LECTURE 5 - BASIC FILE STRUCTURES
BASED IN PART ON UNIT 43 - DATABASE CONCEPTS I - OF THE 1990 NCGIA CORE
CURRICULUM IN GIS
Initial html-ization by Brian Klinkenberg, University of British Columbia
Compiled with assistance from Gerald White, California State University,
Sacramento
A.
INTRODUCTION
B.
CONCEPTS IN DATABASE SYSTEMS
C.
DATABASE MANAGEMENT SYSTEMS
D.
HIERARCHICAL MODEL
E.
NETWORK MODEL
F.
RELATIONAL MODEL
REFERENCES
EXAM
AND DISCUSSION QUESTIONS
NOTES
LECTURE 5 - BASIC FILE STRUCTURES
A.
INTRODUCTION
-
very early attempts to build GIS began from scratch, using very limited
tools like operating systems and compilers
-
suppose you were building such a GIS, and needed to store attributes of
each county in California
-
each county's data would be stored together in the computer, just like
storing on sheets of paper
-
this is a list structure
-
to find SB county's data it is necessary to test every record (look at
every sheet)
-
all forms of computer storage have an inherent order
-
it's easy to tell the computer to go to the next position or block in storage
-
we can take advantage of this by ordering the counties, e.g. alphabetically
-
this is ordered sequential storage
-
a raster does this by storing pixels in order
-
can go straight to a given row, column because its position is known in
advance
-
this won't work for counties
-
computer storage can be indexed
-
e.g. store the address of the first of the counties beginning with S
-
now we can retrieve the SB record more quickly (unless all counties start
with S)
-
today, concepts like these are handled by a layer of software called database
systems
-
today's GIS designer can work at a higher level of sophistication
-
recently, GIS have been built around existing database management systems
(DBMS)
-
purchase or lease of the DBMS is a major part of the system's software
cost
-
the DBMS handles many functions which would otherwise have to be programmed
into the GIS
-
any DBMS makes assumptions about the data which it handles
-
to make effective use of a DBMS it is necessary to fit those assumptions
-
certain types of DBMS are more suitable for GIS than others because their
assumptions fit spatial data better
Two
ways to use DBMS within a GIS
1. Total DBMS solution
-
all data are accessed through the DBMS, so must fit the assumptions imposed
by the DBMS designer
2. Mixed (hybrid) solution
-
some data (usually attribute tables and relationships) are accessed through
the DBMS because they fit the model well
-
some data (usually locational) are accessed directly because they do not
fit the DBMS model
-
ARC/INFO uses a mixed solution
-
INFO is a standard DBMS
-
ARC is a custom component with a specialized model
GIS
as a database problem
-
some areas of application, notably facilities management:
-
deal with very large volumes of data
-
often have a DBMS solution installed before the GIS is considered
-
the GIS adds geographical access to existing methods of search and query
-
such systems require very fast response to a limited number of queries,
little analysis
-
in these areas it is often said that GIS is a "database problem" rather
than an algorithm, analysis, data input or data display problem
-
e.g. ESRI's SDE (Spatial Database Engine), Oracle's SDO
B.
CONCEPTS IN DATABASE SYSTEMS
Definition
-
a database is a collection of non-redundant data which can be shared by
different application systems
-
stresses the importance of multiple applications, data sharing
-
the spatial database becomes a common resource for an agency
-
non-redundant means each item is stored if possible only once
-
it can be changed without changing other items
-
the database cannot lose integrity
-
implies separation of physical storage from use of the data by an application
program, i.e. program/data independence
-
the user or programmer or application specialist need not know the details
of how the data are stored
-
such details are "transparent to the user"
-
changes can be made to data without affecting other components of the system.
e.g.
-
change format of data items (real to integer, arithmetic operations)
-
change file structure (reorganize data internally or change mode of access)
-
relocate from one device to another, e.g. from optical to magnetic storage,
from tape to disk
-
replace one database product with another
Advantages
of a database approach
-
reduction in data redundancy
-
shared rather than independent databases
-
reduces problem of inconsistencies in stored information, e.g. different
addresses in different departments for the same customer
-
maintenance of data integrity and quality when there are multiple users
-
data are self-documented or self-descriptive
-
information on the meaning or interpretation of the data can be stored
in the database, e.g. names of items, metadata
-
avoidance of inconsistencies
-
data must follow prescribed models, rules, standards
-
reduced cost of software development
-
many fundamental operations taken care of, however DBMS software can be
expensive to install and maintain
-
security restrictions
-
database includes security tools to control access, particularly for writing
Views
of the database
-
the database can present different views of itself to users, programmers
-
these are built and maintained by the database administrator (DBA)
-
the internal data representation (internal view) is normally not seen by
the user or applications programmer
-
the conceptual view or conceptual schema is the primary means by which
the DBA builds and manages the database
-
the DBMS can present multiple views of the conceptual schema to programmers
and users, depending on the application
-
these are called external views or schemas
C.
DATABASE MANAGEMENT SYSTEMS
Components
Data types
-
include:
-
integer (whole numbers only)
-
real (decimal)
-
character (alphabetic and numeric characters)
-
date
-
cyclical
-
bulk
-
more advanced systems may include pictures and images as data types
-
e.g. a database of buildings for the fire department which stores a picture
as well as address, number of floors, etc.
Standard operations
-
e.g. sort, delete, edit, select records
Data definition language (DDL)
-
the language used to describe the contents of the database
-
e.g. attribute names, data types - metadata
Data manipulation and query language
-
the language used to form commands for input, edit, analysis, output, reformatting
etc.
-
some degree of standardization has been achieved with SQL (Standard Query
Language)
Programming tools
-
besides commands and queries, the database should be accessible directly
from application programs through e.g. subroutine calls
File structures
-
the internal structures used to organize the data, often proprietary
Types
of database systems
-
several models for databases:
-
tabular (flat file) - data in a single table
-
hierarchical
-
network
-
relational
-
the hierarchical, network and relational models all try to deal with the
same problem with tabular data:
-
inability to deal with more than one type of object, or with relationships
between objects
-
e.g. database may need to handle information on aircraft, crew, flights
and passengers - four types of records with different attributes, but with
relationships between them (e.g. "is booked on" between passenger and flight)
-
database systems originated in the late 1950s and early 1960s largely by
research and development of IBM Corporation
-
most developments were responses to needs of business, military, government
and educational institutions - complex organizations with complex data
and information needs
-
trend through time has been increasing separation between the user and
the physical representation of the data - increasing transparency
D.
HIERARCHICAL MODEL
-
early 1960s, IBM saw business world organizing data in the form of a hierarchy
-
rather than one record type (flat file), a business has to deal with several
types which are hierarchically related to each other
-
e.g. company has several departments, each with attributes: name of director,
number of staff, address
-
each department requires several parts to make its product, with attributes:
part number, number in stock
-
each part may have several suppliers, with attributes: address, price
-
certain types of geographical data may fit the hierarchical model well
-
e.g. Census data organized by state, within state by city, within city
by census tract
-
the database keeps track of the different record types, their attributes,
and the hierarchical relationships between them
-
the attribute which assigns records to levels in the database structure
is called the key (e.g. is record a department, part or supplier?)
Summary
of features
-
a set of record types
-
e.g. supplier record type, department record type, part record type
-
a set of links connecting all record types in one data structure diagram
(tree)
-
at most one link between two record types, hence links need not be named
-
for every record, there is only one parent record at the next level up
in the tree
-
e.g. every county has exactly one state, every part has exactly one department
-
no connections between occurrences of the same record type
-
cannot go between records at the same level unless they share the same
parent
Advantages
and disadvantages
-
data must possess a tree structure
-
tree structure is natural for geographical data
-
data access is easy via the key attribute, but difficult for other attributes
-
in the business case, easy to find record given its type (department, part
or supplier)
-
in the geographical case, easy to find record given its geographical level
(state, county, city, census tract), but difficult to find it given any
other attribute
-
e.g. find the records with population 5,000 or less
-
tree structure is inflexible
-
cannot define new linkages between records once the tree is established
-
e.g. in the geographical case, new relationships between objects
-
cannot define linkages laterally or diagonally in the tree, only vertically
-
the only geographical relationships which can be coded easily are "is contained
in" or "belongs to"
-
DBMSs based on the hierarchical model (e.g. System 2000) have often been
used to store spatial data, but have not been very successful as bases
for GIS
E.
NETWORK MODEL
-
developed in mid 1960s as part of work of CODASYL (Conference on Data Systems
Languages) which proposed programming language COBOL (1966) and then network
model (1971)
-
other aspects of database systems also proposed at this time include database
administrator, data security, audit trail
-
objective of network model is to separate data structure from physical
storage, eliminate unnecessary duplication of data with associated errors
and costs
-
uses concept of a data definition language, data manipulation language
-
uses concept of m:n linkages or relationships
-
an owner record can have many member records
-
a member record can have several owners
-
hierarchical model allows only 1:n
-
example of a network database
-
a hospital database has three record types:
-
patient: name, date of admission, etc.
-
doctor: name, etc.
-
ward: number of beds, name of staff nurse, etc.
-
need to link patients to doctor, also to ward
-
doctor record can own many patient records
-
patient record can be owned by both doctor and ward records
-
network DBMSs include methods for building and redefining linkages, e.g.
when patient is assigned to ward
Restrictions
-
links between records of the same type are not allowed
-
while a record can be owned by several records of different types, it cannot
be owned by more than one record of the same type (patient can have only
one doctor, only one ward)
Summary
-
the network model has greater flexibility than the hierarchical model for
handling complex spatial relationships
-
it has not had widespread use as a basis for GIS because of the greater
flexibility of the relational model
F.
RELATIONAL MODEL
-
the most popular DBMS model for GIS
-
the INFO in ARC/INFO
-
EMPRESS in System/9
-
several GIS use ORACLE
-
several PC-based GIS use DBase III
-
flexible approach to linkages between records comes closest to modeling
the complexity of spatial relationships between objects
-
proposed by IBM researcher E.F. Codd in 1970
-
more of a concept than a data structure
-
internal architecture varies substantially from one RDBMS to another
Terminology
-
each record has a set of attributes
-
the range of possible values (domain) is defined for each attribute
-
records of each type form a table or relation
-
each row is a record or tuple
-
each column is an attribute
-
note the potential confusion - a "relation" is a table of records, not
a linkage between records
-
the degree of a relation is the number of attributes in the table
-
1 attribute is a unary relation
-
2 attributes is a binary relation
-
n attributes is an n-ary relation
Examples
of relations
unary: COURSES(SUBJECT)
binary: PERSONS(NAME,ADDRESS) OWNER(PERSON NAME,HOUSE ADDRESS)
ternary: HOUSES(ADDRESS,PRICE,SIZE)
Keys
-
a key of a relation is a subset of attributes with the following
properties:
-
unique identification
-
the value of the key is unique for each tuple
-
non-redundancy
-
no attribute in the key can be discarded without destroying the key's uniqueness
-
e.g. phone number is a unique key in a phone directory
-
in the normal phone directory the key attributes are last name, first name,
street address
-
if street address is dropped from this key, the key is no longer unique
(many Smith, John's)
-
a prime attribute of a relation is an attribute which participates
in at least one key
-
all other attributes are non-prime
Normalization
-
concerned with finding the simplest structure for a given set of data
-
deals with dependence between attributes
-
avoids loss of general information when records are inserted or deleted
-
consider the first relation (prime attribute underlined):
-
this is not normalized since PRICE is uniquely determined by STYLE
-
problems of insertion and deletion anomalies arise
-
the relationship between ranch and 50000 is lost when the last of the ranch
records is deleted
-
a new relationship (triplex costing 75000) must be inserted when the first
triplex record occurs
-
consider the second relation:
-
here there are two relations instead of one
-
one to establish style for each builder
-
the other price for each style
-
several formal types of normalization have been defined - this example
illustrates third normal form (3NF), which removes dependence between non-prime
attributes
-
although normalization produces a consistent and logical structure, it
has a cost in increased storage requirements
-
some GIS database administrators avoid full normalization for this reason
-
a relational join is the reverse of this normalization process,
where the two relations HOMES2 and COST are combined to form HOMES1
Advantages
and disadvantages
-
the most flexible of the database models
-
no obvious match of implementation to model - model is the user's view,
not the way the data is organized internally
-
is the basis of an area of formal mathematical theory
-
most RDBMS data manipulation languages require the user to know the contents
of relations, but allow access from one relation to another through common
attributes
Example: Given two relations:
PROPERTY(ADDRESS,VALUE,COUNTY_ID)
COUNTY(COUNTY ID,NAME,TAX_RATE)
-
to answer the query "what are the taxes on property x" the user would:
-
retrieve the property record
-
link the property and county records through the common attribute COUNTY_ID
-
compute the taxes by multiplying VALUE from the property tuple with TAX_RATE
from the linked county tuple
REFERENCES
Standard database texts:
Date, G.J., 1987. An Introduction to Database Systems, Addison-Wesley,
Reading, MA.
Howe, D.R., 1983. Data Analysis for Data Base Design, Arnold,
London.
Kent, W., 1983. "A simple guide to five normal forms in relational database
theory," Communications of the Association for Computing Machinery
26:120.
Tsichritzis, D.C. and F.H. Lochovsky, 1977, Database Management Systems,
Academic Press, New York.
The relational model in GIS:
van Roessel, J.W., 1987. "Design of a spatial data structure using the
relational normal forms," International Journal of Geographical Information
Systems 1:33-50.
EXAM
AND DISCUSSION QUESTIONS
1. Compare the four database models (flat file, hierarchical, network
and relational) as bases for GIS. What particular features of the relational
model account for its popularity?
2. Polygon overlay has been called a spatial analog of a relational
join. Do you agree?
3. Summarize the arguments against organizing spatial databases as flat
files.
4. Why do you think the term "relation" was chosen for a table of attributes
in the relational model?