Fault Detection Model Of Distributed Transaction Based On Extended π-calculus

Distributed transaction with autonomy, transparency and sharing features, which makesit the mainstream mode of data processing and has been used in many fields, such as trainticketing system, the banking system, e-commerce systems. Simultaneously, thecharacteristics of distributed transaction is large-scale, the components and data resourcesscattered heterogeneous, which makes the possibility of failure greatly increased. In adistributed transaction, the occurrence of information exchange between nodes will beaffected by the failure, to make matters worse; the system may be out of work. How toaccurately detect and locate faults becomes a hot issue for researchers.Currently, the focus of scholars is on how to ensure the consistency and fault tolerant ofthe transaction, but less attention on fault detection and differentiation. In the detection ofdistributed transaction, scholars usually only concerned with node failures, and assume thatall links are valid, or link failure are equivalent to a node failure, without making distinction.This approach is effective within a certain range, but when there is a node failure and linkfailure simultaneously, information failure cannot be distinguished due to a node failure orlink failure. Under this situation, building reliable database systems becomes undesirable.A hierarchical fault detection method is used for distributed transactions with large-scaleand distributed data resource. This model is close to the real network structure and has a clearpath probing, with fewer messages. The nodes of distributed transaction is grouped based onthe principle of close proximity of the nodes divided into a group before failure detection,such nodes with insignificant structure and communication are generally located within thesame LAN. A master node is elected in each group as an auxiliary coordinator to manage thesub-transaction.Formal methods are used to inspect the correctness of system before it put intoapplication. It’s necessary to formal modeling and deduces interaction. DπF calculus is usedas formal modeling language for distributed transaction, DπF calculus extends from Dpicalculus for distributed transaction failure areas. It owns the notion of location, networkenvironment and types which can naturally describe the features of distributed transaction,also the simulations theory which can effectively distinguish distributed transaction failures.But in DπF calculus, clock features of a distributed transaction cannot be described, which hasbrought great difficulties to analysis of the system interaction. Therefore, in this article DπFcalculus will be extended by adding a timer operator.The main work of this paper is as follows: （1）Adding assisted coordinators in distributed transaction model which effectivelyresolve communication overload problem of the coordinator.（2）Proposing hierarchical fault detection method for distributed transactions, faultdetecting and distinction before transaction commit can reduce unnecessary failure andenhance the reliability and availability of the commit protocol.（3）Extending timer operator to DπF and using its bisimulation theory for distributedfault modeling and distinction.In this paper, DπF calculus is selected as a modeling language for the characteristics ofdistributed transaction, and is extended representation of clock with timer operator. It isnatural to express distributed transaction failures, such as node failure, link failure, andprocesses migration. Finally, distinguish between these failures through bisimulation theory.Analysis the performance and characteristics of the distributed transaction failures can helpdesigners to modify and optimize the system, develop distributed fault detection, andfault-tolerant recovery. It is significance to build reliable distributed database.