Research Of The Long Transaction Handling Mechanism Based On The Workflow

With the computer and network's wide use in daily life, workflow technology has become a critical technology for enterprise on coordinating business process, enhance competitive and strain capacity. Wide scale workflow application possesses transactional properties in distributed and heterogeneous environment. So transactions in workflow applications are different from traditional flat ones. They are typically long running and with coarse-grained operations. And the recovery guarantees of different activities are dissimilar. Further more, semantics of the applications are rich. For example, a workflow instance may terminate successfully in different ways. So even advanced transaction models and methods are not suitable for workflow applications. In the early 1990's, the term transactional workflow was introduced to clearly recognize the relevance of transactions to workflows. The development of transactional workflow technology is of great importance for pushing workflow technology into practice. Transactional workflow technology has been a research focus of both transaction management and workflow areas.In order to resolve the shortage of workflow transactions in transactional concurrent control and failure recovery, first, we proposed a concurrency control based on the isolation sphere, it can avoids the lost update and the dirty read phenomena base on the appropriate setting of the isolation sphere and the isolation property, the user can set the isolation level based on the business process, and improve the concurrency of the process instance. Second, we propose HFR, a hierarchical failure recovery algorithm to ensure semantic atomicity in case of failures. Different from traditional methods, it supports both transactional and non-transactional sub processes, and sub processes of a workflow are nested structured. HFR determines the end compensation point dynamically in a hierarchical bottom-up manner, so the compensation sphere is confined and compensation costs is reduced. We prove the correctness of the proposed algorithm. Performance analyses show that it is preferable than non-hierarchical methods.To sum up, we have studied on several key problems in transactional workflow area in this dissertation. Our contributions listed above are helpful for enhancing the reliability of workflows in both theory and practice.