Jeremy Zawodny posted yet another great article over at Linux Magazine on some improved defaults that can save you a lot of grief when your network fails intermittently. In summary:

Faster replication heartbeat (old default is 3600 seconds = 1 hour):

slave_net_timeout 10

Disable DNS hostname lookups:

skip-name-resolve

Sane connection timeout (may need to be raised if your network is flaky):

connect_timeout 5

Disable host blacklisting after x number of failed connections:

max_connect_errors 1844674407370954751