Research On Transaction Model Based On LCWA And Transaction Decomposition

Serializability has been widely accepted as the correctness criterion for concurrent access in database systems. However, in some newer challenging areas, serializable exection may not be feasible. Although serializability guarantees the correctness of concurrent access, it meanwhile causes lots of loss in concurrency and brings down the system throughput. In order to improve throughput and reduce response time, most current applications of transaction processing run at the isolation level lower than serializability. Therefore, many research put forward some new correctness criterions'based on weak consistency, such as CAP theory, BASE model, ultimate consistency of data, etc.,. These new criterions aim to making the tradeoff between correctness and efficiency, which is maxmizing the system effiency based on an acceptable level of correctness.First, based on the local world assumption and in the view of complex system, data can be viewed as the resources of respective sub-systems. Based on the assumption, each site is viewed as relatively independent sub-systems conntected with other sites. These sub-systems exchange information primarily through data replication. In the paper, it redefines the correctness of transaction from the point of individual transactions and introduces a novel transaction model based on the traditional ACID transaction. A multiple version control method for the existing multiple versions of data in this model is introduced. This version control method provides flexible isolution levels to maximize the system concurrency.Second, with the mathematical definations of transaction semantic, a new method of semantic-based decomposition of transactions is proposed. According to this method, any transactions can be decomposed into steps with smaller granularity that is excuted as an atomic unit. One schedule is an execution sequence of steps. If the transaction semantic is correct, the corresponding schedule is current; or else, the schedule is false. When the execution is false, transaction compensation instead of the inefficient transaction rollback is used to remedy it. In order to implement extensive simulation, the step schedule problem of the database is then we then simplified as a classical NPC problem, Maximum Set Packing Problem. Based on this simplification, a simulation system is designed. An algorithm based on searh with pruning strategy is provided to optimize the step schedule of decomposed transactions. Finally, the theoretical analsis and the simulation results show that our semantic-based transaction decomposition can improve the throughput of the database system significantly.