160 likes | 268 Views
Simulating MDBS Transaction Management Protocols. Ramon Lawrence, Ken Barker, Aruna Adil Department of Computer Science University of Manitoba, Canada {umlawren, barker}@cs.umanitoba.ca. Outline. Introduction The MDBS architecture and the transaction management problem Previous work
E N D
Simulating MDBS Transaction Management Protocols Ramon Lawrence, Ken Barker, Aruna Adil Department of Computer Science University of Manitoba, Canada {umlawren, barker}@cs.umanitoba.ca
Outline • Introduction • The MDBS architecture and the transaction management problem • Previous work • Simulation architecture • LDBS simulation • MDBS simulation • Simulation results and comparisons • Future work and conclusions
Database Terminology • transaction management - a protocol for mediating access to database data • global transaction - a transaction spanning more than one database • multidatabase system (MDBS) - a collection of autonomous, local databases participating in a global database system to share data
Global Transactions GTM subtransactions GTS GTS GTS GTS LDBS LDBS LDBS LDBS Local Transactions MDBS Architecture • Global Transaction Manager (GTM) • processes global transactions • insures information in all LDBSs is consistent • submits subtransactions to the GTSs for each LDBS • Global Transaction Servers (GTSs) • one for each LDBS • converts subtransactions from the GTM into a form usable by the LDBS and vice versa • Local Database Systems (LDBSs) • databases combined into MDBS • not changed in MDBS as still process local transactions
The Transaction Management Problem • Transaction management (TM) is the process of ensuring data access results in correct and consistent data • TM is harder in a MDBS because: • the global-level TM has no control over how LDBSs store and access data • databases are distributed so network delays affect query performance • must handle different database models and systems • data may be incomplete or inconsistent between databases
Previous Work • There has been several algorithms proposed for TM in a MDBS including: • Ticket serializability (Georgakopoulos) • Database serializability (Barker) • others … • The major problems with these algorithms is that they are too inefficient to be implemented in a real system.
Motivations and Goals • Motivations: • No simulation system capable of comparing current and proposed protocols has been implemented • Examine problems with current protocols and determine ways to improve their performance • Goals: • create a MDBS simulator capable of handling different configurations, load conditions, and protocols • simulate current protocols to determine their performance and design new protocols based on simulation results
MDBS Simulator Overview • A MDBS is simulated by combining several local database simulators into a logical entity. • Each local database simulator: • simulates a relational strict-2PL database • used in products by Oracle, Sybase, and IBM • models database structure, transaction frequency, and database management protocols • processes transactions originating from local and global queries
MDBS Simulator Overview (cont.) • At the global level, the global transaction manager: • processes global transactions and sends queries to the LDBSs • enforces global-level consistency • compiles statistics on transaction residence times and number of aborts
Using the MDBS Simulator • The MDBS simulator was used to compare two GTM protocols: • Ticket method protocol • Global Serial Scheduler (GSS) • Each protocol was simulated on the identical MDBS configuration with identical transaction loads at both the local and global levels
The Ticket GTM Simulation Results • The Ticket GTM protocol uses tickets at each database to detect conflicts and insure the data is consistent • optimistic algorithm with the potential for high concurrency and performance • Simulation results: • algorithm creates too many conflicts between global transactions which causes global deadlocks, global transaction aborts, and local database overloading. • Thus, the Ticket GTM protocol would not be a good protocol to use in a production environment.
The GSS GTM Simulation Results • The Global Serial Scheduler (GSS) schedules some transactions serially to prevent conflicts at local databases • prevents conflicts (pessimistic algorithm) • Results show that it has good performance despite some serial executions because it: • has no possibility for global deadlock or abort • executes transactions in the order they are submitted • Thus, the GSS algorithm is a better choice in a production environment.
Conclusions and Future Work • Conclusions: • Defining GTM protocols is difficult because of performance concerns • Simulating a MDBS provides insight into the protocols and possible improvements • Current protocols are inadequate for use in production multidatabase systems • Future Work: • allowing object-oriented local databases • simulating other GTM protocols and MDBS configurations • designing new, more efficient GTM protocols