摘要
We have considered a method called Enhanced Rollback Migration Protocol, which potentially has the effects of compressing the period of compensations in a long-lived transaction, since before. In general, a compensation transaction can recover an irregular status of a long-lived transaction into the original status without holding unnecessary resources by making its consistency tentatively loose. However, it has also been pointed out that there is a difficulty of maintaining the isolation between a pair of transactions when executed in parallel. In particular, this could be more prominent under modernized scalable cloud environments. Thus, there is a proposal for concurrency control for the service level. However, there is still another risk that more computer resources will be consumed than actually necessary and an unnecessary stagnation of the processing will be caused if concurrency control is naively applied without careful consideration. Therefore, we need to implement a functionality which can optimize the processing of a long-lived transaction by selecting a suitable method between concurrency control and compensation transactions. In this paper, we propose a method in which optimistic concurrency control is applied for long-lived transactions. Furthermore, a pair of verification phases is carried out. At the beginning from a safe point, an attempt of verification is done. Then if the difficulty of isolation on a long-lived transaction executed under a competitive situation is estimated, concurrency control for the service level is applied. Alternatively, a long-lived transaction without any concurrency control is executed. At the next reachable safe point, another attempt of verification is performed. Then if a failure of serialization is detected, a set of compensation transactions is invoked to recover the original long-lived transaction by returning to the first safe point. We evaluated this approach by using numerical simulations and confirmed the basic features. This approach can realize optimizing and enhancing the performance of a long-lived transaction. We regard this approach applicable even to the modernized scalable cloud environments.
We have considered a method called Enhanced Rollback Migration Protocol, which potentially has the effects of compressing the period of compensations in a long-lived transaction, since before. In general, a compensation transaction can recover an irregular status of a long-lived transaction into the original status without holding unnecessary resources by making its consistency tentatively loose. However, it has also been pointed out that there is a difficulty of maintaining the isolation between a pair of transactions when executed in parallel. In particular, this could be more prominent under modernized scalable cloud environments. Thus, there is a proposal for concurrency control for the service level. However, there is still another risk that more computer resources will be consumed than actually necessary and an unnecessary stagnation of the processing will be caused if concurrency control is naively applied without careful consideration. Therefore, we need to implement a functionality which can optimize the processing of a long-lived transaction by selecting a suitable method between concurrency control and compensation transactions. In this paper, we propose a method in which optimistic concurrency control is applied for long-lived transactions. Furthermore, a pair of verification phases is carried out. At the beginning from a safe point, an attempt of verification is done. Then if the difficulty of isolation on a long-lived transaction executed under a competitive situation is estimated, concurrency control for the service level is applied. Alternatively, a long-lived transaction without any concurrency control is executed. At the next reachable safe point, another attempt of verification is performed. Then if a failure of serialization is detected, a set of compensation transactions is invoked to recover the original long-lived transaction by returning to the first safe point. We evaluated this approach by using numerical simulations and confirmed the basic features. This approach can realize optimizing and enhancing the performance of a long-lived transaction. We regard this approach applicable even to the modernized scalable cloud environments.
