| In this work a new Two-Level Checkpointing technique for distributed systems is presented. This technique is mainly devoted to distributed control system although it can be implemented in general distributed environments.; For the development of the new technique, first, an analysis and classification of the existent Checkpointing methodologies in distributed systems is carried out.; Then, the results of a study that analyze the feasibility of the implementation of a Checkpointing technique in a typical industrial control system are presented. Later, this study will be enlarged with the study of the performance of the system with different Checkpointing techniques.; Next, the study of the special features of distributed control systems leaves to the statement of a number of requirements for a Checkpointing technique suitable for this kind of environment.; Starting from these requirements and keeping in mind the results of the previous developed studies, a new Checkpointing strategy is presented. This method establishes two levels of Checkpoints (local and remote). Once the technique is described, its implementation on different distributed system architectures is commented.; Later, the results of the evaluation of the new Two-Level Checkpointing technique are presented. Both the performance and the dependability (mainly, the reliability and security) of the system with this new technique is studied. In these studies, the influence of the internal architecture of the system's nodes is also taken in account.; As a result of all these analyses, it can be observed that the performance loss of a system implementing a Two-Level Checkpointing technique is quite acceptable. Only much more complex Checkpointing strategies achieve a better performance level. On the other hand, techniques that store two levels of Checkpoints improve the reliability of the system notably. Although this improvement can be observed for all the studied distributed system architectures, it is much more important for architectures with low level of redundancy. This type of architectures is just the most frequently used in distributed control systems.; The new Two-Level Checkpointing technique supposes, therefore, an improvement of the reliability of distributed control systems with a minor loss of performance. |
