Implementing the Databese Server session 02

Implementing the Database
Server SW###
Session 2

Database Hardware
• RAM - Random Access Memory
• Caching - Memory is 10,000 times faster than disk!
• Locality of Reference - if is a piece of data is
accessed, it is likely the next piece of date will be.
• Procedure Cache - precompiled procedures.
• Paging - shuffling of data between memory, and
disk.
• Virtual Memory / Swap Files – the anti-cache…

Database Hardware
• Disk Space - How much for SQL functions?
• NOT ALL - utilities.
• Disk Mirroring / RAID – improve throughput
and/or reliability by making multiple disks act as a
single device. Also allows for larger volumes.
• “Data located on segment” – home-made RAID, for
throughput only.
• Should always be done below SQL Server level if
possible.

Database Hardware
• Processor(s).
• SMP - Symmetric Multi Processing
• Dual-core processors.
• The more the merrier?
• Faster vs. more?

Database Hardware
• Network.
• Adapter speed (100MB, 1GB)
• Protocols - Banyan Vines, anyone? TCP/IP is
the standard world-wide.
• Multiple Adapters? Only on multiple networks,
usually.
• Problems – latency, bandwidth.
• Sniffers may help?

Hardware and Performance
• Simple tests by looking at the performance
monitor.
– If the processor is “floored”, a faster processor
will help.
– If lots of PAGING is occurring, then more
memory can be used. (physical vs. virtual reads).
Cache hit ratio may be a clue.
– If reads and writes are queuing, more disk
bandwidth may help.

Hardware and Performance
• Final Idea - sometimes throwing hardware at a
problem is cheaper than engineering a solution.
Other problems cannot be solved with even an
infinite amount of hardware.
• “Big O Notation” - order of growth of
algorithms.

ODBC, MDAC, ADO, etc.)
• “Standard” interfaces between an application
and the database.
• An API?
• Hands On - Settings | Control Panel | ODBC
Administrator.
• Must be set up on each client.
• The server becomes known to the client system
by one or more ‘Driver Names’. Allegedly
allows for moving of data without application
changes (but local changes).

“SQL Links”

• Name directly stolen from Borland software.
• Referring to any API that is proprietary and
designed to interface between a single
programming environment and designed to talk
to a single database type.
• Oracle SQL*NET.

Devices
• Device : A storage area which SQL Server
can read from or write to. Most frequently
this is a file. Databases and transaction logs
occupy one or more devices (files).
• Dump Devices : Devices used for backups.
Also called ‘backup devices’. Can be a disk
file, a tape, a named pipe.
• Used for dumping or restoring databases and
transaction logs.

Database Devices
• Database Device : A disk file used to store databases and/or
transaction logs.
• Relationship from Databases/Transaction Log to files is 1:N.
A database/log may span multiple files, but databases/logs
may not share files.
• Devices are located in the instance pathDATA by default.
Not always what you want. Data on a different virtual drive?
• The “instance path” is the path where this “instance” of 2005
is running. Can be multiple instances per machine, or
developer edition / standard edition, or use of Reporting
Services, Analysis Services, etc.

Database Devices
• If you create your database in SQL, CREATE
DATABASE and ALTER DATABASE allow the
creation / addition / configuration of devices.
• It is far easier and more common to use SQLEW to
manage devices and databases - recommended.

Databases - LAB
• Register the server.
• Create and configure a database for use
during the semester.
• Should NOT be allowed to automatially
expand!

Databases
• ‘An (organized) collection of (the binary
representation of) data (stored on permanent
read-write media)’
• A self-contained structure, that logically contains
a set of tables, views, rules, stored procedures, etc.
• Can be in one more files.
• Will consume the entire file.

Transaction Logs
• Modifications to databases are atomic.
• A single update to add 10% to all salaries will
affect all rows or none. However, logic dictates
that a computer can only calculate one at a time.
• What happens if the database server crashes in
the middle?

Transaction Logs
• SQL Server uses ‘write-ahead’ transaction logging.
• All changes are viewed as ‘transactions’.
• Transactions are written to the log before the database.
• A ‘midpoint’ is used in the log.
• During a transaction, conceptually the DBMS:
– Identifies all the changes that will be made.
– Writes those changes to the log.
– Writes a midpoint in the log.
– Writes changes to the database, noting each change in the log.

Transaction Logs
• When the server is started, conceptually:
– Any transaction that had not reached the midpoint
never happened.
– Any transaction that had passed the midpoint and
started changing tables can be completed.
• “Conceptually” - the algorithm is more complex
due to possible interaction between transactions.

Transaction Logs
• Pre-7.0, transactions logs could share a device with
the database. This is a very bad idea for both
performance and recovery reasons.
• Also, database (MDF) and log (LDF) files are
position sensitive. Copying moves position and
requires them to be re “ATTACHED”.
• Pre-7.0, .DAT files could not be re-attached
(backing up the files was futile).
• Hint: back up the master DB after changing any
other DB - it holds information on all other DBs!

A Common Error
• [Microsoft][ODBC SQL Server Driver][SQL Server]The
log file for database ‘PRODUCTION' is full. Back up
the transaction log for the database to free up some
log space.

• Logs grow forever unless backed up or
truncated. Limit log space growth and trim
them regularly.
• Even when truncated / backed up, they grow
to a high water mark and remain there.

Generic SQL Data Types
• CHAR(n), VARCHAR(n)
• VARCHAR used when lots of variation can
occur (avoids wasting of space, allows indexes
to be more efficient)
• DECIMAL(p,s), NUMERIC(p,s)
• INT, SMALLINT, LONGINT
• FLOAT(n), REAL, DOUBLE
• REAL and DOUBLE are the same in SQL
Server

SQL Server Data Types
• DATETIME, SMALLDATETIME
• NVARCHAR() et al. (for DBCS)
• TIMESTAMP
• BIT
• BINARY(), VARBINARY()
• MONEY, SMALLMONEY
• TEXT, IMAGE
• IMAGE is basically a BLOB (Binary Large
OBject)

Types of SQL Statements

SQL consists of two groups of statement
types:
Data Manipulation (Modification) Language
(DML) - allows a user to look at, add to, modify,
or remove the data in a database. SELECT,
UPDATE, INSERT, and DELETE are DML
statements.
Data Definition Language (DDL) - allows a user
to alter the structure of the database. Supports
the creation, deletion, and modification of
database objects.

Types of DDL Objects

Tables
Views
Indexes
Rules
Stored Procedures
Triggers

Discussion focuses on Tables and Indexes.

CREATE TABLE
Creates a table with no data.
Defines the structure of the table (the
relation schema).
Basic Syntax :
CREATE TABLE <tableName> (
<columnName1> <dataType1>[(<size1>)],
…,
<columnNameN> <dataTypeN>[(<sizeN>)]
)

CREATE TABLE
The tableName:
May not have spaces unless enclosed in []. Not
recommended.
Is case insensitive (as is all SQL).
Must not be a SQL keyword (can be a reserved word).
Must be unique (for a particular user.table)
The columnNames:
Same rules as tables, but column names must only be
unique within the table.
Hint - Use a naming convention different
from your PL.

CREATE TABLE
dataType = INT, CHAR, VARCHAR, DATE,
etc. (domain)
size = the size of the data.
Generally default is fine for non-textual data.
With CHAR/VARCHAR/NVARCHAR, default of
1 is usually not acceptable.

CREATE TABLE
The user who creates the table is the owner
of the table. Other users must be granted
access, and may have to preface the name of
the table with the name of the owner.
If USER1 creates TABLE1, USER2 sees it as
USER1.TABLE1
In this way users can each have a table with
the same name, e.g., MY_PREFERENCES.
The user name can be omitted if there is no
ambiguity.

EXAMPLE

An Emergency Room needs a table for all the
doctors that work there. Define and create it.

Indexes
“An index is an ordered (alphabetic or numeric) list
of all the contents of a column or groups of
columns in a table.” -Gruber.
Tables by definition are not ordered. Indexes
provide an external structure that gives the table
order.
Indexes make searches, particularly JOINs and
GROUP BYs, much faster.
On the flip side, Indexes make inserts to and deletes
from the table slower, as well as updates to the
indexed fields.

An Index Example
Assume the telephone company keeps a phone # table, where
each phone number is matched with the SSN of the owner of
the phone. Entries are added and deleted to the table as account
information changes. There are 1,000,000 entries in the table.
Assuming we can read 100 records in one second,
approximately how long does it take to find a phone number?
How can we make this faster?
How much faster?

How Indexes (Indices) Work
• They use the concept of a binary search - in an ordered list,
check the middle item and you can eliminate half the list
(see example).
• The external structure is a binary tree (actually, a b-tree of
extents, but we’ll ignore that).
• Each node on the tree can have a left and a right child.
ALL nodes to the left of a node have lower values, all to
the right have a higher value.
• Moving down the the tree is O(log n).
• The tree must be ‘rebalanced’ on occasion; rebalancing is
O(n^2).

Indexes
Indexes work by themselves once created.
Syntax:
CREATE [UNIQUE] INDEX <indexName>
ON <tableName> (columnName1, […,
columnNameN] )

For example,
create index PAGER_NDX on DOCTOR (PAGER_NUMS)

Its nice to have all indexes have some part of
their name in common.

Indexes
If UNIQUE is specified, then no duplicates are allowed
in the field or collection of fields.
No NULLs are allowed in unique indexes.
Attempts to violate a unique index cause an error.
The data in the table must fulfill the requirements for a
unique index before the index is created.
And now that we’ve talked about unique indexes…
don’t use them. Its more clear to use a TABLE
CONSTRAINT and a normal index.

Deleting Objects - DROP
DROP INDEX <tableName>.<indexName>
The reason why indexes are named
Table name should preface the index name.
DROP TABLE <tableName>
ANSI 92 - table must be empty to drop it.
SQL Server - “The command(s) completed
successfully”

ALTER TABLE
Non-standard but widely available.
Allows adding of columns to table (most DBs)
SQL Server allows adding and removing of
columns or constraints as long as no violations
are caused.

ALTER TABLE
Syntax in Transact-SQL Help! (NON-
Standard)
ALTER TABLE EMPLOYEE ADD AGE
INT
The alternative:
Create temporary table with new column.
Select rows from old into new, using nulls for new col.
Delete and drop old table.
Recreate old table with new column
Select rows from temp into old.
Delete and drop temp table.

Implementing the Databese Server session 02

Recommended

More Related Content

What's hot (18)

Viewers also liked (6)

Similar to Implementing the Databese Server session 02 (20)

Recently uploaded (20)

Implementing the Databese Server session 02