Physical Design Practical Issues of Data Placement Access efficiency Driven by number of disk reads Join logic SELECT * FROM A, B WHERE A.KEY = B.KEY AND A.CONDITION = TRUE; Read Key A IF condition = TRUE Read All B’s Match current A Read Next A Improving Efficiency Indexing Locate specific records quickly Denormalization Reduce the number of joins Partitioning Reduce the amount of data INDEXES: PRIMARY KEY INDEXES Maintain record integrity by assuring that no duplicate values exist CREATE UNIQUE INDEX PRODUCT (PRODUCT_NO); PRODINDEX ON Indexes: Nonkey Indexes Allow efficient access to files using nonkey data. CREATE INDEX DESCRIP ON PRODUCT (DESCRIPTION); Indexes: Clustering Indexes Determine the physical storage sequence for data CREATE INDEX DESCRIP ON PRODUCT (DESCRIPTION) CLUSTER; Denormalization Selective violations of normalization principles for access efficiency Entities with 1:1 relationships. It may be wise to combine these into a single table. Denormalization M:N relationships with non-key attributes (gerunds). Extracting attributes from one entity into another requires a join that accesses the link file and the associated entity. It may be worth violating 2NF or 3NF rules to duplicate commonly needed attribute values in the link file or even in the "parent" entity. Denormalization Reference data. These are 1:M relationships in which the 1 side is a lookup table for indicator keys in the "many" file. If these lookups are not used in many files and there are not many instances of the "many" entity for each "one" occurrence, it may be a good idea to move the reference attributes to the parent file and violate 3NF. Partitioning Splitting data into groups for improved performance Vertical Select frequently used attributes for a primary table and transfer others to a sub-class file that is less frequently accessed Horizontal Duplication Distributed Data (Maintaining Concurrency) Two Phased Commit Prepare commit by locking all versions Execute the commit simultaneously