Planning Warehouse Storage Chapter 9

Data Partitioning zBreaking up a data into separate physical units that can be handled independently zEase of: - Restructuring - Reorganization - Removal - Recovery - Monitoring - Management - Archiving - Indexing Order table Drop Other data is not affected Add

Objects to Partition zTables: - Fact - Dimension zIndexes

Horizontal Partitioning zTable and index data are split by: - Time - Sales region or person - Geography - Organization - Line of business zCandidate columns appear in WHERE clause zAnalysis determines requirement

Vertical Partitioning You may use vertical partitioning when: zSpeed of query and update actions is improved by it zUsers require access to specific columns zSome data is changed infrequently zDescriptive dimension text may be better moved away from the dimension itself

Partitioning Methods zRange partitioning (Oracle8 and Oracle8i) zHash partitioning (Oracle8i) zComposite partitioning (Oracle8i)

Star Query Optimization Optimum performance with star schema models 1. Dimensions are queried to create a 2. Cartesian product, computed against 3. Smaller reference table. 4. The result is joined to 5. A fact table to produce a query result.

Star Transformation Time_Table Fact_Table Product_Table Market_Table

Indexing Indexing is used because: zIt is huge cost saving, greatly improving performance and scalability zCan replace a full table scan by a quick read of the index followed by a read of only those disk blocks that contain the rows needed

B-Tree Index zMost common type of indexing zUsed for high cardinality columns zDesigned for few rows returned

Bitmap Indexes zProvide performance benefits and storage savings zStore values as 1s and 0s zUse instead of B-tree indexes when: - Tables are large - Columns have relatively low cardinality

Oracle8 and Oracle8i Index Enhancements zOracle8 index enhancements: - Partitioned index - Index-organized tables zOracle8i index enhancements: - Function-based index - New bitmap index improvements - Online index build and rebuild - Descending index - Statistics can be collected when an index is created

Protecting the Database zRAID is essential with large databases zRAID improves: - Reliability - Storage management zThere are different levels of RAID zYou can eliminate disk contention with disk striping

RAID 0: Striping The file is written to a four-drive disk array: zBlock 1 on Drive 1 zBlock 2 on Drive 2… zBlock 5 in another sector on Drive 1 Disk array controller

RAID 0: Striping zBenefits: - Good for simultaneous reads and writes - No redundancy - Scalable zLimitations: - Not recommended for mission-critical systems - No recovery from data loss - One bad sector affects entire disk of data

RAID 1: Mirrored Disk Disk array controller Disk 1 Mirror Disk 2 Mirror Copy of files stored on mirror disk

RAID 1: Mirrored Disk zBenefits: - Complete data redundancy - No performance penalty - Improves reads - Scalability zLimitations: - Highest cost of all RAID configurations

RAID 5: Independent Disk Array Disk array controller Disk 1Disk 2Disk 3Disk 4 Data striped with parity across array

RAID 5: Independent Disk Array zBenefits: - Efficient data integrity - Data reconstruction - Multiple concurrent seeks across array - Scalable zLimitations: - Disk overhead - Data write rate

Backup zPlan at the design stage zUse hot backups for VLDBs zBack up necessary components: - Fact and dimension data - Warehouse schema - Metadata schema - Metadata zExport/Import utility - Disk space - Time

Summary This lesson discussed the following topics: zExplaining vertical partitioning and horizontal partitioning zDistinguishing the different types of partitioning methods zDistinguishing between B-tree index and bitmap index zUnderstanding why warehouse typically uses RAID 0, or 5 to protect the database