Autonomic Web Services Based on Different Adaptive Quasi-Asynchronous Checkpointing Techniques

Companies, organizations and individuals use Web services to build complex business functionalities. Web services must operate properly in the unreliable Internet infrastructure even in the presence of failures. To increase system dependability, organizations, including service providers, adapt their systems to the autonomic computing paradigm. Strategies can vary from having one to all (S-CHOP, self-configuration, self-healing, self-optimization and self-protection) features. Regarding self-healing, an almost identical tool is communication-induced checkpointing (CiC), a checkpoint contains the state (heap, registers, stack, kernel state) for each process in the system. CiC is based on quasi-synchronous checkpointing where processes take checkpoints relying of control information piggybacked inside application messages; however, avoiding dangerous patterns such as Z-paths and Z-cycles; in such a regard the system takes forced checkpoints and avoids inconsistent states. CiC, unlike other tools, does not incur system performance, our proposal does not incur high overhead (as results show), and it has the advantage of being scalable. As we have shown in a previous work, CiC can be used to address dependability problems when dealing with Web services, as CiC mechanism work in a distributed and efficient manner. Therefore, in this work we propose an adaptable and dynamic generation of checkpoints to support fault tolerance. We present an alternative considering Quality of Service (QoS) criteria, and the different impact applications have on it. We propose taking checkpoints dynamically in case of failure or QoS degradation. Experimental results show that our approach has significantly reduced the generation of checkpoints of various well-known tools in the literature.