A Concurrency Control for Transactional Mobile Agents

Mobile agents are autonomous objects that can migrate from one host to another within a computer network. Concurrent execution of multiple mobile agents requires a concurrency control for data consistency. Some mobile agent applications, such as information retrieval, do not need a concurrency control, but most applications, for example workflow and electronic commerce, require such a control on their mobile agents because a task dependency exists between them. In this paper, we define the requirement of transactional mobile agents and compare the performance of some concurrency control methods to select one having the highest concurrency ratio for mobile agents. According to the results of our experiments, we suggest using a timestamp-ordering protocol having alternative tasks for concurrency control on mobile agents. If long-lived and short-lived transactions coexist, it is much better to use a timestamp-ordering protocol having alternative tasks than a locking-based protocol according to performance aspects. Since the timestamp-ordering protocol having alternative tasks assures global serializability, we can maintain the data consistency across the system. To use the timestamp-ordering protocol having alternative task as a concurrency control for mobile agents in distributed environments, we also describe how to solve the cascading abort and global time synchronization problems.