)
4.15: Summary
Reading and Review
Chapter 6: Integrity and Security
6.1: Domain Constraints
6.2: Referential Integrity
6.3: Assertions
6.4: Triggers
6.5: Security and Authorization
6.6, 6.7: (ignore)
6.8: Summary
Reading and Review
Chapter 7: Relational-Database Design
7.1: First Normal Form
7.2: Pitfalls in Relational-Database Design
7.3: Functional Dependencies
7.4: Decomposition
7.5: Desirable Properties of Decomposition
7.6: BCNF
7.7: Third Normal Form
7.8-7.10: (ignore)
7.11: Summary
Reading and Review
Chapter 11: Storage and File Structure
11.1: Overview of physical storage media
11.2: Magnetic Disks
11.3: RAID (not responsible for this)
11.4: Tertiary Storage
11.5: Storage Access
11.6: File Organization
11.7: Organization of Records in Files
11.8: Data Dictionary Storage
11.9: (ignore)
11.10: Summary

Database as Model - UMD Department of Computer Science