I have a setup where I do batch computation using spot instances in an auto-scaling group.
The batch computation is coordinated by a "coordinator node", which is just an EC2 instance. I use an elastic ip to ensure the spot instances can connect to the coordinator node.
This has downsides: AWS limits the number of elastic IPs an instance can have & it seems unnecessary since I don't want to expose the coordinator to the outside world.
Is there a better way to have the instances in the auto-scaling group connect to the coordinator node?