Living on the Edge, Safely or: Life without Consensus

摘要

The centre-of-gravity of cloud is moving towards the edge. At edge scale, the opposition between the requirements of availability and ensuring correctness precludes any single simple answer. Choosing the right trade-off is a most vexing issue for application developers. To address this, we propose an application-driven approach, Just-Right Consistency (JRC). JRC derives a consistency model that is adapted to the specific application, being sufficient to maintain its invariants, otherwise remaining as available as possible. In order to maintain its invariants, even sequential code follows some standard patterns. We leverage mechanisms that uphold several of these patterns while maintaining availability: Conflict-Free Replicated Data Types (CRDTs) ensure that concurrent updates can be merged; Causal Consistency preserves relative ordering; Available Transactions preserve grouping. Together, these mechanisms form the TCC+ model. Furthermore, our CISE logic and analysis tools distinguish cases, in the remaining pattern, where the application's semantics requires synchronisation or not. This talk presents the challenges of edge-scale computing and the basics of the JRC approach by following the concrete example of a healthcare network. This research is supported in part by European projects SyncFree and Light-Kone, and by ANR project RainbowFS.