DevConf.CZ 2019

This talk explains the key challenges of data replication of cloud/on-prem based Big Data clusters. To show how the replication strategies of the current storage systems (eg. Hadoop HDFS, Ceph)could be improved.

I. DATA

* What is Apache Hadoop Ozone
* main strength and vision (TLDR; multiple interfaces, not just hadoop: +s3 +posix fs).
* Explain why data replication is a key element of a cloud native storage.

II. SCIENCE

* Commonly used strategies (eg. random replication)
* What are the problems? (multiple independent node failures guarantee data loss)
* How they could be improved (larger data loss can be accepted if it has lower possibility).

This part is based on the copyset/hyperset papers but visualized and presented in a simplified way to easy to understand