UNIFOR Global Semantic Serializability: An Approach to Increase Concurrency in Multidatabase Systems Angelo Brayner University of Fortaleza Brazil Angelo Brayner Theo Härder University of Kaiserslautern Germany CoopIS - Trento, September 2001 1 Contents UNIFOR Motivation Multidatabase System Model Global Semantic Serializability Concurrency Control Protocols Conclusions Angelo Brayner CoopIS - Trento, September 2001 2 Motivation (1) UNIFOR Integration of heterogeneous databases is a strategic requirement Integration of heterogeneous databases in a enterprise Integration of heterogeneous web databases Web as a large collection of distributed autonomous and heterogeneous databases Integration of ubiquitous databases mobile heterogeneous databases providing data everywhere Angelo Brayner CoopIS - Trento, September 2001 3 Motivation (2) UNIFOR Multidatabase technology Efficient solution for integrating a collection of autonomous and heterogeneous databases Local databases Created independently without considering the possibility of being integrated in the future Operate autonomously Local autonomy is a key feature Multidatabase Angelo Brayner Collection of local databases CoopIS - Trento, September 2001 4 Motivation (3) UNIFOR Multidatabase System (MDBS) Software component to manage a multidatabase Provides DBMS functionalities Multidatabase environment Global transactions Submitted to the MDBS Access and update local database objects Local transactions Submitted to local database systems Angelo Brayner CoopIS - Trento, September 2001 5 Motivation (4) UNIFOR Classical transaction-processing Model "Syntactic" serializability Serialization order of all active transactions must be known For identifying correct execution of concurrent transactions Efficient criterion for synchronizing operations of short transactions Angelo Brayner CoopIS - Trento, September 2001 6 Motivation (5) UNIFOR Concurrency control problem in MDBSs Global transactions Involve operations on multiple local databases Long-living transactions MDBS does not have any information about the execution (serialization) order of local transactions Classical transaction model is inefficient for solving the CC problem in MDBSs Angelo Brayner CoopIS - Trento, September 2001 7 Multidatabase System Model (1) UNIFOR Global Transactions Gi Gj MDBS Global Scheduler Interface Server 1 Local Transactions SUBi1 SUBj1 Interface Server n SUBin DBMS DB Angelo Brayner Log Global Recovery Manager LDBS1 DB CoopIS - Trento, September 2001 Global Log Log SUBjn DBMS Global Transaction Manager Local Transactions LDBSn 8 Multidatabase System Model (2) UNIFOR MDBS 1. A set LD={LDBS1, LDBS2, … , LDBSn} of local database systems 2. A set L={L1, L2, … , Ln} Each LK represents a set of local transactions executed at LDBSK 3. A set G={G1, G2, … , Gn} of global transactions Angelo Brayner CoopIS - Trento, September 2001 9 Multidatabase System Model (3) UNIFOR Local Schedule SK Models the execution of interleaved operations belonging to local and global transactions Executed at LDBSK Global Schedule SG Models the execution of all local schedules Angelo Brayner CoopIS - Trento, September 2001 10 GS-Serializability Model (1) UNIFOR Assumptions An MDBS integrates a collection of preexisting local databases (LDBs) A collection of disjoint sets of objects Each set represents a single local database Semantic Unit An update operation executed by a global transaction G on an object of a particular semantic unit does not depend on values of objects belonging to other semantic units previously read by G Angelo Brayner CoopIS - Trento, September 2001 11 GS-Serializability Model (2) UNIFOR Module-structured Transaction Operations are grouped into subsequences Modules Encompasses operations on objects of only one semantic unit Example DB={A, B, C, D, E, F, G} SULDBS1={A, B, C} SULDBS2={D, E, F, G} T1= r1(G) w1(E) w1(C) r1(B) Module T2= r2(G) w2(C) w2(E) r1(B) Angelo Brayner CoopIS - Trento, September 2001 12 GS-Serializability Model (3) UNIFOR GS-Serial Global Schedule Local schedules are conflict serializable and Serial execution of modules belonging to global transactions Example G1=r1(G)w1(E)w1(C)r1(B); G2=r2(A)w2(B)w2(D)r2(E) SC= r2(A)w2(B)r1(G)w1(E)w2(D)r2(E)w1(C)r1(B) SC is GS-Serial SC is not conflict serializable Angelo Brayner CoopIS - Trento, September 2001 13 GS-Serializability Model (4) UNIFOR GS-Serial Schedules preserve multidatabase consistency Correctness criterion for MDBSs GS-Serializable Schedule S Local schedules are conflict serializable and The execution order of global transactions in S is conflict equivalent to the execution of a GS-Serial schedule over the same set of transactions Angelo Brayner CoopIS - Trento, September 2001 14 GS-Serializability Model (5) UNIFOR Identifying GS-Serializable Schedule Since existing DBMSs yield conflict serializable schedules The GTM has solely to verify the execution order of global transactions A graph-based method Angelo Brayner The Semantic Serialization Graph (SSG) CoopIS - Trento, September 2001 15 Concurrency Control in MDBSs UNIFOR Concurrency Control Protocols Conservative Based on a locking mechanism Aggressive Management of an always acyclic graph Angelo Brayner Based on the SSG CoopIS - Trento, September 2001 16 Conclusions UNIFOR GS-Serializability Model Increases concurrency in MDBSs More permissive than syntactic serializability Increases concurrency in mediator-based systems Each web database can be seen as a semantic unit Can be applied to control concurrency in ubiquitous database Mobile database can be defined as a semantic unit Angelo Brayner CoopIS - Trento, September 2001 17