[I answer my own question: a tip of the day]
When Whirr 0.6.0 spawns a Hadoop cluster (CDH or Apache) for EC2,
fs.default.name is set up with the public DNS of the master node. When
EC2 nodes talk to each other via private ip addresses, there is no
charge for data transfer, but there is a charge for transfer over public
ip addresses.
Is this limitation?
No, because DNS resolution inside the cluster returns a private ip
address for a public DNS name like ec2*1.compute.amazonaws.com. The
*.compute.internal DNS name resolution returns the same private ip address.
Paul
