1.1.1 Relational database systems
•Information is stored in 2D-tables
•Rows in a table are records
•Columns in a record are fields
•A key, consisting of one or more fields, that uniquely identifies a record
•A dictionary is a table that describes all the tables
•Fully-partitioned: if each table is stored at exactly one physical site
•Fully-replicated: if each table is stored at all physical sites
•Natural distribution: data are kept at the local site
•Query operations:
- Select - picking records
- Project - picking fields
- Join - merging of tables
•Predicate: a condition between fields used to manipulate the queries
1.1.2 Issues for distributed database systems
• Distribution of tables to sites
• Natural distribution of data at various sites
• Fully partition of systems
• Fully replicated systems
• Important factors
• Replicates the dictionary at every site
• Frequency of request to a table from a site
• Storage capacity at each site
• Communication costs between sites
1.1.2.1 - Query processing
• Query response time (for interactive applications)
• Total bandwidth consumed (for batch applications)
• Approach
- Optimizing specific query based upon specific statistical conditions
- The query site will either “estimate” or request the related sites to report the related time and cost of moving the data before deciding on an actual query sequence
1.1.2.2 - Concurrency control
• Maximize the amount of parallel activity while maintaining the semantic integrity of the data
• Approach
- A transaction: a set of reads, followed by some processing, and then a set of writes
- A log is the time ordered sequence of reads and writes performed on the database
- A log is serial if each read is immediately followed by an appropriate write
- There is no known algorithm that allows serial logs, all serializable logs, and all other logs that leave the database consistent
- Most algorithms achieve serializable logs by allowing transaction to lock part of the database
- The lock could be applied on the full database, or some tables, records, fields, and physical sectors
- Deadlock occurs when two queries want to lock certain resources that have already been locked by each other