Every Byte Counts. Why Datatype Choices Matter. Andy Yun, Database Architect/Developer

Transcription

1 Every Byte Counts Why Datatype Choices Matter Andy Yun, Database Architect/Developer

2 Thank You Presenting Sponsors Gain insights through familiar tools while balancing monitoring and managing user created content across structured and unstructured sources. Unifying computer, storage, networking, and virtualization, Cisco UCS is the optimal database and business intelligence platform for SQL Server. Supporting Sponsors Solutions from Dell help you monitor, manage, protect and improve your SQL Server environment

3 Planning on attending PASS Summit 2015? Start saving today! The world s largest gathering of SQL Server & BI professionals Take your SQL Server skills to the next level by learning from the world s SQL Server experts, in over 190 technical sessions Over 5000 attendees, representing 2000 companies, from 52 countries, ready to network & learn Contact your Local or Virtual Chapter for an additional $150 discount. $1795 until July 12 th, 2015

4 Andy Yun SQL Server Architect & Developer Working with SQL Server since 2001 Chicago Suburban User Group Chapter Leader SQL Saturday Presenter Simple-Talk 2014 Tribal Awards Best New Community Voice - Runner Up Grant Fritchey s May 2014 Speaker of the Month sqlbek.wordpress.com /in/sqlbek 4

5 Data Types: Who Chose What? FirstName NCHAR(255) ZipCode INT Address NVARCHAR(MAX) DateCreated CHAR(10) OrderVolume BIGINT DailyRevenue FLOAT(53) Why not just use NVARCHAR(255), NVARCHAR(MAX), BIGINT, & DECIMAL(28,14) everywhere? 5

6 Presentation Goals Answer: How does SQL Server really store data records? Answer: What are some performance consequences of my data type choices? Takeaway: Assess your own systems & implement some best practices 6

7 Internals: How Data Is Stored 7

8 Employee Table CREATE TABLE dbo.employee_large ( EmployeeID FirstName LastName Gender SocialSecurityNumber Title ManagerID DateHired Salary Address HomeOfficeLocationCode VacationDaysRemaining RecordCreatedDate RecordCreatedBy RecordLastModifiedDate RecordLastModifiedBy ) UNIQUEIDENTIFIER PRIMARY KEY CLUSTERED, NCHAR(255), NCHAR(255), INT, NCHAR(50), NVARCHAR(255), UNIQUEIDENTIFIER, DATETIME, FLOAT, NVARCHAR(255), NVARCHAR(4000), INT, DATETIME, UNIQUEIDENTIFIER, DATETIME, UNIQUEIDENTIFIER 8

9 Fixed & Variable Length Data Types Fixed Width Data Types TINYINT SMALLINT INT BIGINT DATETIME SMALLDATETIME CHAR NCHAR DECIMAL FLOAT UNIQUEIDENTIFIER VARCHAR Variable Width Data Types NVARCHAR 9

10 Fixed & Variable Length Data Types Fixed Width Data Types & Storage Requirements TINYINT 1 byte SMALLINT 2 bytes INT 4 bytes BIGINT 8 bytes DATETIME 8 bytes SMALLDATETIME 4 bytes CHAR n bytes NCHAR (n * 2) bytes DECIMAL 5-17 bytes FLOAT 4-8 bytes UNIQUEIDENTIFIER 16 bytes Variable Width Data Types & Storage Requirements VARCHAR n + 2 bytes NVARCHAR (n * 2) + 2 bytes 10

11 FIXVAR Format Tag A Tag B Fsize Fdata Ncol Nullbits VarCount VarOffset VarData 1 byte 1 byte 2 bytes Fsize bytes Ceiling (Ncol / 8) 2 bytes 2 x VarCount n bytes Tag A & B Fsize Fdata Ncol Nullbits VarCount VarOffset VarData Status Bits A & B Total Fixed-Length Size Actual Fixed-Length Data Total Number of Columns NULL Bitmap (1 bit per column in table) Total Number of Variable Length Columns Variable Column Offset Array Actual Variable-Length Data 11

12 Fixed Width Columns Employee ID Gender Social Security Number Manager ID Date Hired UNIQUEIDENTIFIER INT NCHAR(50) INT DATETIME F-Data Size Fixed Data 144 bytes Employee ID 16 bytes Gender 4 bytes SSN 100 bytes Manager ID 16 bytes Date Hired 8 bytes 12

13 Variable Width Columns Title Address NVARCHAR(255) NVARCHAR(255) VarCount VarOffset Variable Width Data President al@countingbytes.com 2 bytes 2 bytes 2 bytes 18 bytes 40 bytes VarCount VarOffset Variable Width Data Database Administrator james@countingbytes.com 2 bytes 2 bytes 2 bytes 44 bytes 46 bytes VarCount VarOffset Variable Width Data Intern isabella@countingbytes.com 2 bytes 2 bytes 2 bytes 12 bytes 52 bytes 13

14 Employee Large CREATE TABLE dbo.employee_large ( EmployeeID FirstName LastName Gender SocialSecurityNumber Title ManagerID DateHired Salary Address HomeOfficeLocationCode VacationDaysRemaining RecordCreatedDate RecordCreatedBy RecordLastModifiedDate RecordLastModifiedBy ) UNIQUEIDENTIFIER PRIMARY KEY CLUSTERED, NCHAR(255), NCHAR(255), INT, NCHAR(50), NVARCHAR(255), UNIQUEIDENTIFIER, DATETIME, FLOAT, NVARCHAR(255), NVARCHAR(4000), INT, DATETIME, UNIQUEIDENTIFIER, DATETIME, UNIQUEIDENTIFIER Min. Size = 1304 bytes 14

15 Right Sizing Data Types Data Type Storage Value Range TINYINT 1 Byte 0 to 255 SMALLINT 2 Bytes -32,768 to 32,767 INT 4 Bytes -2,147,483,648 to 2,147,483,647 BIGINT 8 Bytes -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807 DATE 3 Bytes to SMALLDATETIME 4 Bytes :00:00 to :59:59 DATETIME 8 Bytes :00: to :59:

16 Right Sizing: Examples SocialSecurityNumber NCHAR(50) 1. Date? 2. Number? 3. String? 1. CHAR 2. NCHAR 3. VARCHAR 4. NVARCHAR CHAR(9) VacationDaysRemaining INT 1. Date? 2. Number? 3. String? 1. TINYINT 2. SMALLINT 3. DECIMAL DECIMAL(4, 2) 16

17 Employee Narrow CREATE TABLE dbo.employee_narrow ( EmployeeID INT IDENTITY(1, 1) PRIMARY KEY CLUSTERED, FirstName VARCHAR(100), LastName VARCHAR(100), Gender CHAR(1), SocialSecurityNumber CHAR(9), Title VARCHAR(40), ManagerID INT, DateHired DATE, Salary DECIMAL(10, 2), Address VARCHAR(100), HomeOfficeLocationCode TINYINT, VacationDaysRemaining DECIMAL(4, 2), RecordCreatedDate SMALLDATETIME, RecordCreatedBy INT, RecordLastModifiedDate SMALLDATETIME, RecordLastModifiedBy INT ) Min. Size = 104 bytes Employee_Large: Min. Size = 1304 bytes 17

18 Internals: Data Pages 18

19 Data Page: Overview 3 Types: In-Row Data Row-Overflow Data LOB Data A All Pages have Fixed Size of 8KB or 8,192 bytes B Comprised of 3 components: Page Header: 96 bytes Data Records Record or Slot Array: 2 bytes per record C 19

20 The 8KB Limit Functionally limited in what can fit in a data page Because I/O executed via data PAGES, not data RECORDS, in best interest to maximize # of records per data page Due to page size limit, potential for wasted space Typical data is stored In-Row VARCHAR(8000) & NVARCHAR(4000) use Row-Overflow Data Pages VARCHAR(MAX) & NVARCHAR(MAX) use LOB Data Pages 20

21 How Many Per Page? Employee Large Records Per Page Wasted Bytes Min Record Size 1304 Bytes bytes Employee Narrow Records Per Page Wasted Bytes Min Record Size 104 Bytes bytes 21

22 Why do we care? SELECT FirstName, LastName FROM Employee_Narrow WHERE LastName LIKE S% SELECT FirstName, LastName FROM Employee_Large WHERE LastName LIKE S% 22

23 What s the Difference? Employee Large Employee Small Number of Records Pages Size Pages Size KB 1 8 KB 30,000 5, MB MB 1,500, ,000 2,000 MB 19, MB 25,000,000 4,166,667 33,333MB 324,676 2,597 MB 23

24 Transaction Logging 24

25 Transaction Log Overhead INSERT, UPDATE, & DELETE statements all require writing to Transaction Log Does record size impact how much Transaction Log gets written? Larger Transaction Log means larger & longer backup operations Be mindful of ALL I/O. More I/O = more work for SQL Server 25

26 Demo sp_help sys.dm_db_index_physical_stats STATISTICS IO sys.dm_tran_database_transactions 26

27 Implicit Conversions 27

28 Alike or , :00:00 This is a String This is INT DECIMAL(6, 3) DECIMAL(6, 2) , VARCHAR(9) DATE :00:00 SMALLDATETIME VARCHAR(25) This is a String This is a String NVARCHAR(25) 28

29 Converting w. Data Type Precedence Value Conversions follow preset Precedence rules Smaller Data Type always upconverted to larger Data Type Condensed Precedence Chart 1. DATETIME 2. SMALLDATETIME 3. DATE 4. TIME 5. DECIMAL 6. BIGINT 7. INT 8. SMALLINT 9. NVARCHAR 10. NCHAR 11. VARCHAR 12. CHAR 29

30 Demo Implicit Conversions 30

31 Finding Implicit Conversions Finding Implicit Column Conversions in the Plan Cache By: Jonathan Kehayias 31

32 Suggested Best Practices Right-size your data types Remember the 8KB data page size limit Match your data types Be mindful of Internals related performance impacts 32