Table Partitioning:
Secret Weapon for Big Data Problems
October 15-18, 2013 John Sterrett, Sr. Database Admin Advisor
Charlotte, NC DELL
About John
2
How to contact me?
http://johnsterrett.com/go/partition
Blog: http://JohnSterrett.com/
Twitter: @JohnSterrett
LinkedIn: http://linkedin.com/in/johnsterrett
3
Agenda
Big Data starting to slow you down? Data growth putting your
maintenance tasks in danger of not meeting your SLAs? Wish you
could archive old data with minimal impact to your tables during the
archive process or that you could eliminate most of the data in your
tables when you query them? If so, it’s time you consider implementing
table partitioning to help with general performance and reduce your
window for completing maintenance tasks.
• Learn if Partitioning can help you manage big data.
• Understand how partitioning works.
• Learn how to maintain table partitioning with an automated
sliding window.
4
Assumptions…
• You are not familiar with Table Partitioning and want
to see how it works.
• You have Enterprise Edition
• Highly Transactional Tables with over 50 GB of data
• General Maintenance Tasks are in danger of not meeting
your SLA
• Purge and/or Archive process is slowing you down.
5
The Big Question…
HOW CAN TABLE PARTITIONG MAKE MY LIFE
EASIER?
6
How partitioning helps me?
• Reduce Maintenance Tasks
7
How partitioning helps me?
• Reduce Maintenance Tasks
• Improves Purging and/or Archiving
8
How partitioning helps me?
• Reduce Maintenance Tasks
• Improves Purging and/or Archiving
• Improves Performance
9
Big Question…
HOW DOES TABLE PARTITIONING WORK?
10
High Level…
11
High Level…
Partition
Table
Partition
Scheme
Partition
Function
12
Selecting The Partition Column
Only get one column, use it wisely!
Column should be a highly used filter.
• Review index usage statistics
• Review Missing index statistics
• Review Queries and Talk to developers ;-)
Column must be part of clustered index
13
Partition Function
Defines the data type used to distribute data into partitions.
Assigns boundary values to split data between partitions.
Assigns the RANGE for boundary values
Partitions = Boundary values + 1
NULL values go to left most partition
14
Partition Function
Partition functions are not static. They can change over
time with SPLIT and MERGE statements.
Range LEFT is used by default.
15
Partition Function - Range
{min….100}, {101…200}, {201...300}, {301…max}
{min…99}, {100…199}, {200…299}, {300…max}
16
Partition Scheme
Assigns a partition function to a partition scheme
Assigns filegroups to partitions
17
Partitioning
DEMO
October 15-18, 2013 | Charlotte, NC
Big Question…
HOW DOES TABLE PARTITIONING IMPROVE MY
MAINTENANCE TASKS?
19
Improving Maintenance Tasks
Backup and restore filegroups based on business priorities.
Index Maintenance by partition
New Features in SQL 2014
• Rebuild index online by partition
• Incremental Statistics by partition (SQL 2014 CTP 2)
20
Incremental Statistics
Just added in SQL Server 2014 CTP2
From Books Online:
“When ON, the statistics are recreated as per partition
statistics.”
Cannot be used for the following:
• Statistics created with indexes that are not partition-
aligned with the base table
• Filtered Indexes
21
HOW DOES TABLE PARTITIONING IMPROVE
PERFORMANCE?
22
Partition Elimination
23
Skip-Scan: Seek keys
24
Database Compression by Partition
25
Partitioning
DEMO
October 15-18, 2013 | Charlotte, NC
HOW DOES PARTITIONING IMPROVE ARCHIVING AND
PURGING?
27
Real World Example:
28
Quickest way to move billions of rows
Meta data swap is quickest way to move billions of rows
assuming you can get a schema lock.
29
Sliding Window Goals
SPLIT and MERGE with empty partitions
Use SWITCH to do meta-data swaps
Minimize all physical data movement
30
Range Right - MERGE
31
Range Left - MERGE
32
Range Right - SPLIT
33
Range Left - SPLIT
34
Visual Sliding Window Example
35
Visual Sliding Window Example
Partiton1 Partiton2 Partiton3 Partiton4 Partiton5
(EMPTY) (DATA) (DATA) (DATA) (DATA)
FG1 FG2 FG3 FG4 FG5
Visual Sliding Window Example
Partiton1 Partiton2 Partiton3 Partiton4 Partiton5
(EMPTY) (DATA) (DATA) (DATA) (DATA)
FG1 FG3 FG4 FG5
Staging
Table
(EMPTY)
FG2
Visual Sliding Window Example
Partiton1 Partiton2 Partiton3 Partiton4 Partiton5
(EMPTY) (EMPTY) (DATA) (DATA) (DATA)
FG1 FG3 FG4 FG5
SWAP
Staging
Table
(DATA)
FG2
Visual Sliding Window Example
Partiton1 Partiton2 Partiton3 Partiton4 Partiton5
(EMPTY) MERGE (EMPTY) (DATA) (DATA) (DATA)
FG1 FG3 FG4 FG5
Staging
Table
(DATA)
FG2
Visual Sliding Window Example
Partiton1 Partiton2 Partiton3 Partiton4
(EMPTY) (DATA) (DATA) (DATA)
FG1 FG3 FG4 FG5
Staging
Table
(DATA)
FG2
Visual Sliding Window Example
Partiton1 Partiton2 Partiton3 Partiton4
NEXT
USED
(EMPTY) (DATA) (DATA) (DATA)
FG1 FG3 FG4 FG5
Staging
Table
(EMPTY)
FG2
Visual Sliding Window Example
Partiton1 Partiton2 Partiton3 Partiton4 Partiton5
(EMPTY) (DATA) (DATA) (DATA) (DATA)
FG1 FG3 FG4 FG5 FG2
Sliding Window Steps...
Create partition swap table (if it doesn’t exist)
1. Insert partition swap meta data
2. Create staging table
3. Meta-data swap (partition 2 with staging table)
4. Merge Partitions #1 and #2
5. Mark next used partition
6. Split to create new partition
7. Update processed partition swap meta data.
43
Partitioning
DEMO
October 15-18, 2013 | Charlotte, NC
Questions…
http://johnsterrett.com/go/partition
Blog: http://JohnSterrett.com/
Twitter: @JohnSterrett
LinkedIn: http://linkedin.com/in/johnsterrett
45
Thank you
for attending this session and the
2013 PASS Summit in Charlotte, NC
46
October 15-18, 2013 | Charlotte, NC
