We have a Hadoop cluster with ~1.5 PB of data (i.e. ~1500 TB), running on bare metal with CDH 5.7 and without Cloudera Manager. We're planning to decommission the cluster and set up a new one from the scratch (bare metal as well, not cloud), problably switching to Hortonworks (HDP) this time. We're also moving the whole datacenter where it's currently located, so the new one will be on a different location. The idea is to keep all the data (all 1.5 PB of data is relevant, so unfortunately we can't get rid of anything). Just to clarify, we're talking about HDFS data as well as HBase databases/tables.
That being said, my question is:
Assuming we have our brand-new cluster set up and ready to ingest the data, what would be the best method to migrate all 1.5 PB of it to the new one? Needless to say we need to have the least possible downtime while doing all this.
Below is our current cluster's resources:
- 2 NameNodes in HA --> 2.80GHz 6-core / 24GB RAM
- 49 DataNodes:
- 5 of them --> 2.4GHz 6-cores / 72GB RAM
- 38 of them --> 2.3GHz 16-cores / 128GB RAM
- 6 of them --> 2.4GHz 32-cores / 128GB RAM
Thanks in advance!