An effective deadlock prevention mechanism for distributed transaction management

Recently there has been an increasingly use of SOA infrastructure. Most of business applications need the transaction management to guarantee the system consistency, typically by resource locks. High availability and performance of resource managers are extremely critical for business services. Researchs have shown that deadlock can be easily produced when multi-process try to access the shared resources if some basic synchronization mechanism is adopted. In addition, existing resource locking based two-phase commit protocol requires that each transaction obtains all needed locks before the second commit phase. This locking mechanism may lead to potential deadlocks and can pose significant limitations on applications that need to share resources between transactions. In this paper, we propose a novel solution, for distributed transaction services, utilizing the advanced replicas and timestamps mechanisms to prevent deadlocks. Our approach makes use of existing lock policies with replicas to solve guaranteed deadlocks and compares conservative locking and timestamps based restart policy for global transaction deadlock prevention. Further, we also demonstrate that our approach significantly improves the performance for both serialization and deadlock prevention. Our experimental results show that no deadlock occurs after utilizing our prevention mechanisms. Finally, we also observe that timestamps based restart policy is more suitable for global resource allocation and outperforms over related solutions.