Overview
Explore a conference talk on UniStore, a fault-tolerant data store that combines causal and strong consistency for geographically distributed data centers. Delve into the challenges of maintaining liveness despite data center failures and learn how UniStore ensures durability for causal transactions only when necessary. Examine the evaluation results from Amazon EC2 using microbenchmarks and a RUBiS benchmark, demonstrating UniStore's effectiveness and scalability in combining consistency models. Gain insights into geo-replication, fundamental trade-offs, hybrid alternatives, and related work in the field of distributed data storage systems.
Syllabus
Intro
Geo-replication
Fundamental trade-offs
The hybrid alternative
A banking application
UniStore: causal baseline
UniStore: strong txs
US data UniStore: key challenge center crashes replicating to
Related work
UniStore: uniformity
UniStore: other features
Evaluation: results
Conclusion
Taught by
USENIX