Checkpointing and Causal Recovery in Intermittent Distributed Systems
intermittent computing, fault tolerance, checkpointing, recovery, rollback, causal consistency, energy harvesting.
The use of battery-free devices powered by ambient energy harvesting has grown significantly in recent years. These devices rely on intermittent computing, in which execution is frequently interrupted by unpredictable power failures. In such environments, ensuring reliable distributed operation is a major challenge.
Existing checkpointing techniques for intermittent systems typically preserve only the local state of individual devices, ignoring dependencies among them and potentially leading to inconsistent behavior after recovery. Conversely, traditional distributed checkpointing solutions provide strong consistency guarantees but assume stable execution conditions that are incompatible with intermittent operation.
This thesis proposes two complementary solutions for consistent recovery in distributed systems composed of intermittent devices. The first, called IRON (Intermittent Resilient Orchestrated Network), adopts a centralized architecture to preserve causal dependencies and coordinate recovery. The second, called REKINDLE (Recovery and Event Keeping in INtermittent Distributed Low-energy Environments), extends these principles to fully distributed environments by combining lightweight local checkpoints with persistent logging of causal events.
Both approaches ensure the consistent processing of messages, distributed events, and irreversible actions that may be performed in the physical world. Experimental evaluations demonstrate that the proposed solutions preserve causal consistency, prevent incorrect actuator activations, incur low storage and energy overheads, and maintain scalable recovery in highly intermittent environments, thereby advancing the state of the art in fault tolerance for intermittent distributed systems.