Evaluating the impact of a coordinated checkpointing in distributed data streams processing systems using discrete event simulation

Coordinated Checkpointing is a fault-tolerance strategy proposed for Data Streams Processing systems, which handles a continuous, potentially unbounded flow of data under Quality of Service requirements. Although traditional in large-scale distributed systems, there is a lack of study on how a Coordinated Checkpointing may impact the stream processing in both failure-free and failure-prone environments, especially considering the inherent requirement of analyzing and processing data in real-time. This paper presents a study that used a discrete simulation model to investigate the impacts of the Coordinated Checkpoint fault tolerance strategy on a Data Stream Processing System. The results show Coordinated Checkpointing should be avoided since it critically impacts the stream processing and the real-time analyzes of data, increasing latency up to 120%, and discarding up 95% of the processing window during a global checkpoint when a rollback-recovery is required.