The comparison between RDD-tolerant processing and traditional fault-tolerant processing-(video note)

1. HDFs can only be read, or created by other means
2, Transfrmation is lazy.
3, traditional fault-tolerant mode, data checkpoint or record data update
Fault tolerance is the most difficult part of distribution.
Data checkpoint: Replicate large datasets across the network of the data center, between the machines where they are connected. Consumes network and disk.
Record Data update: Many updates, the record cost is very high.
4. RDD Fault Tolerant Mode
All Rdd know where they are coming from and record how their data is generated, and the original lost data is generated by recalculation.

The comparison between RDD-tolerant processing and traditional fault-tolerant processing-(video note)