Cloud native data replication strategies
summary
Explains concrete replication choices for Apache Hadoop Ozone and compares random replication versus copyset/hyperset strategies using small math examples (6 datanodes: choose 3 gives 20 combos) and a 9-node grid where a tuned layout reduces multi-node-loss probability from guaranteed in the random case to about 7% while also increasing parallel recovery sources; shows the tradeoff that two-group designs lower loss probability but throttle rebuild I/O, and reports that the copyset approach is implementable and used in practice.
AI-written summary · about
chapters (5)
- 0:00Start
- 2:03What Is Hadoop
- 2:59Options To Use the Storage
- 7:25Independent Failures
- 8:02Correlated Failures
related talks
33mBalancing data in Apache Ozone
45mEmbracing Big Data Workload in Cloud-Native Environment with Data Locality
20mLightning Talk: Multi-Datacenter Strategies for Data Services
34mData replication patterns 101
47mAll the DataOps, all the paradigms
42mApache Ozone - State of the Union
24mOxia - A Horizontally Scalable Alternative to Apache Zookeeper
30mSecure Apache Ozone with High Availability