60 likes | 206 Views
CSC 556 – DBMS II, Spring 2013. May 1, 2013 DBMS ACID Properties & Concurrency Control. DBMS ACID properties. Atomicity = transaction is all or none. Consistency ensures database integrity. Isolation avoids interaction of concurrent processes interacting with the database.
E N D
CSC 556 – DBMS II, Spring 2013 May 1, 2013 DBMS ACID Properties & Concurrency Control
DBMS ACID properties • Atomicity = transaction is all or none. • Consistency ensures database integrity. • Isolation avoids interaction of concurrent processes interacting with the database. • Durability ensures persistence in the face of crashes, power losses and errors.
FileMultiSet ACID support • Uses expensive SYNC flag to force flushes. • recoveryFile gets a sequential SYNC’d copy of each mutation before mutating main DB. • transact() sets a transaction boundary. • docommit() cleans up recovery records. • dorollback() reads records from recovery file & restores state of main DB. • A crash occurring during a transaction or during a rollback still has recovery records in recovery file to affect repair. • LockedFileMultiSet uses expensive file locking for concurrency control.
More efficient transaction support • Serialize access to DB files via server process or synchronized server processes. • This approach avoids expensive file I/O locks. • http://docs.oracle.com/cd/E11882_01/server.112/e16508/intro.htm • http://www.siue.edu/~dbock/cmis565/module1-architecture.htm • Data versioning notes timestamp at start of transaction. At time of commit, if timestamp of data has changed, roll back and redo.
Functional programming approach • No data structure is ever mutated. • Construct new tuples and abandon references to their predecessors. • Eventually garbage collect 0-reference tuples. • Use lazy evaluation to defer repeated small changes to data structures. Perform mutations in batches the first time they must be read. • Indices (B-trees) must still be mutated.
Concurrency control • Timestamps / version numbers applied upwards from data records through B-tree. • These establish oldest records in a subtree. • Timestamps / version numbers applied downwards from root to data records during lookup-preceding-mutation. • These establish newest records in a subtree. • Use as a basis for tentative updates requiring rollback and restart when initial query data change during a transaction.