Research On Management Of Transaction Versions For Dynamic Component Update

Component-Based Software Development(CBSD) is widely used to improve de-velopment productivity and software quality. These component-based applications of-ten need updates to fix bugs or address changes in their execution environments. How-ever, the traditional off-line update method is not acceptable for mission critical sys-tems, which must be non-stopping. In such situations we need dynamic software update (DSU).In this thesis, we analyzed the existing dynamic update algorithms, and summa-rized their limitations on the system safety, timeliness, disruption, and communication overhead. To address these deficiencies, we proposed a dynamic update algorithm by versioning ongoing transactions. We also conducted experiments to evaluate its effec-tiveness, and applied the algorithm in the Spring framework.The main work includes as follows:● We analyzed the existing dynamic update algorithms, and summarized their limi-tations on the system safety(ensuring system consistency during and after system update), timeliness(dynamic update completion time), disruption(the time that up-date operations take), and communication overhead(communication overhead in-troduced by update operations). To address the deficiencies of existing algorithms, we proposed a dynamic update algorithm by versioning ongoing transactions.● Based on the algorithm, we proposed a mechanism which manages versions of on-going transactions, where transactions are labeled with proper versions, and will be served by the proper component. Besides, the component to be updated can be removed at the safe time, which ensures the system safety.● We conducted experiments through simulations with the tool DEUS. The results showed that, compared to existing approaches, our approach performed signifi-cantly better in communication overhead(90% less), timeliness(10% faster) and disruption(15% lower).● Based on our solution, we extended the Spring framework to support dynamic up-date. In addition, we selected a Spring application to experimentally evaluate the correctness of our extended Spring framework.