etcd was constantly restarting with too many open files until I gave it more room on Ubuntu 14.04
https://gist.github.com/darron/2aadb8f30f3dd6f580bf
This is a more sensible default - but it may not be enough depending on how many minion nodes there are.
This will allow more successful kube-up.sh executions. Since kube-apiserver doesn't start on the first try after etcd first starts up possibly due to the lack of resources on my server.