seminars
Detail
Publication date: 1 de June, 2021Designing for Partition Tolerance with CRDTs
Planet scale systems, or systems with sufficiently high numbers of nodes, will eventually experience partitions. Rather than planning for the non-partitioned case, and betting nothing bad will happen, some developers are embracing the inevitability of partitions, high latency spikes and split-brains, and finding data management mechanisms that are still robust under those conditions.
Conflict-free replicated data types (CRDTs) provide a model that allows a principled approach to eventually consistent, always available, partition-tolerant systems. CRDTs also have a design spectrum with variable requirements with regard to the system model. In this talk we will first consider a system with a stable number of nodes and discuss operation-based CRDTs over causal broadcast. Then we focus on state-based CRDTs, over arbitrary numbers of nodes, and optimizations to state dissemination with propagation of deltas.
| Date | 03/12/2014 |
|---|---|
| State | Concluded |
| Host Bio | Carlos Baquero is an Assistant Professor and teaches Distributed Systems at Universidade do Minho, Portugal. He is a Senior Researcher at the High Assurance Laboratory within INESC-Tec. In the 90s, motivated by mobile computing and disconnected operation for file systems, he studied data types with merge (and fork) operations over semi-lattices, a precursor to state-based CRDTs. He is also interested in causality and in distributed aggregation algorithms. He specially likes all things distributed that eventually merge. |